A vision-based fruit-vegetable picking robot helps to improve picking efficiency by making full use of the information by which the target of harvest can be recognized and located. For harvesting robots, it is important and difficult to calculate and locate the picking point from the recognized main fruit bearing branch of litchi. Hence, calculation of picking point and its stereo match become the research focuses. To meet the needs of picking the whole litchi cluster for litchi picking robot, a scheme of combined algorithms of Harris and improved SIFT to compute picking point of litchi and achieve its stereo matching was proposed. Firstly, corner extraction from the main fruit bearing branch of litchi was carried out by Harris method, and the whole identified area of litchi fruits was taken as a big fruit, the feature information on “centroid” and the maximal vertical coordinate vertex of the MBR (Minimum bounding rectangle) of the big fruit (denoted by Y ) were then attained. Then, taking each Harris corner whose vertical coordinates were bigger than Y as the center of circle, all possible circles were computed and the center of circle whose circle area was the maximum was chosen as the pixel coordinate of picking point in original collected litchi image. Furthermore, the computed picking point was described with a characteristic vector of SIFT with 128 dimensions, and its binocular stereo matching based on cosine distance similarity of SIFT was also proposed. Theoretical analysis and experimental results show that the proposed scheme can satisfy the need of vision of litchi picking robot with successful matching rate of 89.55%, which means that the scheme can improve the computation precision of picking point from main fruit-bearing branch with complex construction.