Techniques for Achieving High Performance in Deep Learning Based Systems for Selective Filtering of Live Video Streams

The research focuses on applications of parallel and distributed processing techniques and deep learning for filtering frames from live video streaming data based on a set of criteria specified by the user. With various video streaming sources around the world, storing the entire video streams and searching the stored data is computationally expensive. A video stream filtering system keeps the frames of interest and discards the rest for fast search and retrieval as well as reduction of storage space. This research concerns the devising of filtering techniques for streaming video data. Parallel and distributed processing techniques including frameworks like Apache Kafka integrated with TensorFlow are used to devise the proposed scalable real-time filtering prototypes. Initial research results that include experiments with live video streams from webcams were conducted on Amazon EC2 cloud to gain insights into system performance including the reduction in processing time that accrues from applying the proposed techniques.