Castro-Garcia, Sergio :

Rosa, Uriel A

Smith, Dave :

Diaz, Ricardo :

Gliever, Chris

Burns, Jackie :

Glozer Dr, Kitren

Krueger, William H

Ferguson, Louise

Damage Assessment and Mitigation of Mechanically Harvested California Table Olives 2007 Annual Progress Report to COC 1 / 10 / 08 Uriel A . Rosa Sergio Castro - Garcia Dave Smith Ricardo Diaz Chris Gliever Jackie Burns Kitren Glozer William H . Krueger Louise Ferguson . A . Summary 1 ) Analysis : Extended 2006 / 2007 off - season studies were conducted on data recorded during 2006 season to identify the causes of injury to olive fruits . Two types of analysis were performed on : a ) sampled fruit collected before processing and b ) stereo high speed video images ( 42 olive fruits were analyzed frame by frame with existing software ) . Results indicated the improved DSE - Korvan canopy harvester needed to be further improved to reduce damage for green table olive production . However , processed fruit seemed to tolerate injury much better according to processed fruit results . Nevertheless , the injury caused to green olive can be immediately seen after harvest and samples were used in this analysis . From previous analysis , we had already learned that the main causes of fruit injury were originated from the actions of the spiked drums and the catching frame . Three main damage sources while the fruit was in the canopy were identified . The most important damage sources were detached olives dropping onto rods and rods hitting attached olives . Other less significant damage sources were lower impact magnitude hits among fruit during canopy shaking . 2 ) Machine Component Incorporation . Based on our analysis and collaboration with DSE the following improvements were suggested to the DSE - Korvan harvester : a ) increase padding on catching frame b ) improve rod padding on drum rods and increase rod tip resistance to wear and tear c ) incline all drums for better removal , less side throw of fruits away from catching frame and increased removal efficiency d ) include padded drum housing to act as velocity breakers to avoid detached fruits entering the lower drums and being re - hit by rods . Mitigating solutions 2a - 2d were incorporated into the DSE 007 harvester by DSE . Extensive tests were performed at UC Davis with material supplied by DSE to determine the proper hardness and durability of rod padding materials . 3 ) Machine Component Results The 2007 field tests at Rocky Hills Ranch have indicated the redesigned padded rods were durable and soft to fruits . However , an in depth quantitative analysis was not performed . In general , the 2a - 2d modifications seemed to improve the DSE 007 harvester . Visual inspections seem to indicate that fruit quality , i.e . , less injury , and Please note : text partially edited for grammar and flow . removal were substantially improved from previous season , however , final results will be obtained after the processors samples are released in 2008 . We collected some data on harvester performance ( yield monitoring system ) and some video images for the DSE 007 operating in Rocky Hills Ranch . Off - season analysis of these data needs to be performed for better evaluation of the previous machine component changes and to possibly propose further machine improvements for the next season . 4 ) Yield Monitoring System : An olive yield monitoring system was developed , incorporated and preliminarily tested during the 2007 season . However , limited data were collected during 2007 season . The system can be further improved and used in the 2008 season . 5 ) Further Machine Improvement Suggestions : a ) Extend catching frame on the opposite side of the harvester , b ) Increase assessment of canopy by studying and adjusting the drum linkage dimensions . c ) Perform a complete machine performance evaluation for 2008 with various drum speeds , amplitudes and ground speeds . 6 ) Section B shows the results of our 2007 studies with data collected with the DSE - 007 canopy harvester . Analysis of data collected in 2007 for further improvement of the harvester is in progress . Please note : text partially edited for grammar and flow . B . Analysis performed at the UCD Bio - Automation Lab in 2007 to improve the performance of the modified DSE - 007 olive canopy harvester with Stereo High Speed Imaging . Contents : 1 . Introduction . 2 . Preliminary Reports . 3 . Removal and damage percent estimation with frame analysis . 4 . Olive damaged evaluation : Fresh olives ( 2006 ) , and canned olives ( 2007 ) . 5 . Determination of olive positions with use of two high speed cameras . 6 . Accuracy analysis of the olive tracking process . 7 . Parameter extraction used for the tracking process . 8 . Results of the olive impact evaluations . 9 . Damage evaluation . 1 . Introduction This section presents in details the analysis developed during May to September of 2007 at Bio - Automation Lab ( BAL ) at UC Davis . The main objective of this investigation was to determine the source , magnitude and nature of the damage caused by the studied canopy shaker for table olive production with the goal to produce viable mitigating solutions for olive fruit injury . Data were previously collected during field tests performed on Oct 18 and 19 of 2006 at the UCD Experimental Station in Lindcove , CA on an orchard conventionally planted with a Manzanillo variety . Preliminary studies using these data identified and quantified olive damage ( see sub - section 2 , preliminary reports ) caused by isolated harvester components of the Korvan - DSE harvester - 2006 : tree canopy shaker head , catching frame , rear belt , rear bin drop and fruit dropped from different highs . After this analysis , we focused our attention in the interaction rod - canopy to understand olive damage and detachment process . Stereo videos 500 fps were recorded in 21 quarter trees ( 10 trees ) in a homogeneous orchard . 2 . Preliminary Reports These preliminary reports show a common introduction and justification about field tests . This work is focused on the image analysis used to evaluate olive damage caused by the canopy shaker . Please note : text partially edited for grammar and flow . â?? Design and Evaluation of Selected Machine Components to Reduce Injuries Caused to California Table Olives Mechanically Harvested . 2006 Project Report to COC January 26 , 2007 - Uriel Rosa , Chris Gliever , Louise Ferguson , Carlos Crisosto , Jackie Burns , Kitrin Glozer Dave Smith of DSE . â?? Mechanical harvester efficiency and damage evaluations . Louise Ferguson , Jackie Burns , Kitrin Glozer , Carlos Crisosto , Vanessa Bremer , William H . Krueger , and Richard Rosecrance . 