关键词: 机器视觉, 立体匹配, 图像识别, 荔枝采摘机器人

Abstract: To resolve the problem on fast recognition and accurate location faced by the robot of picking litchi, a scheme on location of picking point based on parallel binocular stereo vision model is proposed, which is supported by the computation and matching of picking point obtained from combined algorithm of Harris and SIFT. Firstly, litchi cluster, litchi fruit and its main fruit bearing branch are sortedly recognized, by segmenting Cr gray picture of YCbCr with twin-threshold method. Secondly, region feature information on centroid and Minimum Bounding Rectangle（MBR） of litchi is extracted, along with corners from the recognized main fruit bearing branch detected by Harris algorithm, and two-dimensional coordinate pixel on the picking point is achieved. After stereo matching of the picking point based on SIFT vector search is accomplished, the experiment on picking location and its error analysis is finally carried out. The experimental data shows that depth error on location of the picking point commenced by interpolation method is less than 10 mm, under shooting distance of 354~590 mm, which is better satisfaction for litchi-picking robot.

Key words: machine vision, stereo matching, image recognition, litchi-picking robot