Research Of Vehicle Flow Statistics Based On Video Images

Vehicle flow statistics is an important research content in the field of intelligent transportation.For the problems of long testing time and low detection precision existed in traditional statistical methods,a system of vehicle flow statistics based on video images is studied and implemented in this thesis,and the realization process includes vehicle detection,vehicle tracking and statistics.The specific research work is as follows:1.Vehicle detection based on background subtraction.Firstly,the initial background model is established with the statistical average method.Secondly,the background is selectively updated through an improved recursive algorithm,specifically,the corresponding points are updated except the points on the foreground areas.Finally,the foreground object is extracted through subtracting the updated background frame from the current video frame.2.Vehicle tracking and statistics based on video images.The vehicle flow statistics are implemented in this thesis through the tracking method based on combining centroid feature with Kalman filter and setting the counting line.The specific process is as follows,firstly,the tracking object and its location are determined according to the features of centroid and area in the foreground object,consequently,the object matching is implemented through comparing the location of the object with the one predicted by Kalman filter,and the counter is started as the successfully-matched vehicle goes through the counting line.The above methods are initially implemented on the the Visual Studio 2008 platform with OpenCV2.4.3 library,in order to improve the practicality and portability of vehicle flow statistics,the vehicle flow statistics system has been successfully transplanted to Android development board with the JNI technology.The results of experiment show that,for the road videos with 352×240 resolution and 25 frames per second,the average accuracy of vehicle detection is 97.1%,the average accuracy of vehicle Statistics is 96.3%,and basically meets the real-time requirements.