Research On Abnormal Behavior Detection Algorithm Of Dangerous Chemical Transport Vehicle Based On Road Surveillance Video

Abnormal behavior detection of hazardous chemicals transport vehicles has great application value in intelligent traffic management,illegal vehicle investigation,accident early warning and other fields,among then,vehicle recognition,vehicle tracking and abnormal driving behavior detection in road surveillance scenarios are the focus of research,there are still many problems to be solved in abnormal behavior detection of hazardous chemical transport vehicles.Based on the research of highway surveillance video,this paper carries out in-depth research on key technologies such as detection and identification of hazardous chemical transportation vehicles,vehicle tracking,abnormal event detection and so on.The research contributions of this paper are as follows:(1)By analyzing the vehicle detection principle of Faster R-CNN model in detail,the feature extraction network,candidate region generation algorithm and ROI(Region of Interest,ROI)pooling layer in the original model are respectively improved.In the feature extraction stage,a dense connection network is introduced to replace the original feature extraction model,which enhances the fluidity of features and makes the model suitable for vehicle detection tasks with complex background.In the process of generating candidate regions,the number of candidate boxes generated is reduced by half through the prior knowledge of the data set,which greatly improves the detection speed of the model.Finally,bilinear interpolation algorithm is used to optimize the error of target frame mapping and improve the positioning accuracy of the target.(2)The identification and tracking model of hazardous chemical transport vehicle is constructed.According to the sign lights in the appearance characteristics of hazardous chemical transport vehicles,a feature pyramid network based on dense connections was proposed,by integrating shallow features and deep features,the extraction efficiency of small target features is enhanced.At the same time,the dense connection network further strengthens the fluidity from shallow features to high-level features,and solves the difficulty in identifying hazardous chemical transportation vehicles in video sequences due to the small scale of the sign lights.Then,based on the vehicle detection model,a vehicle tracking algorithm based on Deepsort was designed to track hazardous chemical transport vehicles.(3)The abnormal behavior detection algorithm of hazardous chemical transport vehicle is designed and implemented.First of all,typical abnormal driving behaviors(such as illegal parking,reversing,etc)are detected according to the characteristics of the track sequence according to the predictable abnormal behavior pattern.For the unpredictable abnormal behavior pattern,the agglomerate hierarchical clustering algorithm is used to obtain the normal behavior model of the corresponding road section.Finally,the abnormal driving behavior of the vehicle is detected and tracked by measuring the similarity between the trajectory sequence to be detected and the normal driving model.