The entire video surveillance industry has rapidly entered the era of intelligent surveillance. However

the environment is complex and changeable in the video obtained under natural uncontrolled conditions

which poses many challenges for the detection and tracking of various types of pedestrian targets. For a variety of complex scenarios and different goals

how to design a target tracking and recognition technology with high efficiency

good robustness

and strong real-time performance is still a hot and difficult topic in the industry. Therefore

this paper focuses on non-rigid pedestrians in the monitoring field

and uses multi-features collaborative learning to analyze and research object tracking. It is intended to achieve Spatio-Temporal Context model for tracking

especially to improve the accuracy of the descendant tracking in a complex background. Experiments show that the overlapping rate (OR) and tracking center location error (CLE) for our proposed algorithm exceed the existing optimal algorithms

and its tracking speed is more than that of most algorithms.