Research On Algorithms Of Vision System Of Cotton-harvesting Robot

In recent years, as China’s great cotton-producing province of xinjiang, theamount of cotton production has accounted for the proportion of about fifty percent inour country. It spend a lot of labor in picking up cotton due to the traditional way, andthe effect is not ideal when we use cotton-picking machines.These factors restrict thedevelopment of cotton industry in xinjiang seriously.As a result, the cotton-pickingrobot’s researching and developing is a good way to solve the current pickingsituation. Cotton-picking robot use binocular stereo vision system for cottonidentification, and transfer the location information to the manipulator. This paperresearch the binocular stereo vision system of cotton-picking robot, including imagesegmentation, maturity discriminant and image matching.The first part is about segmentation of the cotton image.This paper put forwardmethod of using PSO and K-means mixed clustering segmentation for cotton in theYCbCr color space.With this method,we can overcome the effects of cotton underillumination and shadow effectively,and balancing of the global-searching ability andconvergence speed. I’t can segment cotton image with all kinds of complicatedenvironment effectly.And more stability ande better precision than traditional PSOclustering segmentation method.The second part is about discriminating of mature cotton.this paper put forwardthe method using PSO to optimize the SVM based on shape feature of cottonmaturity discriminant algorithm. The method of cotton after binary imagesegmentation draw the minimum circumscribed rectangle, to extract the characteristicparameters of the shape of the target image, finally the SVM parameters wereoptimized by PSO algorithm can distinguish cotton maturity of the SVM model isestablished. The experimental results show that the method has high accuracy, shorttime, can meet the requirements of the cotton-picking robot in real-time. The third part is about image matching, this paper proposes a fast imagematching algorithm based on the characteristics of the cotton. The method to detectthe use of FAST and SURF to description of feature points, through the description offeature points using the algorithm of advanced KD-Tree in another image to find thecorresponding matching points, and uses the limit constraints and RANSAC for errormatching removal. The experiments shows that this method solves manyproblems,such as too little matching point, matching algorithm is slow and weak,edge features less faults.In addition to this,it can still matching successful to therotating images.