A Study on object detection in complex background

Abstract Recently, a demand for video serveillance system using surveillance cameras has been increasing because of the uneasiness to secure the public peace. At present, It is often the case that people look video transmitted to a monitoring center. A variety of video recognition technologies to find abnormality are developed, and demand for a smart surveillance by the machine is increasing. Though object detection is an important element of smart serveillance system, the detection accuracy is often degrated in a complex background. This kind of situation can be seen in the environmental change like shakes of the tree. In this paper, we propose a method of object detection in such a complex environment. First, regions in the image is divided according to a similar level of the color information and the direction of the motion vector. Then, the difference of the luminance value in each area between frames is calculated. Finally, forground regions are detected using a threshold. Experimental results show that the proposal method can detect a moving object effectively.