An automatic target tracking system based on binocular vision was designed. After the image information was acquired by the binocular camera of the system， the Yolov5s real-time detection algorithm was used to quickly identify the target， the kernel correlation filtering method was used for target tracking， and finally the semi-global stereo matching algorithm combined with the weighted least squares method was used to measure the target depth to obtain distance information and location. After the position of the target was obtained， the speed and direction of the system were controlled， so that the system could maintain the relative orientation and set distance from the target， and realize automatic tracking. From the results， robots could replace reporters to perform real-time tracking reports of dangerous tasks， solving the safety problems of reporters.