Even though my attempt at OpenCV was successful, it took a long time to execute and finish. I tried finding “myself” in 13,000 photos. It took several hours to complete and it was only able to find me 11 times in 13,000 security camera images. While it was running, my CPU usage was at 100% the entire time. Mind you, I initially ran the code on my PC (i7-6700K, overclocked), and a GTX 1080Ti GPU, and CPU/GPU was still the bottleneck! So I knew I had to either increase CPU power, create better models, or simply give up with the face recognition idea, but I usually don’t give up so… 🙂Here’s Intel’s version 2 of it Neural Compute Stick. It runs Myriad X as opposed to Myriad 2, and is supposedly 8x faster than the previous generation. It’s a Vision Processing Unit (VPU) that off-loads processing from the host’s CPU and GPU to the stick. Amazing thing! Look at the video below and see how instantaneous it is. Was able to find my face and recognize my gestures on an average 15ms! Yes, FIFTEEN milliseconds! I also ran it on a smaller laptop, since the code has a lot of things that require system-level changes (python libraries). Im running XEN on my laptop to create a virtualized environment for Ubuntu. THis is usually where I code python.
Check the top command output, the CPU usage barely budged! The ultimate goal is to process my security cam feed (20+ cameras) and be able to detect human faces, animals, and objects. With proper coding, I should be able to detect when packages arrive, picked up, or stolen in real-time. Don’t mind the beard, it’s my first attempt 🙂