The video sequence of the monitoring system is often affected by environmental noise, numerous moving targets and target occlusion, aiming at the problem that traditional video monitoring cannot effectively detect, track and count the workers, an intelligent multi-target tracking and monitoring system based on ARM was designed. The hardware design scheme and software environment construction were described from the perspective of system integrity. In terms of algorithm implementation, the target detection and tracking were realized based on the improved adaptive Gaussian mixture model and Kalman filter, the Hungarian algorithm was introduced to solve the task assignment problem of multi-target tracking by data association, and the occlusion problem was solved by using the Euclidean distance between detection target and prediction target and kalman filter. Experimental results show that the system can effectively track 6 moving targets under the limited field and camera perspective, and its average processing capacity remains at 18 frames/s.