Region MoG and texture descriptor-based motion segmentation under sudden illumination in continuous pan and excess zoom

摘要

Recently violent crimes have increased, and hence, the society is in need of intelligent surveillance system for security applications, like scenarios including ATM, banks, traffic surveillance. An advanced sensor technology like PTZ (Pan Tilt Zoom) camera-based computer vision techniques can provide detailed information of data over static cameras. It has a functionality to rotate, tilt, zoom, and it can pan to cover a wider area. The PTZ camera-based background modelling faces challenges like ego motion, motion parallax, and illumination changes which may cause false detection of foreground. A majority of existing work cannot handle drastic movements like continuous pan and excess zoom due to the changing of pixel at every instant. This task becomes further challenging in sudden illumination conditions. A new combination of algorithms has been proposed to solve the aforementioned issues for moving object detection in the PTZ camera-based video surveillance scenario. It utilizes the Region-based Mixture of Gaussian (RMoG) algorithm, used for foreground extraction to cope with any dynamic fast movement in the related background. Moreover, if there is any false positive in the background due to sudden illumination, it can be eliminated by the Extended Center Symmetric Local Binary Pattern (XCS-LBP) descriptor. Finally, the output is fine tuned using morphological operators for more accurate ROI (Region of Interest) segmentation. Experimental results are evaluated as case studies such as continuous pan, excess zoom, and sudden illumination on various surveillance benchmark datasets including Change Detection (CDnet 2014) dataset to show the robustness of the proposed work.