The “Marauder’s Map” is a magical artifact from the Harry Potter franchise. That kind of magic isn’t actual, however as Arthur C. Clarke famously identified, it doesn’t have to be — we’ve got expertise, and we will make our personal magic now. Or, moderately, [Dave] on the YouTube Channel Dave’s Armoury could make it.
[Dave]’s ironmongery store is perhaps in a tough neighborhood, because it has 50 cameras’ price of CCTV protection. On this case, the stockman’s loss is the hacker’s achieve, as [Dave] has talked his method into accessing all of these numerous digicam feeds and is utilizing machine imaginative and prescient to trace each single human within the retailer.
After all, finding people in a video feed is simple — to find them in house from that feed, one first wants an correct map. To do this, [Dave] first 3D scans all the retailer with a rover. The scan is in full 3D, and it’s no small quantity of information. On the rover, a Jetson AGX is required to deal with it; on the bench, a beefy HP Z8 Fury workstation crunches the purpose cloud right into a map. Fortunately it got here with 500 GB of RAM, since simply opening the mesh file generated from that time cloud wants 126 GB. That’s processed right into a easy 2D flooring plan. Whereas the workflow is spectacular, we will’t assist however surprise if there was a neater method. (Perhaps a tape measure?)
As soon as an correct map has been generated, it seems NVIDIA already has a turnkey resolution for mapping video feeds to a 2D spatial map. When processing a lot information — keep in mind, there are 50 digicam feeds within the retailer — it’s not very best to be passing the picture information from RAM to GPU and again once more, however fortunately NVIDIA’s “Deep Stream” pipeline will do object detection and monitoring (together with between totally different video streams) all on the GPU. There’s additionally pose estimation proper in there for extra correct monitoring of the place an individual is standing than simply “inside this crimson field”. With 50 cameras, it’s all a bit a lot for one card, however fortunately [Dave]’s workstation has two GPUs.
As soon as the coordinates are spat out of the neural networks, it’s comparatively easy to place footprints on the map in true Harry Potter vogue. It truly is magic, within the Clarkian sense, what you are able to do when you throw sufficient computing energy at it.
Sadly for show-accuracy (or happily, when you choose to keep away from gross privateness violations), it doesn’t monitor each particular person by identify, but it surely does exhibit the chance with [Dave] and his robotic. If you would like a map of one thing… else… possibly take a look at this yard venture.
