Research On Pedestrian Abnormal Behavior Recognition In Parking Lot Based On Multi-feature Fusion

As a place where people often go in and out of life,the management of parking lot has become a serious problem in the process of urbanization.There are also pedestrian abnormal behavior events in the parking lot.Therefore,how to accurately and quickly identify the abnormal behavior of pedestrians in the parking lot has become a difficult problem to be solved in the current parking lot management.The dissertation takes the behavior of pedestrians in the parking lot as the research object,and proposes an abnormal behavior recognition method based on multi-feature fusion.The main reaseraches are as follows:(1)The YOLOv5 pedestrian target detection algorithm fused with attention mechanism is proposed,which solves the problem of low accuracy and poor real-time performance of pedestrian target detection in parking lots.Aiming at the YOLOv5algorithm’s insufficient accuracy rate when quickly identifying targets,the dissertation improves the accuracy of the YOLOv5 algorithm in detecting pedestrian targets by adding hybrid attention.The result shows that the algorithm has higher recognition accuracy and good real-time performance.(2)An improved method of pedestrian tracking in parking lot based on Deep Sort is proposed.Aimed at the problem that pedestrians in parking lots are blocked by vehicles or other obstacles.Through the Deep Sort target tracking method,combined with the YOLOv5 target detection algorithm,the traditional target tracking algorithm has solved the defect of low tracking accuracy due to occlusion.Experiments show that the algorithm can effectively improve the accuracy of pedestrian target tracking and meet the accuracy requirements in practical applications.(3)An abnormal behavior recognition algorithm based on the fusion of trajectory features and skeleton features is proposed,which solves the problem of low recognition rate of abnormal behaviors in parking lots with single feature.The dissertation uses the Alpha Pose model to extract the key point information of the pedestrian’s posture,combines the trajectory information obtained in the target tracking process,uses the multi-feature fusion method to characterize the pedestrian behavior in the parking lot,and uses the multi-layer perceptron algorithm to classify.Normal behaviors include: walking,running,and meeting;Abnormal behaviors include: falling,fighting,and chasing.The experimental results prove that the recognition rate of abnormal behavior is 91.45%,which can identify abnormal behavior well.The dissertation designs a multi-feature fusion parking lot pedestrian abnormal behavior recognition algorithm to solve the problem of low recognition rate of parking lot pedestrian abnormal behavior and poor real-time performance,and the feasibility of the proposed algorithm is verified in experiments.