Research On Moving Objects Detection Based On Background Subtraction

Moving object detection is one of the key techniques for automatic video analysis,especially in the domain of video surveillance. As a classic and effective algorithm for motion detection, background subtraction has been widely applied. Various background subtraction methods have been developed, with each of them having its own characteristic, strength and weakness. The purpose of this paper is to improve the effects of codebook and ViBe(Visual Background Extractor) detection algorithm in motion detection. The main tasks of this paper are as follows,Firstly, this paper introduces some typical detection algorithms which are based on background subtraction, then analyzes and estimates the performance of these algorithms in detail by compare.Secondly, to solve the problem of low ability in adaptive dynamic background and low accuracy of detection under complex environment, this paper proposes a novel adaptive codebook target detection algorithm based on regional information analysis. Firstly, it utilizes adding two learning rates in order to update adaptively the background model in the presence of background motion. Secondly, it combines the observable codebook with neighboring codebooks to obtain more accurate moving targets. Lastly, it models and updates the foreground and background respectively, and converts them to eliminate false detection caused by the partial change of background. The experimental results show that the improved algorithm has higher recognition rate and better robustness even in complex environment conditions with illumination variation and random noise. It basically meets the demands of moving object detection in dynamic scenes in terms of precision, speed, noise resistance, light adaptability and so on.Thirdly, considering the design limitation of ViBe algorithm, this paper combines with the advantages of some other algorithms, and proposes a novel target detection algorithm based on improved ViBe model. In the algorithm, first of all, it joins the time samples to the space samples. Secondly, it utilizes adding the TOM(Time of Map) in the process of updating the background model in order to fuse quickly the changing foreground and background. Finally, it adds a shadow removal and morphological connected component inthe post-process to get better and more complete moving objects. Several experiments are implemented and the results demonstrate its effectiveness in detecting the moving objects in the complicated condition.