Aiming at the problems of low accuracy and long matching time of image matching methods for traditional robots,a new improvement algorithm based on SURF was proposed.The transformation from high dimension to binarization was realized by replacing the descriptor with the BEBLID descriptor based on SURF feature point detection.The method of adaptive setting threshold instead of artificial threshold setting was used to reduce the impact of matching.Combined with the elimination method of false matching point pairs by PROSAC optimization strategy,effective matching point pairs were obtained.The experiments show that compared with the expired improved algorithm,the matching correct rate and matching time of this algorithm are improved by 12.01% and 9.66% respectively.Therefore,in the extraction and matching of feature points,it has higher real-time and accuracy,and can meet the real-time requirements of product detection robot sorting.