A System Of Human Detection And Tracking Using Multi-feature Fusion Based On HOG

Aiming at the problems of slow detection speed and unstable tracking in traditional methods with the fixed camera scene,a multi feature fusion pedestrian detection and tracking system based on HOG is proposed.This paper proposes HOG pedestrian detection algorithm based on inter frame constraints,to improve the pedestrian detection speed of the traditional HOG method;a pedestrian tracking algorithm with particle filter based on multi feature fusion is proposed to improve the pedestrian tracking accuracy of the traditional particle filter method;an system control method based on multi information filter is proposed to improve the robustness of the whole pedestrian detection and tracking system.Aiming at the slow detection speed of the traditional HOG method in fixed camera scene,the three frame subtraction method's results of moving object detection provides temporal information constraints for HOG detection.The area of the relevant regional complex operations for HOG detection is reduced to the scene of the movement related area.In view of the above,the speed of HOG detection is improved without affecting the accuracy of pedestrian detection.Aiming at poor tracking accuracy of the traditional pedestrian tracking algorithm based on particle filter in the case of complex background,target occlusion,etc,a particle filter tracking algorithm integrating pedestrian color features and SUSAN features is proposed.When the color feature is stable,only the target color feature is calculated,to reduce the computational complexity of the algorithm.When pedestrians pass through the complex background or occlusion,the SUSAN features of the target are re calculated by back calculation to ensure the accuracy of pedestrian tracking under special conditions.An operation control method based on multi information filter is added to solve the large computational complexity and low tracking accuracy in simple pedestrian detection and tracking system.Using the spatial continuity of pedestrians in adjacent sequence frames,a comprehensive evaluation of size information and shape information is carried out to detection results and tracking results of the system.The operation of the system is adjusted according to the decision result,to ensure the accuracy and real-time operation of the system.Experimental results show that: compared with the three frame subtraction method and the traditional HOG detection algorithm,the pedestrian detection method proposed in this paper has faster detection speed under the premise of ensuring detection accuracy.Compared with the traditional particle filter tracking method,tracking method proposed in this paper under the premise of ensuring the speed of operation has better tracking accuracy.