Improvement of Background Subtraction Method for Real Time Moving Object Detection

Accurate moving object extraction is a critical stage in a measurable 3D photogrammetric surveillance system. Several methods have been introduced in spatial and frequency domain by now. Background subtraction is a basic method in spatial domain which has great potentials to be implemented in the live videos due to its simplicity and high rate. However, simple subtraction will not lead to the expected results due to the different kind of noise (e.g. illumination changes noise) and contaminations. In this research, in order to improve the precision of this method in extracting the shape of moving objects, an interactive thresholding has been applied to the subtractions. Moreover, to cover the holes which have been produced due to the same tone of moving objects and background, two methods of colour processing and adaptive filtering is introduced. For this purpose and to reduce the noises and falsedetections, morphological operators have been utilized. The implemented strategies show improvements in the overall accuracy of object detection and separation.