Moving Object Detection Algorithm In Video Sequences

With the rapid development of computer technology, the real-time video surveillance technology applied in real life has been made possible. In the traffic environment, the technology, such as monitoring the flow of a road traffic, positioning of the vehicle target accurately, real-time measuring the vehicle speed, all will be beneficial for the recognition of the vehicle and its license plate. Applying the video surveillance system in a intelligent building can improve the building’s safety index and avoid the waste of manpower and financial resources effectively. In terms of military, target detection technology in radar technology can detect some unknown objects in the air and make warning in time. Therefore, it is valuable to study the technology of moving target detection in video sequences.Theoretically, the moving object detection has good application prospects in many areas. However, with the influence of various random factors in natural scene, the background certain areas could be detected as the moving target, which cause error detection. Most resent methods for moving target detection were based on the ideal background conditions and with unsatisfactory detection effect. In this thesis, with the analysis and summary of the existing detection technology, we present three kinds of histogram-based detection algorithms for the target detection in a video sequence captured with fixed monocular camera. The first algorithm is based on histogram similarity matching, which mainly detect target by the difference of histogram similarity between background sub-block and target sub-block in adjacent frames. This algorithm use the histogram directly, which is relatively simple but with poor experimental effect. The second algorithm is based on the joint histogram. With the standpoint that the similarity between adjacent frames could be described by the joint histogram, the joint histogram and constructed similarity index are used for removing the background block gradually with the purpose of detecting target. The last method in this thesis is based on successive block difference histogram, in which the different distribution of the difference histogram between the background region and the target area are used to divide the two adjacent frames and difference image into blocks successively, and the distribution of the difference histogram in each block is analyzed to remove background gradually, and then the moving targets is detected. Numerical experiments show that the latter two algorithms can detect the complete moving object, and avoid the void appeared inside moving target, meanwhile the pixel changes in the background caused by factors(such as the shaking branches) can also be removed commendably, which is beneficial for the further analysis of the target tracking.