RGB-D Abandoned Object Detection Based on GrabCut Using Kinect

Abstract

Intensity-based techniques for video surveillance (e.g., moving object detection and tracking, object classification, human motion analysis, and activity understanding) are vulnerable to noise and misleading results. Brightness changes and shadow are typically incorporated during the detection phase, resulting in problems. This paper presents a specifically-purposed and novel design for abandoned object detection (AOD) system using the low-cost Microsoft Kinect sensor. Our system encompasses: (1) fully automated abandoned object segmentation by introducing 3D GrabCut in surveillance cases, and (2) a robust AO detector by fusing RGB and depth information gathered by Kinect. We provide both quantitative and qualitative measurements that show our suggested method is effective.