Real-Time Detection of Abandoned and Removed Objects in Complex Environments

We present a new framework to robustly and efficiently detect abandoned and removed objects in complex environments for real-time video surveillance. In our system, the background is modeled by a mixture of Gaussians. Similar to Tian et al. [18], this mixture model is employed to detect the static foreground regions (i.e., static blobs potentially corresponding to abandoned or removed objects) without extra computation cost. Several improvements are implemented to the background subtraction method for shadow removal, quick lighting change adaptation, reduction of fragmented foreground regions, and stable background update rate for video streams with inconsistent frame rates. Then, the types of the static regions (either abandoned or removed) are determined by using a method that exploits context information about the static foreground masks, significantly outperforming previous edge-based techniques. Based on the type of the static regions and several userdefined parameters, a matching method is proposed to trigger alerts indicating abandoned and removed objects. Our method can handle occlusions in complex environments with crowds. The robustness and efficiency of the method was tested on our real time video surveillance system for public safety application in big cities and evaluated by several public databases such as i-Lids and PETS2006 datasets.