Research On YOLO Pedestrian Vehicle Object Detection Algorithm And Implementation Based On Multi-scale Feature Fusion

With the continuous innovation and development of science and technology,the productivity of the automotive industry has been steadily improved.While improving people’s living standards,it has also brought obvious negative impacts such as traffic congestion.This has undoubtedly caused tremendous problems for travelers and traffic managers.In order to improve these phenomena,the concept of intelligent transportation system was proposed.Vehicle detection and pedestrian detection,as the core components of the intelligent transportation system,have become a hot research direction in the object detection subject because of their great practical significance.In recent years,the field of object detection has developed rapidly and has become increasingly mature,providing powerful technical support for pedestrian and vehicle detection tasks in complex traffic scenarios.A YOLO-based object detection algorithm is proposed based on the YOLOv3 network to optimize the network model and improve the target detection effect,which is based on the complexity of urban traffic scenes and the large variation in the scale and number of vehicle and pedestrian targets.In order to better reflect the application of the algorithm in real traffic scenes,pictures of different road sections in Shandong Province are collected,and the data are collected,labeled,and divided to build a traffic environment dataset PVD2020 based on Shandong Province.The large size of YOLOv3 prediction layer and inadequate feature extraction result in inaccurate prediction results.To solve this problem,a feature fusion network structure with five different scales is implemented to form a multiscale prediction network while enhancing the feature expression capability,which improves the network detection performance.Compared with the original YOLOv3 algorithm,the improved YOLOv3 algorithm achieves a higher mean average precision(map)on the PVD2020 dataset.Using the trained model,a traffic-oriented pedestrian and vehicle detection system is implemented to determine the traffic congestion by counting the traffic flow,pedestrian flow,and vehicle occupancy on a specific road.Finally,the system is tested in detail in terms of function and effect.The YOLO pedestrian-vehicle detection algorithm based on multi-scale feature fusion has achieved certain results on the PVD2020 dataset,which fully proves the superiority of the improved algorithm.