3 . Removal and damage percent estimation with frame analysis First , we used the information obtained from the two high speed cameras to learn about the canopy shaking process . Each video ( right and left camera ) recorded 1024 frames , this represents 2.048 seconds into the whole shake sequence . In this time , the first and last frames were used to estimate the number of olives on each branch . Olive number was counted by visual identification in the common area displayed in both cameras ( Fig , 1 ) . Figure 1 . Visual olive identification at first frame ( left , 60 marked olives ) and last frame ( right , 33 marked olives ) Source : R3T3Qsw The removal was computed as the ratio of the olives counted in the initial video frame and the umber of olives counted on the last inspected video frame . This ratio represents the removal efficiency during the time interval recorded by the high speed video cameras . Then , by playing the sequence at low speed ( 15 fps ) the displayed removed olives were counted and isolated into two groups . We have the assumption : Vertical drop olive : Fruits that can be seen dropping in a downward direction , perpendicular to the ground . It is assumed those olives fall freely , without abrupt detachment . They can be representative of non - damaged olives . Please note : text partially edited for grammar and flow . Horizontal drop olive : Fruits with a different way to vertical drop . It is assumed that those olives had been hit by a rod , branch or they are the result of an abrupt detachment process . They are representative of potentially damaged olives . Twenty tests were analyzed but only fifteen tests had enough olives or branches displayed . Results obtained are displayed in Table 1 . Other Parameters description : Data , Row , Tree , Quarter and Vert : Reference dataset about the location and information of each olive tested . Drum speed : Theoretical rotation speed of the drum ( 220 / 180 rpm ) . AM / PM : Test time Left - Cam : Left camera results Right - Cam : Right camera results Initial Olives : Number of olives fruits found in the first frame of the video Ending Olive : Number of olives fruits found in the last frame of the video Sum olive removed : Total number of olive with vertical and horizontal drop Table 1 . Removal and damage percent estimation with frame analysis Left - Cam Right - Cam Row Tree Quarter Vert Drum speed AM / PM Initial OlivesEnding olive % RemovedVertical Droporizontal Dro Sum Initial OlivesEnding olive % RemovedVertical Droporizontal Dro Sum 3 3 NW BOTT 220 AM 71 59 16.9 17 1 18 94.44444 5.555556 100 152 122 19.7 6 0 6 3 3 SW MIDD 220 AM 60 33 45.0 40 7 47 85.10638 14.89362 100 60 46 23.3 28 6 34 3 4 NW BOTT 220 AM 33 10 69.7 54 7 61 88.52459 11.47541 100 35 32 8.6 51 4 55 3 4 SW TOP - MIDD 220 AM 183 132 27.9 33 8 41 80.4878 19.5122 100 134 110 17.9 42 14 56 3 13 SW MIDD - BOTT AM 41 26 36.6 34 17 51 66.66667 33.33333 100 87 27 69.0 52 19 71 3 13 NE BOTT PM 17 9 47.1 21 4 25 84 16 100 32 17 46.9 22 4 26 3 10 NE MIDD PM 147 109 25.9 39 8 47 82.97872 17.02128 100 140 90 35.7 29 8 37 3 3 SE MIDD PM 16 15 6.3 32 8 40 80 20 100 31 15 51.6 37 10 47 3 3 NE BOTT PM 52 16 69.2 72 10 82 87.80488 12.19512 100 60 43 28.3 125 17 142 4 2 SW BOTT 220 AM 57 30 47.4 23 6 29 79.31034 20.68966 100 86 29 66.3 41 8 49 4 7 SW TOP 180 AM 37 18 51.4 28 8 36 77.77778 22.22222 100 54 28 48.1 35 6 41 4 7 SE MIDD 220 PM 22 11 50.0 11 5 16 68.75 31.25 100 68 72 20 6 26 4 4 SW MIDD 220 AM 28 17 39.3 19 4 23 82.6087 17.3913 100 39 24 38.5 11 4 15 4 10 SW TOP 180 PM 78 65 16.7 41 7 48 85.41667 14.58333 100 78 70 10.3 25 16 41 4 4 NE MIDD 180 PM 46 10 78.3 4 1 5 80 20 100 61 27 55.7 8 3 11 Results displayed possible linear correlations between studied parameters . However , each camera showed different information about the same branch because they used a display area from different point of view . Images are high resolution and allowed identification and the ability to follow fruits during the shaking . In some videos , removed olive numbers were higher than the initial olive numbers . It was more important for bottom and middle drum position videos . This error was due because there were two drums working at different heights in contact with the canopy . In this way , the bottom drum also hit detached olives from top drum . But , video tests were recorded in short time mainly at the beginning of the tree or with well defined branches . At this point , error was small because only was in contact with the canopy the recorded drum . We found positive and significant correlations between the initial and final olive numbers in both cameras ( Table 2 ) . Although olive numbers between cameras were different , it is logical find a correlation between them . Table 2 . Correlations coefficients among counted olive in both cameras in initial and end frame Please note : text partially edited for grammar and flow . Initial Left End Left Initial Right End Right Initial Left 1 0.978 0.790 0.739 End Left 0.978 1 0.819 0.779 Initial Right 0.790 0.819 1 0.739 End Right 0.739 0.779 0.739 1 Results show a high liner correlation value between initial and final olive number in each camera ( 0.978 and 0.739 , respectively ) . We suspect that the removal ratio doesnâ??t depend of the number of initial olive in branch ( keep the consideration that test was for only two seconds ) but it is possible that it depends on the speed drum , orientation of the rods , time of the tests or number of rods . Beside , there are not correlation between initial olive number and removal ratio in both cameras ( R2 value of 0.247 and 0.338 ) . However , results of the removal process are different between cameras ( correlation = 0.024 ) ( Fig . 2 ) . 90.0 80.0 70.0 Removed y = 0.1231x + 36.67 60.0 2 = 0.0129 R 50.0 % 40.0 Left - Cam 30.0 20.0 10.0 0.0 0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 Right - Cam % Removed Figure 2 . Relation in removal ratio estimated between cameras It is difficult compare removal ratio between cameras . Cameras show similar removal ratios ( left = 38.41 ± 19.43 % ; right = 37.14 ± 19.84 % ) but with high variability data . It may be more appropriate to study each camera alone . We classified detached olives into two groups ( vertically and horizontally dropped olives ) . The correlation value between these groups could be discussed for each camera ( left = 0.445 and right = 0.603 ) . They have a low R2 value ( 0.32 and 0.41 , Fig . 3 ) and exhibit a positive correlation of the damaged olive ( horizontally drop olives ) with the number of the vertically dropped olives . The damaged olives are a positive function of the olives removed , and therefore the probability of damage increases with the number of olives removed but not necessarily with removal ratio . Please note : text partially edited for grammar and flow . Left - Cam Right - Cam 80 140 y = 2.5216x + 14.221 70 2 120 = 0.3201 R 60 Olive olive 100 y = 3.211x + 8.7081 50 2 = 0.408 R drop80 drop 40 Vertical Vertical 60 30 40 20 20 10 0 0 0 2 4 6 8 10 12 14 16 18 20 0 2 4 6 8 10 12 14 16 18 Horizontal drop olive Horizontal drop olive Figure 3 . Relationship between numbers of isolated removed olives in both cameras Results of the all removed olives in percent are present in Table 3 . There are no significant differences between ratio values in both cameras . Table 3 . Removal percent olives in both cameras ( mean and sd values ) Left Camera : Right Camera : All olive removed % 41.8 ± 20.8 37.1 ± 19.8 Vertical drop % 81.7 ± 7.3 80.4 ± 9.4 Horizontal drop % 18.4 ± 7.3 19.6 ± 9.4 Conclusions : - There is a relationship between the numbers of olives displayed in both cameras , and the olives before and after the shake process . However , removal ratios are different according with the camera selected . - The removal ratio is poorly correlated with the number of olives on the branch . Besides , this ratio presents values with a high variability . Despite tests being carried out under similar condition ( speed drum , orientation of the rods , time of the tests or number of rods and so on ) there is variability in the harvesting efficiency ( close to 40 ± 20 % ) - Visual damage evaluation shows hits by rods is the main source of olive damage . - The ratio of olive damage during shake process was close to 20 % of the all removed olives . 4 . Olive damaged evaluation : Fresh olive ( 2006 ) and canned olive ( 2007 ) This study compares the results obtained in 2006 during the green olive harvesting ( Report January 26 , 2007 ) and 2007 damage evaluation with canned olives ( Musco and Bell - Carter Companies ) . 