Moving Object Detection Based On Gaussian Mixture Model

Moving object detection based on image sequence is one of the important research orientations in the field of computer vision. It is the fundamental and pivotal technology of intelligent surveillance, human-machine interface, mobile robots navigation, industrial robots hand-eye system, etc. This thesis focuses on the background modeling and moving object detection, and lays strong emphasis on Gaussian Mixture Model (GMM) and shadow removing.The main contributions of this thesis are as follows:(1) Some typical background modeling algorithms are compared on the basis of many experiments and a conclusion is presented. It elaborates on the merits and weaknesses of these algorithms from three aspects: processing speed, memory requirement and detection effect.(2) A systematical deduction and analysis of GMM is carried out, which is valuable for relative research.(3) An improved algorithm of moving object detection, based on GMM, aiming at intelligent surveillance system using fixed-camera, is proposed. This algorithm adopts an improved parameter adaptation mechanism, which can prevent the variance of GMM from decreasing endlessly and reduce the computational complexity of GMM.(4) A shadow removing algorithm based on color space transformation and GMM, aiming at intelligent surveillance system using fixed-camera, is proposed. Firstly, it carries out a moving object detection in image sequences. Secondly, it transforms the pixels that have been regarded as foreground from color space RGB to HSV. Thirdly, it uses a decision mechanism to detect the suspicious shadow pixels. Finally, it makes use of the classification ability of GMM to identify the real moving object pixels from those have been transformed to color space HSV. The experiment result shows that this algorithm can effectively remove the shadow cast by moving object and maintain acceptable computational complexity.