Vehicle Detection And Tracking In Video Intelligent Surveillance Technical Research And Vehicle Classification System Design

With the rapid development of my country’s economy,culture and technology in recent years,people’s living standards have been greatly improved.As the most common and most important means of transportation,automobiles carry the huge mission of material and cultural exchanges.The number of motor vehicles has also shown an increase of geometric orders of magnitude,making transportation planning and management facing with huge challenges,reliable and efficient supervision of motor vehicles has become an urgent problem to be solved.With the development of my country’s security industry,video surveillance of road vehicles has gradually become a mainstream supervision method.However,the traditional method of recording vehicle information in surveillance videos through manual statistics by staff is cumbersome and inefficient,and it is difficult to deal with long-term vehicles.Time and intense work,so the use of computer vision related technology to intelligently monitor vehicles has become a new research hotspot.Vehicle detection,vehicle tracking,and vehicle classification technologies are the basis for vehicle statistical records and behavior analysis.Therefore,this thesis focuses on vehicle detection,tracking and classification in video intelligent surveillance as the main research content,and has the following research work:(1)In the vehicle detection research,this thesis analyzes the shortcomings of traditional foreground detection algorithms when used in vehicle detection scenes,improves the calculation method of the background update rate of the model,and the overall update strategy.And through the morphological optimization of the foreground mask image and the calibration of the foreground vehicle contour,the vehicle area position is determined to realize the detection of the vehicle.Finally,the vehicle detection comparison test is carried out with other foreground detection algorithms.The results show that the improved gaussian mixture model modeling method in this thesis performs well in the test scene,especially in the slow-vehicle traffic jam scene,which has obvious advantages in vehicle detection.(2)In the research of vehicle tracking,this thesis proposes a multi-vehicle tracking method based on centroid point matching.In order to achieve the matching of the detected vehicles in the front and rear frames,the position relationship between the front and rear frames when multiple vehicles are driving is analyzed,and the matching strategy for the same vehicle is given.The applicable conditions of the matching strategy and the applicability in general monitoring scenarios are analyzed and provided process steps for vehicle tracking.Finally,the tracking effect and processing speed in the multi-vehicle monitoring scene are tested.The results show that the tracking effect is good in the multi-vehicle tracking scene,and the algorithm calculation is small,which meets the needs of real-time tracking.(3)In the design of the vehicle classification system,apply the vehicle detection and tracking methods mentioned above,combined with the classification method based on image features,design the overall system scheme of vehicle detection,tracking and classification of road video surveillance scenes,and give the specific system implementation process.The test was carried out with road surveillance video,and the test results of vehicle detection and classification were analyzed.Regarding the deficiencies that still exist in the system test,the future improvement directions and suggestions are provided in the outlook section.