Research On Dangerous Feature Detection Technology Based On Video Surveillance

The country's economic conditions directly affect people's travel methods.Motor vehicles have almost entered millions of households.While providing convenience for our travel,it has also brought a heavy workload to the traffic management department.In order to reduce the occurrence of traffic accidents,in the field of intelligent transportation,how to achieve accurate detection of vehicle dangerous characteristics in video surveillance has become a hot issue.Because traditional human monitoring wastes a lot of time and energy,it is of great significance to promote the research of dangerous feature detection systems on urban roads as intelligent transportation systems.Aiming at the problems such as incomplete target detection by the traditional Vi Be algorithm,an improved Vi Be algorithm based on gray-scale projection motion estimation is proposed to obtain a more complete and clear foreground target.Aiming at scenes where small target detection is difficult,an improved GDT-YOLOv3 high-precision target detection algorithm is designed using deep learning technology.In order to more accurately judge the occurrence of dangerous behaviors of vehicles,the trajectory data obtained after vehicle tracking is analyzed,and the criteria for judging vehicle retrograde,lane changing and turning are proposed.The main work and innovative research results of this paper are as follows:1)In order to reduce the interference of background information on target detection and improve detection accuracy,this paper proposes an improvement of Vi Be algorithm based on gray-scale projection motion estimation.This algorithm can separate the foreground and background information of the panoramic video image,which is helpful for subsequent Object detection and classification.The experimental results show that the improved algorithm in this paper has good advantages in eliminating ghosts and resisting interference from dynamic background models.2)In order to solve the problems of low accuracy and slow speed of detecting targets in video images.This paper proposes an improved GDT-YOLOv3 target detection algorithm.First replaced the IOU with GIOU,which can solve the problem that the original IOU cannot directly optimize the non-overlapping parts.Second,after drawing on the idea of densely connected networks,replacing the three residual blocks in YOLOv3 with three dense blocks can realize the multiplexing and fusion of features.At the same time,Max Pooling was introduced to enhance the transfer of features between densely connected blocks..In terms of detection accuracy,the improved GDT-YOLOv3 performs better than SSD512,YOLOv2,and YOLOv3.3)In this paper,when vehicles with dangerous characteristic behaviors are detected,the Kalman filter algorithm is used to track the vehicles.The vehicle's position information and motion trajectory can be determined.The obtained trajectory information is combined with three types of dangerous behavior judgment rules to accurately detect Dangerous characteristic behavior to the vehicle.The experimental simulation results and data can prove that the judgment method in this paper can be applied to the detection of dangerous events in the actual road environment.