4.1 . Fresh olive injury evaluation in 2006 Please note : text partially edited for grammar and flow . Tests carried on 2006 with fresh olives harvested in 2006 had the objective of isolating olive injures among different parts of the harvester . Results obtained in 2006 showed that most of the olives were injured and there was a great variability among tests , even with control tests . However , the most important samples - compared with the control values - were : 12 â?? drop olives , Bin drop olives and the samples from the spiked drum at 220 rpm ( Fig . 4 ) 100 90 Percent Olives Damaged for each test 80 70 60 50 40 30 20 10 0 1 Control Soft Tarp Bin drop test Rear Belt 4 â?? drop 12 â?? drop 180rpm 220rpm drum Figure 4 . Percent damaged in fresh olives from all harvester component isolation tests The Independent - Samples T Test procedure compares means for two groups of cases ( Table 4 ) . In each row , mean values followed by the same letter are not statistically different at the 5 % level . Table 4 . Percent damaged in fresh olives from all harvester component isolation tests Mean SD 1 2 3 4 5 6 7 8 1 Control 37.5 9.0 a 2 Soft Tarp 46.9 9.9 a a 3 Bin drop test 53.7 14.5 b a a 4 Rear Belt 30.9 12.7 a b b a 5 4 feet drop 33.9 7.8 a a c a a 6 12 feet drop 60.5 8.3 c c a b b a 7 180 rpm 79.2 14.0 d d d c c b a 8 220 rpm 64.3 17.9 e e a d d a a a Main significant differences can be highlighted : â?? Olives from Bin , 12 feet , 180 rpm and 220 rpm tests had more damage than the control olives . This result fits with the original report 2006 where the most important damage resulted from the interaction with the canopy and high drops . â?? Olives form 4 feet test presented less damage than olive form 12 feet test . â?? There was no difference in damage value between 180 rpm and 220 rpm test . â?? 180 rpm test had the highest damage value . â?? Control tests had high variability . Please note : text partially edited for grammar and flow . 4.2 . Canned black olive injury evaluation in 2007 In order to compare results , subsequent damage evaluations with canned black olives in 2007 were performed . Scores evaluations were transformed to percent of damaged olives ( from 40 score = for perfect olives , to 28 score = for most damaged olives ) ( 100 % = 28 and 0 % = 40 ) . Canned evaluations were made by two different companies : Musco ( whole olives ) and Bell Carter ( whole and pitted olives ) . Both evaluations were compared for same tests and scores were codification to remove the negative effect of olive stem because was considered like damage . Note that 0 value is showed with a very small column in plots . Figure 5 shows percent damage in whole canned olive for all tests . 100 Percent Whole olive damaged for each test 90 80 70 60 50 40 30 20 10 0 1 Control Soft Tarp Real Belt Bin drop 4 â?? drop 12 â?? drop 180rpm 220 rpm drum Figure 5 . Percent damaged in whole canned black olives from all harvester component isolation tests The Independent - Samples T Test procedure compares means for two groups of cases . In each row , mean values followed by the same letter are not statistically different at the 5 % level ( Table 5 ) Table 5 . Percent damaged in whole canned black olives from all harvester component isolation tests Mean SD 1 2 3 4 5 6 7 8 1 Control 16.7 13.9 a 2 Soft Tarp 23.5 13.2 a a 3 Bin drop test 14.5 12.6 a a a 4 Rear Belt 19.7 11.2 a a a a 5 4 feet drop 23.4 15.0 a a a a a 6 12 feet drop 16.7 16.9 a a a a a a 7 180 rpm 25.0 27.7 a a a a a a a 8 220 rpm 14.8 12.8 a a a a a a a a We found no significant difference among the mean values of the tests . Please note : text partially edited for grammar and flow . In a similar way , damage comparison were made with pitted canned olive ( Fig . 6 ) 100 Percent Whole olive damaged for each test 90 80 70 60 50 40 30 20 10 0 33 Control Soft Tarp Real Belt Bin drop 4 â?? drop 12 â?? drop 180rpm 220 rpm drum Figure 6 . Percent damaged in pitted canned black olives from all harvester component isolation tests The Independent - Samples T Test procedure compares means for two groups of cases . In each row , mean values followed by the same letter are not statistically different at the 5 % level ( Table 6 ) Table 6 . Percent damaged in pitted canned black olives from all harvester component isolation tests Mean SD 1 2 3 4 5 6 7 8 1 Control 27.7 6.5 a 2 Soft Tarp 13.8 15.4 b a 3 Bin drop test 16.6 15.5 a a a 4 Rear Belt 17.25 15.6 a a a a 5 4 feet drop 20.0 13.7 a a a a a 6 12 feet drop 23.5 13.2 a a a a a a 7 180 rpm 21.8 21.1 a a a a a a a 8 220 rpm 26.2 14.0 a a a a a a a a We found only one significant difference : - The olives from soft tarp tests present less damage than control olive . Maybe , this results is because the low data number . Conclusions : - Green olive harvesting showed the most important damage in the interaction of the canopy - drum and high drop tests . Previous reports showed that improvements in catch frame , using different soft padding materials reduced the olive impact magnitude . Therefore , we need to focus our work on understanding what happens in the canopy to know the source , magnitude and nature of the olive damage . Please note : text partially edited for grammar and flow . When processed into black olives , whole or pitted , damage produced by the mechanical harvester is apparently lessened and the olive grade well . . Therefore , evaluation as a green olive immediately after harvesting tests are required to analyze and evaluate the mechanical harvesting to obtain top quality harvested fruits . 5 . Olive Position determination with two high speed cameras The main objective is to knowing the source , magnitude and nature of the olive damage sources due to mechanical harvesting with canopy shakers . 3D high speed cameras were used to evaluate the fruit position , velocity , acceleration , impulse and energy in olive impacts , and also the rod effects in the canopy . Tests were carried out with a canopy shaker ( Korvan ) in Manzanillo olives . We have used the data collected in 2006 harvesting season ( Report January 26 , 2007 ) . 5.1 . Located the interested points in both cameras ( left and right ) Review each video and camera with a low number of frames per second ( 2 to 5 fps ) . Normally , one camera showed better illumination , position , focus and quality than other . When one of the cited events was located , the most difficult task was tried to find it in other camera . We used reference points easily placed in both cameras ( special small branches or leafs ) to locate the event . It was important know the frame number and the point coordinates ( it mean pixel number : x , y ) in the main frame where happen the event ( left and right ) to locate it during in subsequence analysis . Then , using play and reverse play we can select the video duration to analyze ( usually 40 frames before and after the event ) . We have used Photron Fastcam Viewer v . 2.4.3.5 . software to view , extract and save the frames sequence . 5.2 . Extracting the interested frames from . avi videos into . tif images Photron Fastcam software led us to save . avi videos into . tif imagen ( saved file name had form like : View file left or right_number_event.tif ) . It was important because then we can set the time between frames ( 500 fps ) . However , sofware cannot let doing this with . avi video ( normally configured with 30 fps ) . We saved several frames before and after each event with 80 frames per events at least ( 0.16 s ) . Short sequences only showed us the event selected ( however , during tracking process you can find more interesting event in this sequence that can be considered ) and it made the computer worked quickly . On the other hand , large sequences required a lot memory , it is slow process and is difficult to work with too frames number . 5.3 . Calibration frame process Tests were carried out with a calibrated target . Target was make up by two wood sticks ( Fig . 