Research On Intelligent Monitoring System Of Expressway Tunnel Parking Incidents Based On Image Recognition

Real time online monitoring is very important to ensure the safety of highway tunnel operation and event management.Once a random and unpredictable accident occur in the tunnel,it Will cause traffic jams,worse is the second accident,pose a serious threat to the safety of people’s lives and property.Aiming at the complex environment of highway tunnel,the difficulty of vehicle moving target recognition and the difficulty of real time online monitoring,studied an intelligent monitoring system for expressway tunnel parking events based on image recognition.In order to realize on-line monitoring of parking target in complex environment of tunnel,and researched on background modeling and updating method,vehicle moving object extraction method,vehicle feature tracking method,parking object recognition method in complex environment and quadratic decision of parking events based on convolutional neural network.An intelligent monitoring system for expressway tunnel parking incidents based on image recognition was constructed,and the monitoring system deployed was applied to a section of expressway tunnel in Chongqing.The main research contents and innovations are as follows:1)Aiming at the problem of complex background modeling and updating of expressway tunnels,the characteristics of tunnel environment are analyzed.Compared with the existing typical environment modeling methods,a background modeling method based on non-parametric probability is proposed.Epanechnikov function is selected as the kernel function,and the probability distribution effect of the calculated local function is taken as the estimated value of sampling points.The experimental comparison shows that the modeler can effectively estimate the probability distribution of the local function.Compared with other methods,this method has better advantages in model effect and calculation speed.A multimodal background updating scheme is proposed based on the comprehensive analysis of the impact of environmental characteristics on background updating.The experimental results show that this method is better than the traditional averaging method and frame difference method in the case of external environmental noise interference,and can effectively suppress the interference of tunnel illumination change.2)Vehicle moving object extraction adopts background difference moving object monitoring algorithm,real time background image transmitted from tunnel camera tomonitoring center server is acquired based on environment background modeling and updating.Using current background image with previous background image get on difference and binarize the image,the target region coordinates are extracted by connected domain analysis and pseudo target filtering,finalize the calibration of dynamic targets with a rectangular frame.Experiments show that the algorithm is simple and robust,and can solve the problem of global light slowly changing and mutation in complex environment,and accurately detect vehicle moving targets.3)Based on the criteria of anti noise,distinguishability and size characteristics,a local feature tracking method is adopted to replace global feature tracking in vehicle feature tracking.By matching the features of different targets,we can recognize and track the visible features when the target is partially occluded.4)Aiming at the identification problem of parking objects in tunnels,based on vehicle feature tracking,a method for hierarchical recognition of parking objects was proposed.Based on the number of successful vehicle tracking,the initial recognition of parking target is carried out,and the final recognition is carried out according to the change of vehicle centroid coordinates.This method can effectively obtain the moving track of vehicle target and guarantee the recognition rate of parking.5)Aiming at the problem of video image interference and low detection rate of highway tunnel,a high speed parking event detection method based on convolution neural network is proposed.The second judgment is made on the alarm information obtained by the above methods.Suspected stationary foreground target position obtained by video processing is matched with the vehicle target and training sample obtained by image processing,realizing classification of static objects,complete static vehicle detection and alarm.6)Guided by the needs of monitoring system,based on development environment of VC++2015 and Open CV3.0,constructed a parking incidents monitoring system in tunnel environment with system modular design,and applied the monitoring system to the Chongqing-Hechuan section of the expressway tunnel to monitor parking incidents.Application research shows that the system has strong anti-interference ability,fast tracking and monitoring of targets and improve the system monitoring rate and correct.In summary,this paper develops an intelligent monitoring system for expressway tunnel parking events based on image recognition,and proposes the methods of background modeling and updating,vehicle moving target extraction,vehicle featuretracking and parking target recognition under complex tunnel environment.An intelligent monitoring system for expressway tunnel parking events based on image recognition is constructed and installation and application the Chongqing-Hechuan section of the G75 Lanzhou-Haikou Expressway Tunnel,which makes a positive contribution to the safe operation and management of the tunnel.