This paper uses Shi-Tomas method for video search of sports competitions. Firstly, the ONVIF356-8950 4K HD network image data decoding processor is selected, and RS485 interface actuator is used as the main interface. Secondly, we build the running environment of sports game video, extract the similar playback screen of sports game video, and identify the boundary of the result. In the tracking link, this method adds a Hima point which has better ability to describe the object and is insensitive to light and rotation. During the operation of the tracking device, the corner points are tracked by optical flow algorithm, and the deviation test is carried out in two directions. Then the target window position is determined according to the remaining fixed corner points. Experiments show that this method can track the motion and deformation of moving objects effectively. The algorithm can effectively eliminate the motion error caused by smooth vertices and improve the stability of the system. The amount of computation improves the real-time performance by about 20% when encountering fast update scenes. This method can improve the search efficiency and shorten the query speed when the search keywords are exactly the same.