BACKGROUND SUBTRACTION METHODS FOR MOTION DETECTION IN VIDEO SURVEILLANCE SYSTEMS

A main activity of many computer vision systems is foreground object detection. Motion is used in a video sequences to identify target objects. Many background subtraction methods have been proposed to identify motion that belongs to a moving object. Each has advantages and disadvantages when it is applied in different conditions. An overview of five background subtraction methods is proposed in this paper. The selected methods described and compared are: temporal averaging, frame differencing, Kernel Density Estimate, Σ−∆, and mixture of Gaussians. One of the features that was found, was that the more complex methods, such as Kernel Density Estimate and mixture of Gaussians achieve high precision and recall when applied to practice. A lower precision and recall were the results of the more basic methods. However the more basic methods are cheaper in computational cost