Abstract: Aiming at the problem of high location precision and real-time in robot servo grasping, we proposed a fast algorithm for feature matching and target location. Firstly, Shi-Tomasi detection algorithm was adopted to extract the feature points. Secondly, a new definition method of feature descriptors was proposed, the sub-image was intercepted with the feature point as a center, and the feature direction was determined by the partial derivative of two-dimensional Gaussian function, the local image was rotated according to the feature direction, and the local binary pattern of the standard local image after rotation was extracted as the feature descriptor, which had good locality as well as invariance of translation and rotation. Finally, the feature matching was realized by calculating the Hamming distance between the feature descriptors, the homography matrix was estimated to locate the position and direction of the target in the scene. The experimental results show that the algorithm has the advantages of fast 
matching speed, high location precision and good stability, and can meet the requirements of location precision and real-time in robot servo grasping.

Key words: Shi-Tomasi detection algorithm, feature detection, feature matching, target location, local binary pattern (LBP)