A Video Surveillance Detection of Moving Object Using Deep Learning

Moving object detection is a fundamental and important research field in computer vision because this technology can be used for many applications, for example, video surveillance, tracking, and abnormal event detection. Most traditional moving object detection methods first establish a background model, and then subtract the original image to obtain the foreground object. However, this process encounters challenges under certain complex conditions, such as changes in lighting, dynamic backgrounds, and shadows, which break up the foreground and produce undesirable noise. This paper proposes a new two-stream neural network which combines the traditional background modeling method with a deep learning network to detect moving objects. The input for the deep neural network is the original image and its corresponding foreground image, while the output is the bounding boxes of the moving objects in the image. Traditional CNN methods cannot distinguish moving objects from static objects, but the method in this paper successfully solves this problem.