Research On Moving Objects Detection Algorithms Based On Background Subtraction

Moving objects detection extracts interesting moving objects in the images from an image sequence. It is the first step in an auto detection system, and is the fundamental work for objects classification and objects tracking. In this paper, we focus on the research of detection algorithms which are based on background subtraction, and propose an improved background subtraction algorithm using adaptive Gaussian mixture models. The main work of this paper includes:(1) Introducing some typical detection algorithms which are based on background subtraction, then analyzing and estimating the performance of these algorithms in detail by comparing.(2) After the comparison of the typical algorithms, we choose the adaptive Gaussian mixture models as the main algorithm for motion detection in complex backgrounds. Some improvements are added to the model to improve the performance of the algorithm, making it more real-time, and more reliable.(3) A square neighborhood matching algorithm is put forward in this paper to compensate the motion of the camera. The algorithm matches some sample pixels to their neighborhood to get the motion information of the camera, and then adjusts the whole input image with the motion information, making the image prepared for detection.(4) Based on the fact that shadows are relatively stable in that their values have little to do with the objects that cast them, we propose a method to eliminate shadows, which presents each shadow of a pixel with a Gaussian model. The speed of the algorithm is relatively fast, because only a little extra calculation is needed to detect and eliminate shadows.(5) In the post-detection process, we take a step further to remove some more noise from the moving areas that are obtained with background subtraction, thus lower the false positive detection, making the detection more accurate. In addition, we use the math morphology techniques to cure the target areas.