Fast Moving Detection Method Based On Multi-channel Imaging

Moving object detection based on image sequences is widely used in many areas,such as production controling, medical testing and intelligent transportation. In recent years, with the development of computer vision, moving object detection in image sequences tends to be real-time, high precision and large field of view detection. These themes have become hot topics in many researches. In order to solve the problem of poor real-time detection performance, this paper proposed a new method based on the multi-channel bionic compound eyes system to accelerate the detection speed. The main achievements of this paper are as follows:First,this paper study the features of multi-channel conpound eyes system, and conect the features which were captured by SIFT algorithm to achieve the goal of scenes ranging and high-resolution reconstruction. Second, Analyse three different optical flow detection methods. Achieve rapid moving object detection by Farneback optical flow under GPU parallel programming. Last, this paper proposes a multi-channel variable-resolution method for the fast detection of moving object based on the multi-channel imaging system model. By taking full advantages of the motion-sensitive characteristics of the sub-channels, determine the “general area” in the sub-channels quickly.This step include rough moving object detection and image morphology. Then detect the “general area” accurately in the main channel by Farneback optical flow.This method is tested by experiments in laboratory and simulations with out-door videos.The results show that the variable-resolution detection method proposed in this paper is much faster than the traditional detection methods based on single-channel system.This paper uses the GPU parallel computing mechanism to deal with multi-channel image arrays,and improves the speed of moving object detection significantly. This kind of research is meaningful in various areas.