With the rapid development of computer artificial intelligence, the number of cameras is increasing, and the amount of video data is also increasing rapidly. The security monitoring and tracking of humanoid trajectory in video is an important research direction of large-scale intelligent monitoring system. Considering that the difference of illumination and darkness of different cameras in different security control scenarios and the human angle and size of each frame will affect the accuracy of human tracking, Correct Single Shot multibox Detector (CSSD) network with advantage of fastness and associated analysis are proposed for human tracking. Based on the pedestrian multi-object tracking technology, this study proposes a CSSD network for model detection, and uses ordinary Kalman filter to track and predict the position of the target, predicts the position of the detection box, and uses IOU method and Hungarian algorithm to solve the problem of video frame target matching before and after. It has been proved that this method can effectively improve the accuracy of humanoid targets, alleviate the large changes caused by epigenetic mutation or partial occlusion, and adapt to the size, distance, and angle changes of targets to the greatest extent.