Research And Implementation Of Tree Fruit Localization Algorithm Based On Binocular Vision

Automatic tree fruit picking technology is a research hotspot in the field of intelligent agriculture,and the accurate location information of tree fruit is the premise for the successful acquisition of tree fruit by picking robot.In this paper,the tree fruit positioning algorithm based on binocular vision is studied.At present,the algorithms used for tree fruit location mainly include the algorithm based on feature point matching and the algorithm based on region matching.The location algorithm based on feature point matching has the advantage of good real-time performance,but due to light changes and background occlusion,the depth value of fruit center point will be missing,so that the location can not be achieved.The localization algorithm based on region matching performs full pixel matching for fruit region,with good localization accuracy but poor real-time performance.In addition,due to the influence of light changes and background occlusion,parallax values in some regions are missing,resulting in low localization accuracy in this region.To solve the above problems,this paper proposes a tree fruit localization algorithm with high positioning accuracy and good real-time performance.The algorithm combines AKAZE feature point matching,which can effectively match edge and detail features,with Patch Match Stereo region matching,which supports matching of tilted region and spherical region.The algorithm firstly performs channel conversion on binocular images of tree fruits.Then,the fruit regions in the image are segmented and extracted according to the color features.AKAZE feature matching is performed for all fruit regions in the left and right views,and the correct feature point pairs are obtained to calculate the depth details of the fruit region,and the number of fruits in the region is estimated to achieve the matching and positioning of feature points in the fruit region.Furthermore,the feature point positioning information was used in the Patch Match Stereo region matching and positioning of tree fruit images to improve the accuracy and real-time performance of the region matching and positioning algorithm,and obtain the dense disparity map of fruit region.Finally,the neighborhood four-direction quadratic matching method was used in the disparity filling step of the Patch Match Stereo algorithm.In order to improve the quality of the disparity map obtained by the algorithm,the correct disparity value of the disparity missing point was found from the disparity values of the four effective points in the neighborhood of the disparity missing point in some regions.The algorithm in this paper ten of the tree fruit pictures area to locate,the results show that the algorithm is the tree fruit location in the X direction of the maximum relative error is 1.20%,in the Y direction of the maximum relative error is 1.58%,in the Z direction of the maximum relative error is 0.18%,compared with the depth of the feature point matching algorithm is value and maximum error reduced by 1.06%;The maximum relative error of the fruit depth value was0.46%.In addition,the algorithm has high operation efficiency,and the time consumption is reduced by 92.6% compared with the original Patch Match Stereo regional localization algorithm,which can meet the practical application requirements of tree and fruit localization.