Moving Object Detection Algorithm Based On Stereovision And Markov Randow Field Model

In video image processing, the technology of the moving object detection is the first step, even more is the key step and the foundation of subsequent image understanding. There are wide application value in traffic,national defence and the man-machine interaction, and most of the sports scene is no longer simple, the change of lighting conditions, the shadow of sport goals, the shade of between the object and environment or between objects and the camera movement will influence inspection accuracy, so the moving targets detection technology has the urgency and necessity. This paper will be based on stereo vision and markov field model as the main research method. Main tasks are as follows:Firstly, in this paper, on the basis of the V-disparity theory, according to using stereo vision for detecting moving targets of large amount of calculation, this paper proposes a three resolution architecture stereo vision target detection algorithm. The algorithm calculates the foreground/backgroundt to the original stereo image and reduce the data quantity. And then established three layer gaussian gradient pyramid. Finally, calculated stereo disparity based on invariable SAD search range. Experiments show that adopt this method can achieve fast speed in extract obstacle information. But low reliability, lead to error detection, the reason is not high matching precision.Secondly, point to the above question, after further studies, using Markov Random Field(MRF)model for stereo matching. It introduces a data function that integrates gray with gradient to build a MRF global energy function,and uses belief propagation to minimize the global energy function in order to get a disparity map,which is then calibrated by MRF causal systems.Finally, this paper proposes a moving object detection algorithm based on spatiotemporal markov model and edge information. Pointing at light cases, use the combination of the edge information and difference of three frame method. Pointing at the calculating of MRF energy, use the iterated conditional model to find the image's least energy. At last, to deal with the threshold segmentation and morphological. Experiment results show that the algorithm is effective.