Deep Learning driven computer vision ‘YOLOv3’ powered by Darknet, a neural network framework written in C and CUDA.
Several optimizations and improvements were made as per my Honours level research in aerospace engineering and AI computer vision allowing the real-time detection and tracking capabilities seen.
Objective of this video is to demonstrate these optimizations allowing capability of ultra-high resolution input streams. 4K in, 4K out, with a deep learning neural net sitting in between achieving 25FPS+ (better than real-time).
Run on Google Cloud Platform Compute Engine with NVIDIA Tesla P100 supercomputer GPU.
Formation flight credit to Emirates, SkyDive Dubai and Aerovision.