Research For Image Anomaly Detection Of Traffic Monitoring Video And Key Information Reconstruction

In recent years,with the leap-forward development of the urban construction concept,the infrastructure of public transportation has been developed rapidly,and various types of monitoring equipment have increased significantly,but the potential problems in the monitoring system also emerge accordingly.Due to the influence of various realistic factors,surveillance video images often appear some abnormal states in the actual operation of monitoring system,which results in a decline in the quality of video,and seriously affects the normal operation of the system,not only making the monitoring system cannot play its proper role,but also hardly conducive to the extraction of incident information.In addition,considering the cost of the system,surveillance shooting,video transmission,storage and other objective constraints,some of the monitoring video resolution is not high.The key information related to public safety incidents often occurs in these monitoring scenarios,which has serious impact on the relevant work of the traffic control personnel.This thesis developed an image quality detection system for the traffic monitoring video,and also constructed a method of key information reconstruction based on super-resolution.(1)This thesis studied the characteristics of these common abnormal video images in the spatial domain,and developed an image anomaly detection system for the traffic monitoring video,to achieve the purpose of automatic detection for video quality during the monitoring system,and accordingly provided the result of abnormal style as the basis for manual examination.(2)Aiming at the low resolution monitoring scenes related to public security incidents,this thesis constructed a new method of hybrid super-resolution reconstruction based on adaptive weight,to reconstruct the key information of the image without video quality.This method takes advantage of the prior knowledge of external training data and internal self-similar data,and computes an adaptive weight to select the suitable method to minimize super-resolving error,which fully mixes the advantage of the both methods.(3)Based on the actual traffic surveillance video data,this thesis tested stability and operability of the designed system.It studied some typical super-resolution methods and also compared their advantages and disadvantages.The effectiveness of the proposed method was verified by comparing the experimental results in some representative dataset.