Abstract: The existing algorithms of SBD show low robustness when there exist camera or object movements, light changes in the scene. An SBD algorithm based on ORB was proposed in order to effectively deal with these changes, in which shot boundary was located by computing the matching degree of ORB descriptor between adjacent images. In the detection process, the matching number and matching rate of the feature points were used to describe the characteristics of shot boundary, and the similarity degree curve was used to represent the relationship between image frames. Experiment results show that the proposed algorithm is effective to solve false and miss detection caused by the above problems, with a sharp rise in procession speed.