Virtual Reality Rendered Video Precognition with Deep Learning for Crowd Management

We describe an AI driven model based on the Nethra Video Analytic platform, optimised for overhead detection and designed specifically for CCTV and which can detect and categorize people from any angle. The model runs in real time in crowded and noisy environments and can be in- stalled on devices as in edge analytics or applied directly to existing video feeds. By mapping an entire space, we link together individual camera feeds and data points to calculate the total number of people to assist with capacity planning and to pin point bottlenecks in people flows. Solving the problem of aggregating multiple 360-degree video camera feeds into a single combined rendering, we further describe a novel use of interactive Virtual Reality. This model renders St Pancras International Station in VR and can track people movement in real time. Real people are rep- resented by avatars in real-time in the model. Users are able to change their viewpoint to look at any angle. The movement of the avatars exactly mirrors what can be seen in the cameras.