Real-time robust vehicle flow statistics based on adjacent frames clustering

In this paper, a clustering method of adjacent frames is proposed for vehicle flow statistics to overcome the fault of low robustness of video-based detection algorithms in complex environments. In the method, the boundaries of the abrupt or gradual visual content changing in consecutive video frames are described by color and intensity histogram method. The clustered frames containing different vehicles are segmented by these boundaries. In order to enhance the robustness of the algorithm, the interaction of adjacent multi-frames is weighted by gauss function which is the function of the interval between frames. Vehicle flow statistics are accomplished by detecting local maxima of the objective function. Experiments show that this method works stable in different weather and traffic status with the real-time performance of over 30 frames per second and with the mean detection precision of 93.2%.