An Approach for Unattended Object Detection through Contour Formation using Background Subtraction
Neelam Dwivedi, Dushyant Kumar Singh, Dharmender Singh Kushwaha · Procedia Computer Science · 2020
Any unattended object in public places is generally considered as a suspicious object. Identification of such objects in public places is one of challenging tasks in computer vision. Robust object detection system with prompt and accurate identification of safety breaches augments security measures at public places such as shopping malls, airports, railway stations, etc. This paper proposes a new and efficient approach for the detection of unattended objects in a well illuminated environment from video captured through a single camera. The proposed approach extracts foreground objects using background subtraction method. These foreground objects are further categorized into static and moving objects by comparing their respective coordinates of centroid in consecutive frames. The foreground object is classified as an unattended object if that object is static over a predefined period of time and its size lies in the predefined range. The performance of the proposed approach is evaluated by conducting the experiments on ABandoned Objects DAtaset (ABODA) dataset and also on few real time videos recorded by the authors. The results are derived for evaluating the experiments using Correct Object Detection Rate (CODR), Object Success Rate (OSR), and False Alarm Rate (FAR) metrics. The CODR & OSR rates so attained for the proposed approach with static background are 100% each while FAR is 0%. Average CODR, average OSR, and average FAR of static and dynamic background are found to be 70.83%, 67.25%, and 35.41% respectively for the proposed approach. The proposed approach updates/replaces the original background frame to minimize the scope of false alarm rate in case of dynamic background. It is generic and can be efficiently used for object detection in various public scenarios.