ABANDONED OBJECT DETECTION VIA SUBSPACE LEARNING

Abstract

In this study, a novel video surveillance algorithm is developed for detection of abandoned objects in public scenes. Two different foreground model is used to detect moving and temporarily static objects. A subspace learning method, called GoDec, is used to detect foreground objects. By using GoDec algorithm, shape and contour of foreground objects are obtained more precisely than the traditional methods. Algorithm is tested on different public datasets and on a new dataset prepared in various environments by GTU, Gebze Technical University. According to test results, proposed method gives better results especially for background initialization and occlusion problems.