A temporal smoothing technique for real-time motion detection

Abstract

Motion detection can be achieved using a comparison between two or more image sequences. Reference frame differencing, which needs a valid reference frame, can be used for that purpose. This paper addresses the application of a temporal smoothing technique to segment images of moving objects provided by a static camera. A valid reference image is built and updated every video frame by combining it with the incoming image on a pixel by pixel basis. The image processing techniques are discussed, performance evaluation experiments are reported and some field trials demonstrating the system ability to monitor a real world traffic scene are referred. The system configuration is briefly explained, with particular emphasis given to describe the software and hardware modules which maintain the background reference frame and perform image segmentation.