Research On Abnormal Behavior Detection Algorithm Of Dangerous Chemical Transport Vehicle

With the increasing transportation of hazardous chemicals,the potential risks are also increasing rapidly.The intelligent transportation of hazardous chemicals road safety vehicles has great application value.The key supervision of hazardous chemicals transportation vehicles is of great significance to ensure the safety of people’s lives and property.Given the characteristics of hazardous chemicals vehicles,this paper proposes an improved YOLOv7(named YOLOv7-DM)and DeepSort vehicle tracking algorithm,using a dual detection strategy(detecting external features and small logos at the top of the vehicle)to address the poor detection results of hazardous small logos on the top of the vehicle.Meanwhile,the position information of the target vehicle and clustering algorithms are used to detect differentiated abnormal behaviors of hazardous chemicals transportation vehicles.(1)Improved vehicle detection using the YOLOv7-DM model.Based on highway surveillance video,the Dangerous chemicals transport vehicles-L(DCTV-L)dataset was created for hazardous chemicals transportation vehicles.This paper uses YOLOv7 as the foundation and introduces the attention mechanism CPA and RepGFPN module to construct a sub-dense connection in the neck network,which enhances the fusion of different level features.At the same time,Q-fusion fusion method is used to reduce the model’s delay.On the DCTV-L dataset,the model achieved an accuracy rate of 99.1%,a precision rate of 98.9%,and a detection speed of 47 frames per second.(2)Vehicle tracking model.Based on the characteristics of hazardous chemicals transportation vehicle data,this paper improves the re-identification network model based on the object detection and uses the DeepSort algorithm to extract better features of the target,achieving more stable tracking of the target vehicle.The tracking accuracy in different scenarios reached 83.7%,the tracking precision reached 89.6%,and the tracking speed reached 41 frames per second.(3)Designed a method for detecting abnormal behaviors of hazardous chemicals transportation vehicles.For regular and predictable abnormal behaviors,the vehicle’s trajectory is determined by its coordinate information and its position relative to the lane lines.For irregular and unpredictable behaviors,clustering algorithms are used to determine the abnormal driving behavior.The distance function used in this paper achieved an overall accuracy of 94%.Finally,a user-friendly interface was designed using QT software for easy operation by staff.