The Key Technology Of Automatic Video Recognition Of Vehicle Illegal Activities Based On Machine Learning

With the expansion of Chinese cities and the rapid development of road traffic in recent years,my country’s car ownership has shown an explosive growth trend,which has led to frequent vehicle violations.Based on this background,the intelligent traffic video image processing system can effectively reduce traffic.The labor cost of monitoring improves the detection accuracy of illegal activities and simplifies the determination process.Aiming at the core technical problem of automatic video recognition of vehicle illegal activities,this paper enters into the method of machine learning,and conducts in-depth research on vehicle target detection and positioning,vehicle trajectory tracking,vehicle typical illegal behavior determination and prediction capture algorithms and gives a complete system achieve.This paper studies the vehicle target recognition and location method based on the YOLO v3 target detection model in a traffic monitoring environment.The detection accuracy and computational efficiency of the algorithm for target detection have a significant impact on the system.The YOLO v3 detection network used in this paper is on the same network The classification,recognition and positioning calculation of the target are given in the,and it is not affected by the environmental parameters of the lane,and excellent detection accuracy is achieved.This paper studies the algorithm implementation strategy of vehicle trajectory tracking in traffic monitoring,and presents a calculation method that combines the maximum area matching of the associated frame and the improved Kalman filter algorithm to complete the calculation,tracking and matching of the vehicle target trajectory,based on the area The matching algorithm guarantees the algorithm’s excellent real-time calculation performance and response speed,and the designed improved Kalman filter algorithm can effectively solve the problem of vehicle target missed detection.After algorithm calculation,the complete historical trajectory of all vehicle targets in the surveillance video can be accurately obtained.This paper studies the judgment and calculation strategies of typical illegal behaviors of vehicles in actual traffic monitoring scenarios,mathematically modeling the lane background and typical illegal behaviors of vehicles,and effectively fits the boundary of the lane and the boundary of the region of interest through the least square algorithm Based on the position relationship between the vehicle position and the boundary,the identification characteristics of the typical illegal behavior of the vehicle are defined.The calculation method designed takes into account the calculation efficiency of the algorithm and achieves excellent detection accuracy.This paper studies the target prediction method based on the historical trajectory of the vehicle target in the pixel coordinate system.By selecting the coordinate points in the historical trajectory of the vehicle in line with the trend of speed change,combined with the derived fractional model mathematical function,the overdetermined equation set can be formed to calculate The predicted position of the vehicle in the surveillance video at the future time,and the surveillance dome camera is mobilized to zoom to the preset preset position in time to realize illegal capture.This article presents a complete set of hardware and software systems for automatic video recognition of vehicle target violations.The software system includes an initialization module,a vehicle target detection model,a trajectory tracking module,a vehicle violation determination module,and a predictive capture module.For vehicle prediction and capture,we have developed a visualized automatic observation module,which can provide intuitive and accurate observations when illegal acts occur,and the test results are good.