Research On Traffic Flow Detection Method Based On ARM

The traffic flow detection methods through surveillance video includes target detection technology and multi-target tracking technology,because of wide variety of objects in real-time traffic scene,accurate and real-time identification and statistics of vehicle targets have become a great challenge for video traffic detection technology based on embedded systems.In this thesis,three kinds of vehicle detection methods were successfully transplanted by using the RK3399 of ARMv8-A architecture as the detection platform: frame subtraction method,You Only Look Once Version 2(YOLO v2)detection method,Haar Adaboost classifier detection method,combined with threshold and Kernelized Correlation Filters(KCF)two tracking algorithm,to achieve the detection of traffic flow,and researched the real-time and accuracy of the three methods;Try to parallelize the Haar Adaboost classifier detection KCF tracking method and test the performance after parallelization;Finally,the system was tested at the road site.The main contents of this thesis are as follows:(1)Build the compilation environment and algorithm execution environment.Use the compiler to cross-compile the configured U-Boot,ARM-Linux kernel,file systems,and the software libraries on which the algorithms depend,and then port them to the embedded system.(2)The vehicle detection algorithm,tracking algorithm,counting three parts were combined to realize the detection of traffic flow,and all the algorithms were cross-compiled on the computer,and then the algorithm was transplanted to the embedded system.(3)Three algorithms were tested and compared in embedded systems through road scene video.In order to further improve the detection speed of the system,the Haar Adaboost classifier detection KCF tracking method was parallelized because of high accuracy and execution speed,and the region of interest was divided according to lanes,and each region of interest was allocated to each sub-thread,After the sub-thread completes the entire detection process,the detection result was transmitted back to the main thread,and the main thread counts the data of each sub-thread and displays the final result.(4)In the actual road scene,the traffic flow detection system that transplanted the Haar Adaboost classifier detection KCF tracking method use camera real-time recorded,and the performance test was carried out.