Digeiz was cofounded in 2015, when real-time computer vision AI was just starting, with one idea in mind: accurately measure pedestrian flows in crowded areas.
We count, we segment by gender and age, we reidentify the same outfit appearance (no facial recognition or any biometric data) to extract GDPR-approved aggregates.
We started with stereoscopic cameras, quickly moved to neural networks. The original code could run 10 cameras at 1 FPS on 8 GPUs with 1 neural network. Now, we run 70 cameras at 10 FPS on 3 GPUs and 3 custom neural networks.
After deploying 25 shopping malls in Europe, we also had to build numerous internal tools for the OPS team, from our own google maps to study camera scenarios to maintenance camera tools - health status, environment anomalies...