Application Of Kalman Filter Algorithm Based On BP Optimization In Pedestrian Tracking

In the field of modern computer vision,pedestrian tracking has been a hot issue concerned by many scholars.With the breakthrough of various technical barriers in recent years,pedestrian tracking has been widely used in life.In recent years,pedestrian tracking method based on detector and tracker has become one of the hot research directions.Aiming at the problem of ID switch caused by algorithm’s erroneous judgement in multi-target tracking of pedestrian tracking,this paper proposes a pedestrian tracking algorithm based on the combination of yolov5 and Kalman filter.Above all,the Kalman filter algorithm is selected as the tracking module,because it is based on prediction to track and has fast operation which ensure the stability of tracking.In view of the large tracking error of Kalman filter algorithm in nonlinear system,this paper proposes a method of using BP neural network to optimize Kalman filter algorithm,and then through the simulation experiment comparison with other filtering algorithms,it is proved that this optimization method can effectively improve the performance of Kalman filter algorithm.Secondly,we choose YOLOv5 as the detection module for its extremely fast reasoning speed,by which can greatly reduce the time-consuming of pedestrian detection.In order to further improve the detection accuracy of YOLOv5,besides adjusting the training parameters reasonably,by adding additional negative samples with pedestrian occlusion and small target pedestrian on the basis of the original data set,we improve the training results of the YOLOv5 which can be obtained by comparing the inference results.Finally,build the overall algorithm framework.When tracking multi-target pedestrians,whether to give a new ID or inherit the previous ID is determined by judging whether the detection frame has a prediction frame matching with it.Finally,by comparing with other tracking algorithms on MOT challenge datasets,it can be verified that our pedestrian tracking algorithm has achieved the expected effect,and can effectively reduce the misjudgment of ID in multi-target pedestrian tracking.