Recognition And Location Of Tomato In Greenhouse Based On Binocular Stereo Vision

Robots based on machine vision have been widely applied in facility agriculture. Machine vision mounted on the robot, played an important role in obtaining environment information. This research studied the application of stereovision system to recognition and location of ripe tomato for tomato harvesting robot. The main research contents and results were showed as follows:1. The research achievements in the field of facility agriculture robot were reviewed, especially the application of machine vision in agriculture robot. The current developments of recognition and location of object by stereovision system were introduced, and the problems were put forward.2. A suitable stereovision system was set up for this research. This system is composed of two CCD cameras, two frame grabbers, and an industrial computer.3. Recognition of ripe tomato based on machine vision. Through the analyses of linear transformation of R, G and B in images shot by color CCD, the color image was transformed to grayscale by proper color feature, and then was segmented by threshold. The image processing technology was used to recognized ripe tomato from background. At last the centroids and minimum rectangle bounding the tomato were extracted.4. The algorithm based on feature-based stereovision methods were designed to detect the position of ripe tomato. The centroids and edge feature were extracted. The centroids of two images were directly utilized to matching for estimating the approximate position of gravity center. The edge was divided into many same long parts. These parts were expressed with chain code. The two directions matching method was applied to find corresponding points, and then the depth was computed. The different picking points were chosen in terms of respective end-effectors. The time was used in computation of this algorithm was less than 0.07s, and when distance is less than 550 mm, errors ranged within ±10 mm..5. The algorithm of detecting the position of ripe tomato based on area-based method stereovision was researched. The ROI (region of interest) was built according to the centroids and minimum rectangle bounding the tomato. The suitable color feature was used as gray value. Then the tomatoes in left and right image were divided into many matching windows. The candidate matching region was set up in right image in which the corresponding point was determined according to the similarity of window gray. The twice threshold was used to distinguish the results. The picking points were chosen in terms of respective end-effectors. The time was used in computation of this algorithm was less than 0.35s, though the time used by area-based method is longer than that of feature-based method, this algorithm can get more points for conveniently choosing picking points. When the distance was less than 350mm, the errors were near 0mm, and when the distance less than 450mm, the errors ranged within ±13 mm.