7 ) , but for future tests it could be better used a board with calibrated point targets . Problem is that you need all target points in the same plane ( z = 0 ) , so grid will be easier than you use two different planes ( one in each wood sticks ) . Please note : text partially edited for grammar and flow . Grid was generated later ( Adobe Photoshop 5.0 ) drawing lines in different layers based on target marks . To use this final image it is necessary save it as . tif image without any compression ( saved file name had form like:Calibration file left or right.tif ) Figure 7 . Target used to calibrate frames in tracking process in same frame in stereo cameras . Source T7R4se 5.4 . Upload.tif image sequence , calibrate and synchronize process Time and position in each event was determined using ImageExpress MotionPlus software . ( Note : software help is not really good but you can find a good tutorial and a manual about the program into the install file ) . Use WIZARD button menu to complete the test . The most important steps to complete the analysis were : WIZARD New Experiment . Name Calibration file left.tif . tif imagen without compression from Photoshop Apply Geometry Calibration Note that you have to start with the point Generate data from imagery in this file . 1 ( 0,0,0 ) , place this point in the left down corner to avoid negative coordinates . Please , use zoom function . User Defined used the maximum point displayed in both cameras camera Export data ( it is important , in this way you will save the input coordinates again in the other camera calibration image ) these data will lose if you close the program Left Add New File to Experiment View file left_1 . tif the software transforms into . tif internal sequence Apply Geometry Calibration Import Pre - Process data from another file Calibration file left.tif Specify 3D image sequence Please note : text partially edited for grammar and flow . First , Sync , Last frame Add new file to the experiment : Calibration file right.tif ( tif without compression ) Apply Geometry Calibration Generate data from imagery in this file User Defined ( First , Sync , Last frame ) camera First Import , then place the numbers ! ! Add New File to Experiment Right View file right_1 . tif the software transforms into tif sequence Apply Geometry Calibration Import Pre - Process data from another file Calibration file right.tif Specify 3D image sequence First , Sync , Last frame Close calibration images saving always all changes . Then â?? Sync Images â?쳌 and â?쳌 Lock Images â?쳌 at first frame . For each quarter tree ( experiment ) , you only need to make one calibration image per camera . Different events ( it means different tif sequences ) can be added to the same experiment but with different name , and then we can applied them the corresponding calibration file . For example , in a typical experiment with three events we can found the following files : Calibration file left.tif View file left_1 . tif View file left_2 . tif View file left_3 . tif Calibration file right.tif View file right_1 . tif View file right_2 . tif View file right_3 . tif 5.5 . Tracking the interesting points Usually , we used for tracking first option , exact cursor ( Fig . 8 ) because it led us select the interesting point in different frame . Sometimes , when point is clearly displayed or exists a great contrast with the background we can try to use White point , Black point or Feature point for automatic tracking ( better quality process ) . Please note : text partially edited for grammar and flow . Figure 8 . Tracking point toolbar There are some considerations that are not in the manual : â?? Program hasnâ??t undo button . It is important , be careful and save the data when you are sure that all is right . â?? Software doesnâ??t refresh data . If you make an important change ( like a new tracking , move , delete or change a point ) you need to save the file again to upload changes . Note when you want to track it an error window will appear . Then , save process , you need to â?? Sync Images â?쳌 at the beginning of the sequence . â?? Program doesnâ??t understand that you can track a fixed point ( fixed point in movement ? ) and it will show you an error window or only tracks the first and second frames and then you will see the plot with only two points ( however , it works only a few times and you can track a fixed point â?¦ ) One solution is : to track a point use the exact cursor option , find in each frame and in each camera the interesting point and completed the manual tracking ( it is not necessary that you start at frame 0 ) Then , go to the first frame in your manual tracking , and delete the first and second frame . Track the first frame with a Feature Match point with the same number point . So , program will ask you an advice window similar to : â?? you have inputted a Feature point for this number , and this number in other frames has different properties , do you want to change the properties of this number in the others frames ? â?쳌 You click yes . Now , we have an â?? automatic â?쳌 tracking which program can understand but we had done it manually . Note that second frame hasnâ??t got a number . This number will appear when we will track all sequence to obtain a result . Some common errors that the program doesnâ??t show you information about them : â?? Numbered points for tracking ( 1,2,3 â?¦ ) can have different properties between cameras but they have to have the same properties for the all frames of one camera . â?? Numbered points for tracking have to start and end in the same frames in both cameras . 5.6 . Getting position results Option menu let you set time between frames . It is important if you want to obtain the velocity and acceleration data from tracked point . Actually , I couldnâ??t be able to obtain a realistic measurement of velocity and acceleration with this software . For this reason , I only use the position and time for each point , saving each plot as a text file ( file.out ) . Then , velocity , acceleration , impulse and energy data can be calculated from this information . Please note : text partially edited for grammar and flow . Finally , effective number of frames tracked ( it means only events used later in the analysis ) without considering â?? Other olives in movements without hit â?쳌 were 5230 , it is 10.468 seconds . 6 . Accuracy analysis of olives tracking process Tracking process produces three dimensional position of point in time . This process , manual or automatic , generates an error in position data due to olive is a big and homogeneous target , leaves and shadows make lost data and low resolution zoom in ( Fig . 9 ) . In spite of this error is small in position it could be important with instantaneous velocity and acceleration computations because time frame is 0.002 s . In order to reduce random error to it is necessary to obtain the average value of a set of data able to give a representative value . Main objective is to determine the number of points necessary to obtain a mean value with a minimum error to compute velocity or acceleration values . Figure 9 . Tracking noise sources : olive fruit is a big and homogenous target ( left ) , leaves and shadows produce lost data ( center ) and zoom in with low resolution ( right ) Different events were considered in rod - olive interaction . Finally , we selected a well know event to evaluate and improve the data analysis methodology . Then , parabolic movement produced when freely dropping olive impacts a rod was selected to analyze ( Fig . 