CCTV Footage Summarization for Increasing Efficiency in Surveillance (using ML)

Surveillance cameras are ubiquitous in this era of the future. Whether a house, office, shopping complex, or highway, CCTV cameras are ubiquitous that monitor the activities on a day-to-day basis. Surveillance is a significant activity. They help in identifying any suspicious activity and serve as effective evidence. Yet, the quantity of video recording is huge. Useful information it holds most of the time is extremely minimal. It then becomes an extremely challenging task to sift through such many hours of video to pick out useful information. Storage space for the videos is also wasted on a vast scale. So, there's a need perceived for surveillance products that can make such long recordings into short ones without losing essential events. Our system can condense hours of video recorded by CCTV cameras into a single clip that displays all interesting events at once. All moving objects in the video are detected and tracked by our software. These events are overlaid on one clip, and the timestamps for each event are also displayed, thereby allowing the user to conduct surveillance for multiple events. Our model employs the KNN model for background subtraction and video extract in addition to a special Object Tracking algorithm to identify moving objects that are overlapped on the extracted background.