Research On Vehicle Illegal Stay Detection Based On Improved YOLO3 For Surveillance Video

In recent years,with the rapid development of big data,artificial intelligence and other new generation information technology,intelligent transportation system has been widely used and popularized in the world.This is because the system can reduce the cost of human resources,avoid human errors,and automatically identify abnormal situations.At present,the number of urban vehicles is increasing and the traffic congestion is serious.In order to monitor the abnormal situation,monitoring equipment is usually arranged in public places.However,a small number of drivers are still parking in the no parking area,resulting in more serious road congestion.Intelligent parking detection technology can give full play to the monitoring role of intelligent transportation system,which plays an important role in improving traffic environment and coping with road congestion.However,at present,the monitoring system mainly adopts the way of manual comparison and verification,which requires the staff to stay in front of the monitoring screen all day,which not only causes the waste of human resources,but also prone to human errors,so it is of great significance to give full play to the role of intelligent parking detection technology.This paper analyzes the parking detection algorithm in the outdoor surveillance video,pointing out the shortcomings of the main monitoring algorithm in the use process in the emerging stage:for example,shadow interference is not considered;it is difficult to detect the vehicle staying for a short time in the static target detection;it is difficult to determine whether the foreground object is a vehicle in the way of geometric shape,resulting in low detection accuracy.This paper summarizes the theory of outdoor monitoring video parking detection technology,combs the technical difficulties in the current practice of video parking detection,and improves the existing parking detection algorithm.This paper analyzes the parking detection algorithm,including the following five modules:1.Extract foreground target.When extracting foreground objects,the moving objects in the foreground are obtained by combining the adaptive Gaussian mixture model;2.Detect video shadows.Based on the HSV color space model,the shadow in the foreground is removed to obtain more accurate foreground target;3.Detect still targets.Combined with the historical foreground pixel method,the static targets can be formed firstly,and then the suspicious static regions can be formed according to these targets.The comparison between before and after N frames is made,and the judgment is made based on the improved hash perception method to determine whether they belong to the static targets,Therefore,compared with the conventional method,the static target detection method designed in this study can effectively detect the vehicles staying for a short time.4.Detect foreground vehicles.The accuracy of vehicle detection is improved by improving the spatial pyramid merging unit.5.Detect occlusion.Based on the mixture gauss model,the occlusion detection method is proposed to further improve the success rate of video detectionThe experimental results show that the detection rate of this method on the data sets of iLIDS and UA-DETRAC is 96.58%.