Robust Left Object Detection and Verification in Video Surveillance

Left objects pose a real threat to security in public areas such as railway stations and airports. Detection of these objects therefore forms an important part in any intelligent video surveillance system that is deployed at such locations. Successful left object detection algorithms must operate in real time and produce sufficient detection accuracy with low false positive rates. However in reality, the requirement of both speed and performance is not often achieved due to the huge variation in image appearance caused by illumination, scene, and foreground objects (both dynamic and static). This paper tackles the challenge using a background subtraction scheme coupled with three other techniques. Short-term frame averaging is used to reduce the effect of moving objects such as pedestrians and vehicles. Statistical image background modelling is applied to enhance the visual contrast between the object and the background. Pixel colour modelling is employed to verify the results of left object segmentation. All three techniques are computationally lightweight and thus enable the left object detection to operate in real time.