Content based Person retrieval from Video using Forward Backward Frame Check Algorithm

Content based retrieval systems are spanning fast across all search engines and applications. The need for improved retrieval mechanisms has led to development of context aware content based retrieval systems. The recent requirements in automatic surveillance systems have led to development of many feasible solutions by means of Video Analytics. This paper proposes a novel methodology to analyse videos for required person by means of context aware visual search. Visual search involves the process of Key frame extraction, face detection and face identification. Further, the system is enabled with a forward backward frame check algorithm that generates analytical information about the identified person along with temporal information. The proposed method analyses only the key frames in an input video source using the FFmpeg library for extracting key flames. The forward backward frame check technique enables the system to compare the detected face between current frame and previous frame without the need to search the database again if they are found to be same. This reduces the cost of search and helps to summarize the video with analytical information about the persons occurring in the video. The summarized video is labelled with person information and location of occurrence for context aware search. The proposed system is tested with real time video datasets and found to perform better in terms of time and accuracy.