Suspected Activity Detection in Living Environment using Continual Learning

Abstract

In the 21^st century, almost everyone around the world, including individuals, businesses, and governments, are using some video surveillance technique to enhance security. Millions of CCTV cameras are already installed worldwide, continuously monitoring the activities. Despite such a large-scale initiative, we still need a vast man force to monitor these activities and take appropriate actions quickly. Due to lack of efficiency and human error, many threatening situations go unnoticed and ultimately lead to some preventable disasters. With the integration of IoT sensors and A.I. algorithms, we can still see some improvement in surveillance methods, and the systems can notify the designated authorities about the programmed suspicious activities, but this is not enough. We need to thoroughly understand and design such a system that can understand human behaviour and improve itself over time so that the quality of alerts can be improved exponentially. Hence, we propose a technique to implement this system based on Artificial Intelligence complemented with Continual Learning. This will allow us to cover a wide range of suspicious activities. Moreover, with C.L., we will be able to improve our A.I. algorithm over time without much human intervention required. Real-time prediction of suspicious events could be highly beneficial, and at times we can avoid very big mishappenings. With the combination of A.I. and C.L., the results will be far more promising.