Co-Design of Algorithms and Architectures for Machine Learning Inference at the Edge for Video Analytics

Video analytics involve processing video content in real-time, extracting metadata, sending out alerts, and delivering actionable intelligence insights to security staff or other systems. Video analytics products apply artificial intelligence to cameras to recognize temporal and spatial events. Video analytics are needed in various end applications such as quality inspection, industrial process automation, and workplace security. It is crucial to have video analytics performed at the edge on the multiple streams from on-premises cameras to make automated predictions with high accuracy and low latency. This talk explains the co-design of hardware friendly algorithms and corresponding domain specific accelerator architectures for machine learning inference at the edge for video analytics.