Key Technologies Of Vision System For Cucumber Harvesting Robot

Fruit and vegetable harvesting robot based on machine vision could identify fruits and vegetables to improve picking efficiency by using the ability of information apperceiving, but harvesting robot has not been widely used. One of the most important reasons is that the fruits recognition rate is not high enough. Thus there would be important practical significance for accelerating the practical process on agricultural harvesting robot and changing the current status of the main rely on manual harvesting by studying the target recognition of harvesting robot.In this thesis, cucumber in greenhouse was taken as the study object, in order to solve the problem on recognition between cucumber fruits and leaves or stem which have similar color, the visible light images of cucumber in natural environment was studied. The fruit could be identified correctly by using template matching which using shape features of cucumber, the recognition rate could achieve to 87%, the three-dimensional coordinate location of cucumber picking point could be calculated by using binocular stereo vision, the location error could achieved to 9.46mm. The main research contents and results in the thesis are as follows:1. The color features of cucumber image were studied. The R, G, B and H, S, I component of cucumber fruit and leaf was analyzed, although there is some differences between fruit and leaf on G component but this difference is not enough identify the fruit from the background which could achieve ideal cucumber segmentation. An identification method based on cucumber fruit shape features was proposed. The cucumber fruit border was obtained by using basic image processing on a simple background cucumber image, including preprocessing, morphological operations, region labeling and edge detection.2. Cucumber fruit shape representation and description methods were discussed. 200 pieces of different cucumbers were taken as the sample source, the cucumber images were taken and the fruit border was obtained by image processing. Then the boundary was described by using elliptic Fourier description and was normalized. The average elliptic Fourier descriptors waer obtained by averaging all the descriptors. After that the standard cucumber was obtained by inverse transform of boundary descriptors.3. Template deformation and template matching method was proposed. After 5 times of scale transformation and 13 times of angle transformation on the standard boundary, 65 pieces of cucumber shapes with different size and different angle was achieved. The gray template matching was done on 100 pieces of cucumber image with complex background by using template matching algorithm based on template database, and the correct recognition rate is 87%. The fruit with partial occlusion and similar color can still be recognized correctly. The matching template starting point was taken as the fruit top point.4. The principle, mathematical models and calibration methods of binocular stereo vision was discussed. A binocular vision system was established. The internal and external parameters of the binocular stereo vision system were obtained by camera calibration experiments. And the space coordinates of cucumber picking point was calculated using triangulation measurement method, and the actual position of picking point was measured by using FARO ARM, and the results were compared with the calculation results, the results showed that the max positioning error is 9.46mm relative to the object distance between 500mm and 750mm which could meet the requirement of the vision system..5. The identification and positioning system for cucumber harvesting robot was established including the hardware and software systems which achieve the identification of cucumber fruits and space positioning.