10 ) . After impact the olive movement can be reduced to two dimensions movement in a plane . Besides , analyzing a short period of time between impact and the higher vertical height ( hmax v = 0 ) the friction and turbulence air phenomena can be not y considered . In this situation , olive velocity can be expressed as : v ( t ) = v i + ( v â?? gt ) j ox oy Please note : text partially edited for grammar and flow . Where , horizontal initial velocity after hit , v , is constant and vertical initial velocity , ox v , decreases with the gravity acceleration and time . This velocity variation let us to oy evaluate the instant position measurements and to extract accurate mean data of velocity and acceleration from instant values . Figure 10 . 3D position data from a free fallen olive and hitting a rod In order to compare the necessary points to compute this measured gravity acceleration values different sets with 1 , 3 , 5 , 7 , 9 , 11 and 13 velocity data have been evaluated . Time considered in each velocity set is located in the central value set . In all , eight olives parabolic displacements were tested ( Table 7 ) . The acceleration was computed with this variable number of velocity points after hit and before higher vertical height point . Table 7 . Gravity acceleration estimation with different velocity data sets Points Test 1 Test 2 Test 3 Test 4 Test 5 Test 6 Test 7 Test 8 Mean SD 1 2.21 - 4.10 - 0.56 13.00 8.47 4.07 8.94 11.52 5.44 6.03 3 5.29 9.22 4.17 14.40 13.62 11.51 8.72 10.34 9.66 3.63 5 7.82 7.48 6.62 13.70 13.17 12.47 8.87 12.11 10.28 2.86 7 8.38 8.63 8.27 11.87 11.80 13.00 11.04 10.83 10.48 1.82 9 9.25 8.67 7.95 10.32 10.70 12.63 10.71 10.40 10.08 1.45 11 8.79 10.87 8.15 10.83 11.20 19.41 10.07 10.90 11.28 3.47 13 9.15 7.00 9.82 11.00 11.20 11.92 10.03 10.70 10.10 1.53 Figure 11 show the mean value from nine velocity values shows a closer value to gravity 2 acceleration and the minimum standard deviation ( 10.08 ± 1.45 m / s ) . In this set , data are in a normal distribution and with a T - mean sample there is not significant difference with 2 gravity acceleration value ( 9.81 m / s ) with a level confidence of 0.95 . Please note : text partially edited for grammar and flow . 20 15 ) 2 ( m / s 10 Acceleration Gravity Acceleration 5 2 g = 9.81 m / s 0 - 5 1 3 5 7 9 11 13 Number data Figure 11 . Gravity acceleration estimation with different velocity data sets Conclusions : â?? Tracking process reported excellent results , especially in olive position . â?? To obtain velocity and acceleration estimations from position values is recommendable use nine values to avoid noise effect in measurements and get accurate results . 7 . Parameter extraction from tracking process The following parameters are obtained from tracking 3D imagine analysis . Trajectory angle ( θ ) This angle is composed with lines that define each olive trajectory before and after a hit . Trajectory angle was obtained evaluating 19 frames ( 9 before and 9 after hit and the own hit frame ) . We reduced olive trajectory into two lineal displacements ( movement in a plane ) . To compute this angle we selected 3 well - know points ( x , y , z ) and it was the lowest angle formed in the three point plane ( Fig 12 ) . First point considered was located from 1 to 4 frames before hit , second point was the hit and third point was also located from 1 to 4 frames after hit . This sequence corresponded with an olive mean displacement value from 1.0 to 1.8 cm . These events had duration from 0.01 to 0.018 s . Please note : text partially edited for grammar and flow . Figure 12 . Trajectory , reduced trajectory and trajectory angle θ ( Source : T3R3sw OLIVE 2 POINT 1 ) Trajectory angle is a non - referenced value and it shows the hit trajectory modification caused olive impact with a rod or other olive . In this way , we take into consideration the force direction applied by the rod . A new parameter , Trajectory Change ( TC ) , is used to evaluate θ . This parameter defines small , medium and high trajectory changes established in two different thresholds , 130 ° and 90 ° . Horizontal angles ( α , β ) These angles show horizontal trajectory angle before ( α ) and after hit ( β ) . They are used to evaluate de increment in the linearity momentum ( impulse ) in the hit . Mean velocity and acceleration Three dimensional olive positions have been extracted from tracking process . Before and after the hit frame ( identified with imagine sequence ) the instant displacement between two frames was computed . Mean velocity was obtained following the last calibration process and using nine displacement values before and after the hit . In this way , with 500 fps , time considered in each velocity value was 0.018 s . Mean acceleration values before hit were computed using two set of instant velocity before hit . In similar way , mean acceleration values after hit were computed . Instant velocity and acceleration Instant values , velocity and acceleration before and after hit , were computed from instant displacement values . Due to data position variability and avoiding impact transitional effect ( 0.002 s before and after impact , it is because sometime a hit required more than 1 frame , something similar to a push ) , instant value was measured with an average value of Please note : text partially edited for grammar and flow . three frames . Studied points were placed at 0.004 s from hit frame ; this means closed to 1 cm before and after hit . Linear momentum Increment ( Impulse ) Linear momentum was determined from instant olive velocity before and after hit , reducing the olive movement before and after hit into 2 dimensions ( 3 points ) we can consider easily the increment of lineal momentum ( â?? p ) caused by impact in the olive in function of the olive mass ( m1 ) . We assumption that : - Rod velocity after and before hit are equal - It is a conservative impact ( Q = 0 ) 2 2 2 2 0.5 m1 u + 0.5 m v + Q = 0.5 m v + 0.5 m 2v 1 2 2 1 1 2 â?? p = m a = m â?? v / â?? t = â?? p / â?? t 1 Where : m1 is olive mass , m is rod mass , u is velocity value before hit and v is 2 velocity value after hit - Restitution coefficient is not considered , and we donâ??t know impact time ( â?? t ) - Olive mass ( m1 ) is similar in all tests In this way , â?? p / m1 is impact force estimation in the olive by a rod . In this parameter we consider the impact angle : â?? px = v cos α ± u cos β ; â?? p = v sen α ± u sen β ( where ± 1 1 y 1 1 depends about impact direction ) Lost impact energy per mass unit ( Q / m ) Lost impact energy and instant acceleration were the most useful parameter used to define an impact in other tests with fruits . The problem to determine force impact is to know de restitution coefficient and impact time . For this reason , mayor part of fruit damage tests were carried out in lab , under controlled condition . Field tests made instantaneous acceleration determination value was difficult to measured , because you need to know accuracy impact time and olive is a big fruit for manual tracking . In this way , if you want to use only one acceleration data you can get a large estimation error . However , lost impact energy is a solider value than instant acceleration . We can use velocity measurement form a range of data before and after hit to reduce errors . Besides , energy consideration with small displacement and time are more appropriate . But this parameter has three drawbacks : We assume impacts are not conservative . Therefore , lost energy is transferred in the impact to olive and damages it ( from kinetics energy to damage ) . We considered all rod padding , olives and hits like similar . For this reason , we not considered two events ( T7R4se Olive 1 Point 1 , T3R3nw Olive 2 Point 1 ) because they were a push ( with a long impact time ) and not an impact . They are clearly outliners with high value of impact energy . Please note : text partially edited for grammar and flow . We think rods hit olives during harvesting . It means that lost impact energy present positive value ( â?? Q = E.after - E.before ) , rods push the fruits . However , periodic rod movement causes that sometime olives hit a rod . In fact , we obtain that 15 / 26 impacts where olive placed in a branch hit a rod . Olive fruits losing energy and decreasing theirs velocity value . We donâ??t know each impacted olive mass . Therefore , lost impact energy is showed like lost energy per mass unit ( Q / m ) . Manzanillo tree tested had great fruits and we can guess fruit height between 5 - 6 grams . 