Left Object Detection with Reduced False Positives in Real-Time

Identifying unattended objects in public places efficiently, is one of the major thrust areas of security. This paper proposes a real-time method to identify unattended or left objects in a region of interest that is under surveillance. This is a simple pixel based method for object detection which can be used for both indoor and outdoor environments. This method is robust to changes in illumination in case of high contrast foreground images, which is achieved through normalization. The false positives are eliminated through implementing a method following the ideas of codebook. The proposed method is tested on a live set up. It consists of an IP camera for video capture, an analytics server where the built in intelligence checks for the presence of any left object in the video and a user interface through which the concerned authority is intimated for any timely action. The entire communication in this system follows ONVIF (Open Network Video Interface Forum) standard. Apart from identifying unattended objects, it can be also used to keep designated areas clear of obstructions.