Recognition And Warning Of Abnormal Behaviors:Outside Vehicles Based On Computer Vision

In recent years,the development of intelligent traffic assisted driving has been particularly rapid.The assisted driving technology can effectively guarantee the safety of roads and reduce the probability of traffic accidents,and to protect people's personal safety and property safety to a certain extent.Therefore,this paper adopts multi-target detection technology and multi-target tracking technology,combined with monocular vision ranging algorithm and early warning decision algorithm to complete the design and development of the abnormal behavior recognition and early warning system under the natural scene.The system can accurately Real-time analysis and evaluation of road conditions,timely warning of dangerous situations,and experiments have proved that the system has high application value.This paper mainly carries out related research in the following aspects.(1)In the aspect of multi-target detection,the advantages and disadvantages of YOLO series target detection algorithm are analyzed in detail.The problem that YOLOv3 target detection algorithm cannot be detected quickly on common computing platform is proposed.Based on residual block and FCP module,this paper proposes The RP-YOLO vehicle and pedestrian target detection algorithm adds a residual block to the Darknet53 network to form the Darknet65 network model.The FCP module is connected to the fully connected layer for category prediction and position regression.Through experiments,it is verified on the dataset that the improved RP-YOLO vehicle and pedestrian target detection algorithm have higher detection speed and meet the requirements of target detection accuracy.(2)In the aspect of multi-target tracking,firstly,the Deep-SORT tracking algorithm and existing problems are analyzed in detail.For the Deep-SORT multi-target tracking algorithm,target occlusion and target exchange are prone to error tracking and loss tracking in the vehicle camera scene.In order to better improve the correlation matching between the target detection frame and the tracking prediction frame,the apparent feature information based on the spatial color histogram is added as the matching basis in the association matching phase of the Deep-SORT algorithm,and experiments show that the fusion space color The histogram's Deep-SORT multi-target tracking algorithm can effectively reduce the tracking failure caused by target crossover and target occlusion,and better realize multi-target tracking of vehicles and pedestrians.(3)In the design and development of abnormal behavior recognition and early warning system outside the vehicle,firstly analyze the requirements of the system,divide the functional modules,and then design the system interface,then from the target ranging unit,information storage unit,risk assessment and early warning unit.The implementation details of the vehicle distance monitoring and early warning system are introduced in detail in three aspects.Finally,the system test proves that the system can well identify the sudden braking behavior of the suspected vehicle and the pedestrian's collision behavior in the abnormal behavior outside the vehicle.Early warning.