Research & Implementation Of Moving Object Detection Based On Fish-eye Camera

With the constant development of society, people's inhabitation space becomes more and more crowded, and emergent events and abnormal events are also growing day by day. It is important to find an efficient way to reduce or avoid loss by monitoring. Therefore, states and scolars over the world focus on the second generation monitoring technology. For the purpose of management and security, people should know immediately what's happening in some particular districts, such as banks, militaries, political agencies and departments like that, and crossroads, airports, markets, parking lots, commercial buildings and places like that, and react to the emergency as fast as possible so that remedies produce effect. This is monitoring as we know. Since 9.11 terrorist attack in America, the world have put more and more emphasis on the question how to make all-day, automatic and real-time monitorings in important national security departments and public areas. Intelligent video system is just the efficient way of dealing with that problem.Intelligent video system mainly applies moving detection algorithm based on fish-eye camera in this paper, and the paper focuses on fish-eye image correction algorithm, background modeling, removing shadow when detecting moving object.First, a fish-eye image correction algorithm based on equidistant spherical projected is presented in this paper. The scanning line approach algorithm is applied to determine the optical center and spherical radius of image. Then 2D fish-eye image is projected to 2D without distortion image by equidistant spherical projected, which get the image that is common image from people's view. Thus the moving objects detection become more easily and accurately.As the camera is fited, the fish-eye image which has been corrected takes advantage of improved mean value method to extract moving foreground area. The paper also uses normalized cross-correlation function and image's texture feature to inhibit moving foreground's shadow area, and finally detects and gets the moving object area.The results show that the algorithm can support C100 and C105 camera, and ffectively detect moving foreground for background modeling in the majority of scenes. At the same time, adopting the shadow inhibit algorithm of texture feature based on normalized cross-correlation advoid shadow's effect, improve the accuracy of detcting the moving object, and achieve the more satisfied result of moving object detection based on fish-eye camera.