8 . Impact evaluation results Sixteen olives trees were tested with different speed drum ( 180 and 220 cpm ) . In each test , a sequence of 1024 frames was recorded ( left and right camera ) at 500 frames per second . Finally , ten tree quarters ( six olive trees ) were selected for image analyzes based on number of branches and the fruit position in the branches . Each fruit or rod studied ( in all , 55 olive fruits and 9 rods ) required to know the location in each frame and its displacement before and after the main event . Different events considered and final data obtained are the following : 1 . Rod hitting olive ( 26 ) 1.1 . Red Rod ( with removal = 5 ; without removal = 13 ) 1.2 . Black Rod ( with removal = 5 ; without removal = 3 ) 2 . Olive hitting rod ( 9 ) 2.1 . Red Rod = 9 2.2 . Black Rod = 0 3 . Olive hitting olive without removal ( 7 ) 4 . Other olives in movements without hit ( 13 ) 4.1 . With removal = 8 4.2 . Without removal = 5 5 . Rod position ( 9 ) Damage sources found with high speed videos were : rod hit olives placed in branch , detached olive hit rod and olive hit other olive . Besides , other damage sources rarely times were displayed like a branch hitting olives . 8.1 . Rod hit Olive This damage source is about a rod hits olives attached a branch . After impact olive fruits can be removed or it can be kept in the branches . In both cases , olives are considered damaged . In order to know the damage magnitude though hit magnitude and the olive behavior after hit , we are going to use the data from the tracking olives . Please note : text partially edited for grammar and flow . Data from six Manzanillo olives trees were used . In all , 26 olive fruits were studied theirs rod - olive interaction . Impacts were made with two different kinds of padding rod : soft ( black ) and hard ( red ) rods . Removal percent when a rod hit an olive was 42.3 % . Olive removed by hit could represent near 4 % of fruits harvested ( 0.2 * 26 / 55 * 10 / 26 ) . These olives have an important impact and consequently a severe damage . Could it be better if this fruit keep in the canopy or is better removed it ? Canopy shaker is design to detach olive by vibration , without contact with fruits . Reducing fruit damage is one objective to improve the harvesting machinery . Parameters evaluated results are giving in the tables 8 to 10 . These tables show a high variability in the impact events . The main sources of variability should be : - Different conditions in each impact regarding rod type and olive location . Sometime is difficult to differentiate between impact and push between olive and rod . ( r ) - Rod velocity depends of the impact point location v = Ï? â?? , being lower close to drum and maximum in rod tip . - Detachment olive force show a high variability . Different olive fruits attached to the same tree can show variability up to 30 % and it conditions the removal process . Table 8 . Olive parameters before and after hit . Data are presented with mean value and standard deviation in ( ) Before Hit After hit Removed Olives Not removed Olives All = 26 All = 26 All = 11 Red = 5 Black = 6 All = 15 Red = 12 Black = 3 Trajectory 108.52 108.52 109.42 89.83 125.76 107.85 107.11 110.81 ( 43.80 ) ( 43.80 ) ( 44.67 ) ( 46.71 ) ( 39.25 ) ( 44.72 ) ( 46.20 ) ( 47.32 ) Angle θ 32.47 35.15 27.91 28.20 27.67 40.47 31.50 76.33 α or β Angle ( 22.86 ) ( 23.29 ) ( 18.84 ) ( 18.38 ) ( 20.97 ) ( 25.38 ) ( 19.21 ) ( 8.14 ) 2.54 2.73 2.80 2.67 2.40 2.38 2.47 Mean Velocity 2.05 ( 0.67 ) ( 0.58 ) ( 0.48 ) ( 0.70 ) ( 0.71 ) ( 0.63 ) ( 1.13 ) ( m / s ) ( 0.90 ) Ins . 761.76 776.17 911.43 663.45 751.20 830.71 433.15 662.98 Acceleration ( 342.41 ) ( 349.53 ) ( 311.84 ) ( 364.70 ) ( 349.03 ) ( 342.90 ) ( 127.97 ) ( 337.34 ) 2 ( m / s ) 1.27 2.35 2.62 2.13 1.83 1.96 1.31 Impulse na ( 0.74 ) ( 1.97 ) ( 1.33 ) ( 2.49 ) ( 1.11 ) ( 1.19 ) ( 0.56 ) 1.77 1.71 2.69 0.90 1.82 1.96 1.23 Lost energy na ( 1.26 ) ( 1.38 ) ( 1.00 ) ( 1.12 ) ( 1.22 ) ( 1.18 ) ( 1.41 ) ( Q / m ) Data variability and low olive number per group made difficult find significant differences . There are increments in velocity values before and after impact considering all olives , but we can find significant different . However , with Ins . Acceleration , Impulse and Lost energy , using Wilcoxon Test ( nonparametric tests ) we found : Please note : text partially edited for grammar and flow . - There is a significant different between olive acceleration produced by red and 2 black rods ( 854.46 vs 586.68 m / s ) ( Prob > | z | : 0.0236 ) ( Fig . 13 ) . Hard padding rods increment olive acceleration than soft padding . In this way , we can considerer red rod produces more damage per impact . - Besides , lost energy per mass unit with hard rod is bigger than soft rod ( 2.71 vs 1.01 J / kg ) ( Prob > | z | : 0.0132 ) ( Fig . 13 ) . Hard padding produce lower conservative impacts , we guess this energy is used to damage fruit . - However , we cannot find significant differences in Impulse values between padding materials ( 2.15 vs 1.86 ) ( Prob > | z | : 0.2253 ) 4 1600 3.5 1400 3 1200 acceleration energy 2.5 1000 2 lost 800 1.5 600 1 400 0.5 200 0 1 2 1 2 Padding Rod ( 1 = hard ; 2 = soft ) Column 1 Figure 13 . Instant acceleration and lost impact energy when rod hits an olive ( 1 = hard padding rod material ; 2 = soft padding rod ) For removal olives red rod made bigger trajectory changes than black rod . We can find significant differences between this values because we use a small sample size and variability ( 89.826 vs 125.76 ) ( Prob > | z | : 0.2353 ) . Therefore , data indicate a logical and expected result : softer padding is more adequate to reduce olive damage . We checked that soft padding made less olive acceleration but we can say that they have lees removal efficiency . Impact in soft padding made larger deformation in the material , producing lower changes in trajectories but increments time that fruit is in contact with the rod , generating with less acceleration ( or force ) similar impulse values than hard rods . 