The paper presents an idea of real-time video stream analysis which leads to the detection and tracking of suspicious objects that have been left unattended, which is one of the most crucial aspects to be taken into consideration during the development of visual surveillance system. The mathematical principles related to background model creation and object classification are included. We incorporated several improvements to the background subtraction method for shadow removal, lighting change adaptation and integration of fragmented foreground regions. The type of the static regions is determined by using a method that exploits context information about foreground masks, significantly outperforming previous edge-based techniques. Developed algorithm has been implemented as a working model involving freely available OpenCV library and tested on benchmark data taken from real visual surveillance systems.
展开▼