Automated Shadow Removal for Outdoor Surveillance Monitoring

Automatic video surveillance is an important security system for many sectors to control access, monitor events, and protect assets. The automation process should analyze scenes, such as detecting a person, vehicle, or activity. In order to allow effective monitoring, the surveillance images should be freed from occlusion or unwanted noises. Shadow is one of the major problems in this issue as it frequently appears in images of outdoor environments. The presence of shadows hides important details of images, distorts the shape of target objects, and causes loss of color information. It frequently misleads the results of the related applications which are unable to detect objects, misclassify subjects’ identities, and produce false alarms. Thus, automated shadow removal from an image is an intriguing research area. For low-altitude images captured using CCTV located at a high position (e.g. high floor building, lamp posts, etc.), the shadow removal process is more difficult compared to the ground-based images. This is due to the aerial images containing higher spatial resolution but fewer details of image features. Furthermore, the shadows in the images always appear in various types, shapes, sizes, and brightness depending on time of day, which contributes to more challenges. The state-of-the-art shadow removal methods developed for either ground-acquired images or very high-altitude images, such as satellite images, may not be able to perform well. Shadow removal based on deep learning has achieved good performance for ground-acquired images. However, it remains a challenging task in dealing with shadows of aerial images that contain uneven surfaces and a wide variety of objects.

Such challenges give rise to the requirement of automatic detection and removal of undesired shadows in low-altitude images specifically for outdoor surveillance monitoring. Henceforth the hypothesis of the main problem to be investigated in this project is that an automated shadow removal framework ought to lessen the misclassification of object identification, thus, improving the accuracy of monitoring in surveillance videos. By this hypothesis, the research questions for this study will include the following: (i) are the state-of-art shadow removal methods suitable for shadow removal from low-altitude surveillance images? and (ii) does the proposed shadow removal framework improve the automated object detection. The objectives of the collaboration could include (i) to identify mechanisms for implementing shadow removal for automated object detection, and (ii) to evaluate the effectiveness of the proposed framework. The mechanism includes but is not limited to the use of efficient memory storage, and the implementation of an artificial intelligence framework.