8.2 . Free drop olive hit a rod This damage source is regarding freely fallen olive hits a rod . This impact was analyzed in Section 6 . Olive fruits analyzed with this kind of impact were fallen in a vertical way . Therefore , we assumed these olives were removed without hits from rod ( Rod impacts generate a large displacement in removed olives that usually take them out of catch frame ) . However , we only collect data from olives hitting red rod , because harvester configuration placed black rods in drum top position . In all , 9 olives are tracked dropping and hitting rods . Most of them ( 7 / 9 ) showed negative values of lost impact energy , this is rod stopped olive fall reducing velocity . Main value parameters about these olives are showed in Table 9 . Table 9 . Free drop olive hit a rod . Olive parameters before and after hit . Data are presented with mean value and standard deviation in ( ) Please note : text partially edited for grammar and flow . Trajectory Mean Velocity Mean Velocity after Ins . Acceleration Lost energy Impulse 2 before hit ( m / s ) hit ( m / s ) ( m / s ) ( Q / m ) Angle θ 85.61 2.85 2.01 865.86 4.73 2.99 ( 42.65 ) ( 0.80 ) ( 0.93 ) ( 262.71 ) ( 1.13 ) ( 2.3 ) We can compare these results with previously obtained when a hard padding rod hit a olive . Results show high lost impact energy and impulse values . It indicates that this kind of impact produces deep olive damage . Avoiding this damage source is important because the harvester is hitting good quality olives that have being removed previously . Different possibilities to reduce this damage source are : - Select softer padding rod that absorbed more energy from impact - Reduce drum rod number 8.3 . Olives hit olives During canopy shaking process olives attached by stems hit other fruits , branches and leaves . Then , removed olives fell into the canopy hitting tree canopy , harvester and finally catch frame . However , we cannot evaluate olive hits thought the canopy . This damage sources can be researcher in future works . Canopy shaking movements is similar to rod frequency but with bigger displacements . Olive production is located in external , slim and flexible fruit - bearing branches that attach fruit with a high force . Therefore , olive impacts with branch are small and difficult to detect ( doesnâ??t make trajectory changes ) . Rod amplitude is small compared with branch length and fruit - bearing branches and it is not usual that branches roll up with rod . Damage source that appeared most important in tests was the impacts between fruits , in the same or different branch . Seven impacts have been found and tracked . Principal problems to measured impacts were placed impact frame and found the same olives . In fact , was difficult to isolate between impact and push . For this , data obtained are results of big impacts , featuring best samples to get olive damage ( Table 10 ) . Table 10 . Olive hit olive , impact parameters before and after hit . Data are presented with mean value and standard deviation in ( ) Mean Velocity Mean Velocity Ins . Acceleration Lost energy Impulse 2 before hit ( m / s ) after hit ( m / s ) ( m / s ) ( Q / m ) 1.78 ( 0.79 ) 1.81 ( 0.80 ) 441.12 ( 195.71 ) 3.08 ( 1.61 ) 0.98 ( 0.79 ) Please note : text partially edited for grammar and flow . 8.4 . Impact sources comparison It is assumed that olive damage is proportional to impact magnitude when olive is in the canopy , before will take into catch frame . Therefore , it is difficult to compare several impacts when initial conditions and even material are not the same . We consider in damage source evaluation the impacts : rod hit an olive ( 26 data ) , olive hit a rod ( 9 data ) and finally impact between olives ( 7 data ) . Comparisons for all pairs using Tukey - Kramer HSD with a level of 0.05 were used . Figure 14 . Impact sources comparison with instantaneous acceleration after impacts 1600 Level Mean 2 A 865.86367 1400 1 A B 718.84164 3 B 441.11614 1200 Levels not connected by same letter are AH 1000 significantly different . Iv Acc 800 1 = Rod hit olive 600 2 = Olive hit rod 3 = Olive hit olive 400 Pairs of Means are significantly different : 200 All Pairs 1 2 3 2 - 3 Tukey - Kramer Column 1 0.05 Figure 15 . Impact sources comparison with olive impulse Level Mean 7 2 A 4.7313180 6 3 B 3.0760330 1 C 1.2400639 5 Levels not connected by same letter are impulse 4 significantly different . 3 1 = Rod hit olive 2 = Olive hit rod 2 3 = Olive hit olive 1 Pairs of Means are significantly different : 0 2 - 3 ; 2 - 1 ; 2 - 1 All Pairs 1 2 3 Tukey - Kramer Column 1 0.05 Figure 16 . Impact sources comparison with lost impact energy Please note : text partially edited for grammar and flow . Level Mean 8 2 A 2.9921706 7 1 A B 1.7878502 6 3 B 0.8763169 energy Levels not connected by same letter are 5 significantly different . 4 Loss 1 = Rod hit olive 3 2 = Olive hit rod 2 3 = Olive hit olive 1 Pairs of Means are significantly different : 0 2 - 3 All Pairs 1 2 3 Tukey - Kramer Column 1 0.05 Fig . 14 gives instantaneous acceleration after hit for three canopy damage sources . Sources 2 ( olive hit rod ) and 3 ( olive hit olive ) are significantly different , while source 1 ( rod hit olive ) show intermediate value between 2 and 3 with the higher variability . Fig . 15 shows significant different among three damage sources . Higher impulse values are obtained when olive hit a rod followed by when olive hit other olive into a branch . For this parameter , source 1 shows lower impulse values and low variability data . This result can be explained because olives in source 2 fall with a high vertical velocity component and changed abruptly into horizontal components . However , olives in source 1 showed lower impulse value because the trajectory changes are less abruptly and olive movement in the canopy is mainly horizontal and after the hit olive keep a main horizontal velocity component . Fig . 16 plots lost impact energy among damage sources . This parameter has similar behavior that instantaneous acceleration . Both parameters seem to be a good damage indicator for olive damage evaluation in canopy shaker . Conclusions : Highest damage source is when olive freely dropping and hit a rod . To reduce this damage we can select softer padding material or reduce rod number . Impacts produced among olives placed in the canopy or with branches during shaking process are the lowest important . Highest impacts reported between olives have a damage importance 30 - 50 % ( in terms of acceleration - lost impact energy ) of olive freely dropping and hit a rod . However , it is possible that this damage threshold will be assumed by olive impact resistance . More studies are necessary to establish the olive acceleration or lost energy thresholds to shake the canopy with the minimum damage produced . Impacts made by rod to olives attached to the branch have an importance 60 - 80 % ( in terms of acceleration - lost impact energy ) of olive freely dropping and hit a rod . However , these impacts are more frequently and with higher variability . To reduce theirs importance the previously consideration can be carried out Instantaneous acceleration after impact and lost impact energy have similar behavior to predict olive impact importance . These indicators are solid and coherent and showed a good response for field tests . 8.5 . Rod trajectories and canopy vibration Please note : text partially edited for grammar and flow . Harvesting machine has tree drums located in different highs . Bottom and middle drums have theirs main axis in a vertical position while top drum is inclined . Each drum has 8 layers and 20 rod per layer . In all , 160 rods per drum and 480 rods in harvester machine . A hydraulic engine power each drum thought an inertial wheel . Inertial wheel connected with the drum lets free drum rotation but with a fixed eccentricity . Hydraulic engine revolution defines the drum frequency . Canopy shaker is designed to remove olive without rod - olive contact . In this way , rod transferred energy to canopy in a periodic movement . The aim of this section is know the rod movement and what is about canopy transmission . Harvester controls drum speed by an oil valve . However , rod movement is the results of canopy - rod interaction , harvester velocity and eccentric joint between engine and drum . Each drum revolution is translated as a rod period movement . Amplitude rod is controlled by eccentric joint ( fixed parameter : it express canopy rod penetration ) , canopy interaction ( according rod and canopy density ) and harvester ground speed ( variable parameter ) . Originally harvester ground speed was set at 0.35 m / s and it was kept constant during tests by worker appreciation . Besides , tests were carried out with two drum speed level ( 180 and 220 cpm ) . However , measurement rod frequencies donâ??t let us establish this difference because it was estimated according engine oil flow and data presented high variability between groups and also with low samples ( 180 cpm = 2 samples ; 220cpm = 7 samples ) A typical rod movement is plotted in Fig . 