Video can be used as the input data for the real-time monitoring of machines, products, or processes to which sensors cannot be affixed. Industrial and scientific monitoring applications, compared to other video sources, such as those from surveillance, broadcast, mobile robotics, social media, or entertainment, can often be engineered and structured. Yet, applications of video-based instrumentation in industrial, manufacturing, and scientific experimentation environments are not extensively addressed by the computer vision community.
We discuss the needs, challenges, and recent success in deploying real-time, data-science enabled techniques to efficiently reduce the complexity and dimensionality of raw video data to extract actionable information for real-time feedback and process control, defect detection, and wear and degradation related for factories and the factory subsystem.