A Real-time Detection System For Abnormal Events Of Highway Based On Image

Based on the problems in the real-time monitoring scene of expressway,the traditional manual monitoring method is difficult to detect the occurrence of abnormal events immediately,such as inaccurate detection and high monitoring cost,which makes it impossible to distinguish the running state of traffic efficiently.So,the improved YOLOv3 algorithm is emerged and is verified by experiments.It further improved the detection accuracy under the condition of real-time detection.The vehicle tracking speed is greatly improved by improving the vehicle feature description in deep-sort algorithm.Based on the improved vehicle object detection,tracking algorithm and the real-time monitoring image of expressway,the system finally realizes the real-time detection of vehicle about speeding and congestion.Firstly,the monitoring system of the expressway company was investigated on the spot.After collecting the monitoring videos of the expressway in different sections,lighting,weather and other environments,a complete vehicle data set was established,which provided the data basis for the construction of a deep learning model for vehicle target detection.Secondly,the labeling tool is used to classify and mark all the samples and divide the data set.Based on the study of vehicle target detection algorithm,the loss function of YOLOv3 was improved in the border regression approach by the improved vehicle detection method.Then through the experimental analysis of the different loss function,the scale of training data to build the model of vehicle detection precision verify the model of the algorithm and the upgraded performance and speed of target detection.After the traditional target tracking method is expounded,the deep-sort algorithm based on real-time multi-target tracking is improved.Experiments verify the improved algorithm.In the case that the detection effect is basically the same,the real-time performance of the anomaly detection system is enhanced by improving the tracking speed.Finally,based on the research results of vehicle's target detection and tracking,an anomaly detection algorithm is defined.Taking abnormal events in surveillance video images of expressway as an example,the debugging and analysis of abnormal event detection system can automatically detect the speed and congestion of the video vehicle in the abnormal events;it also achieves the real-time and high accuracy of abnormal event detection.In doing so,this system improves the collection of the traffic information in the highway and the ability of capturing emergency signs in advance.It also provides the strong support for timeliness and on-site rescue,which guarantees the state of the lane traffic running well and improves the overall service quality of highway company.