17 . Three amplitude values can be measured in rod movement : X amplitude : horizontal projection amplitude in the harvester direction movement . It depends mainly about harvester ground speed and rod - canopy interaction . Y amplitude : Theoretically it must be null . It means vertical projection amplitude due to target , camera or drum inclination Z amplitude : deep movement the rod into the canopy . Theoretically , this value has to be constant because it depends on eccentric join engine - drum . Variability data can be obtained because drum can go closer or farer to canopy by a jointed arm . For example , it happens at the beginning and end of the shaking process . Please note : text partially edited for grammar and flow . Figure 17 . Tracked tip rod position into canopy during shaking process Table 11 presents results of 9 tracking rod positions in 7 different tests . Velocity and acceleration values were average value of 4 point considered in each rod cycle . These points were placed at distances of Ï? / 2 rad . Each value was the average of 9 data ( four before and four after the point ) as was setting in section 6 . Measurements were done in the tip of the rod . Table 11 . Rod movement properties into canopy during shaking process Amplitude ( mm ) Frequency Velocity Acceleration 2 ( Hz ) ( m / s ) ( m / s ) X Y Z 148.64 31.55 138.59 5.26 2.37 562.97 Rod ( 11.97 ) ( 12.26 ) ( 20.89 ) ( 0.57 ) ( 0.37 ) ( 278.98 ) Rod movements in horizontal plane were similar during tests . There are not significantly different between X and Y amplitudes . Mean rod horizontal amplitude into canopy was 143.61 mm ( 143.61 ± 17.30 mm ) with a frequency of 5.26 Hz ( 316 cpm ) . Velocity measurement is a realistic value . In this way , considering rod has a circular movement where : r = 71.805 mm and angular velocity Ï? = 2Ï? / 0.19 = 33.069 rad / s , we obtain a velocity value = 2.37 m / s , equal to average velocity measured in tip of the rod . Instant acceleration in tip rod showed a typical variability data because was average from 4 points ( 2 maximum and 2 minimum ) . By other hand , Y amplitude was low values ( 31.55 mm ) and we can consider rod movement into a horizontal plane . Rod movement is the vibration source of canopy . Rods shake canopy and the vibration is transmitted by fruit - bearing branches to remove the fruits . Therefore , canopy vibration has been characterized with instantaneous velocity and acceleration values ( 33 data ) Please note : text partially edited for grammar and flow . before impacts from olive data of sources 1 and 3 ( olives attached to branch ) . Then , transference ratio was obtained to characterize the process . Table 12 shows data used in transmission process between the machine ( measured in tip of rod ) and canopy ( measured in olive before impact ) . Transfer ratio up to 100 % indicates an amplification in the value considered between machine and canopy and value down 100 % indicates transmission reductions . Table 12 . Vibration transference between machine and canopy Rod Olives Transfer Ratio Velocity ( m / s ) 2.37 ( 0.37 ) 2.00 ( 0.87 ) 84.44 % 2 Acceleration ( m / s ) 562.97 ( 278.98 ) 646.80 ( 323.64 ) 114.91 % There are not significantly different at level 0.95 for mean values in rod and olives , this is due to data variability . However , mean value comparisons with Transfer Ratio show a small reduction in velocity values transferred to the canopy ( reduction of 15 % ) and an increment of 15 % in accelerations transference . Checking these results with record videos , we appreciate that olives in the branches follow the rod alternative movement with longer amplitudes and periods ( smaller frequency ) . Olive displacements have longer amplitude because they are attached to fruit - bearing branches and , therefore , need more reaction time to change theirs trajectory . Conclusions : Rod displacements can be considered in a horizontal plane with a 150 mm of amplitude and frequency between 4 and 6 Hz . Canopy is shaking with similar vibration level that rod movement : velocity values close to 2 m / s and instantaneous acceleration of 60 g ( g = gravity acceleration ) . These values presented a mean value modification close to 15 % in the transmission canopy - olive . These excellent results in transfer vibration ratio need to be researched in order to optimize the rod number in the machine ( rod number / canopy volume ) in order to reduce olive damage by rod impacts ( damage sources 1 and 2 ) 9 . Damage evaluation This work has the final objective to understand the interaction between canopy and harvester to avoid or reduce olive damages during shaking . Actually , olive damage sources in the canopy had been isolate and compared . In these field tests , initial harvester setup wasnâ??t constant and it worked in a range of drum speed . For this , we characterize olive movement in the canopy using instantaneous acceleration values before impacts ( 33 data ) . Acceleration values were averaged for trees with more than one olive Please note : text partially edited for grammar and flow . measurement . This parameter expresses the harvesting effect to remove the fruits and it is a good indicator about olive damage severity . We assumed that olive trees are homogeneous into orchard , and then olive impact probability ( with rods , branches or other olives ) only depends on machine effect into canopy . Green olive damage was evaluated for these trees after harvesting in lab ( section 4 ) . This evaluation was adapted to consider only the mechanical fruit injures . Finally , we know acceleration data from 10 quarter trees ( 6 different trees ) and theirs green damage evaluation . Fig . 18 shows relation between olive acceleration and damage evaluated . There is a significant positive linear correlation ( 0.6743 ) and r2 value 0.454697 . 100 90 Damage 80 Green 70 60 50 0 200 400 600 800 1000 1200 1400 1 + 3 Acc IV BH 2 Figure 18 . Green damage evaluation and olive acceleration ( m / s ) in the canopy The significant positive linear correlation indicates higher damage with canopy acceleration . We use a linear correlation to understand the relation between field data with damage estimation . Actually , r2 value is low but it is necessary consider data origin . According with results , a damage - acceleration threshold can be established to isolate and 2 prevent high olive damage . Instantaneous olive acceleration value up to 800 m / s caused higher olive damage ( up to 80 % ) . For this reason , it would be disable reduce the canopy acceleration to decrease olive damage . By other hand , it seems opposite to obtain higher removal percents . More studies are necessary to obtain relation the drum frequency and rod density with olive acceleration . Also , it will be interesting to evaluate different harvester ground speeds . Besides , setting the damage acceptable threshold for green olives is necessary to improve the harvester , optimizing the relation removal efficiency and olive damage . Green damage evaluations after harvesting had been very important . Conclusions : Olive instant acceleration in the canopy is a good indicator of damage Please note : text partially edited for grammar and flow . Olive green damage was linearly correlated with olive acceleration measured with 2 the imaging system . Acceleration values up to 800 m / s produced up to 80 % of olive damage ( fig . 18 ) . More field tests are necessary to improve harvester and green olive damage evaluation . Please note : text partially edited for grammar and flow .