Research On Image Segmentation And Matching Algorithm For Vision System Of Cotton-Picking Robot

With the development of science and technology, robots used in various industrialsectors increasingly and simulated the human eye functions of the computer visionsystem, is one of the important research in robots, primarily through the acquisition oftwo dimensional projected image to get some objective information inthree-dimensional space.This study on the image segmentation and matching technology is the keytechnologies of computer vision system, can be used to get the target depth,three-dimensional dynamic prediction of spatial information, and so on.This articlefirst describes a computer vision system of research and development at home andabroad, presented the theory of binocular stereo vision system with deep informationon how to restore and describes the using simulated hardware that includes cameras,lenses, video capture card and software environments, such as Matlab, driveninterface functions like capture card and Visual C++libraries.And then details the main research contents in this article--image segmentationand match.This article proposes cotton image segmentation algorithm that has basedon YCbCr space and GA neural network, introduced RGB and YCbCr color space andBP neural network, and GA genetic algorithm, descripted process of GA optimizatingBP, used8domain small area to remove the error segmentation regional after imagesegmentation, and through the experiments to simulate effect of image segmentation,and compared currently common of segmentation algorithms,including K-means andBP.The results indicate that the proposed algorithm has higher segmentation accuracy,is up to91.9%, and effectively avoiding the effects of light intensity.This article proposed cotton image matching algorithm which is based on FASTand SURF.Mainly used the FAST to obtain image corners, constructed point SURFdescriptor, quickly search for match points on images from the BBF algorithm andremove false match points using RANSAC and epipolar constraint, increase matching accuracy, prepared the ground for the restoration of three dimensional spatialinformation.By simulating experiment,the results show that this algorithm calculatesfaster and has higher accuracy and more number of match points than SIFT, SURFmatch algorithms.Finally presented for validation of image segmentation withmatching algorithm, through two step in the camera calibration with calibrationparameters inside and outside to recover cotton target depth of information, and thananalysing error sources in the field.