Sunlight Resistant Face Recognition Surveillance

I need a robust real-time face recognition solution that keeps working even when cameras are pointed straight into bright daylight. The system will be deployed both indoors and outdoors, so it must automatically adapt to sudden lighting changes as people move between shaded areas, hallways, and direct sun.

My main expectation is a working prototype that can:
• Detect and recognize faces with high accuracy under intense glare and shifting shadows.
• Run live on-site (edge device, GPU box, or similar) with latency low enough for surveillance use.

If you already have experience mitigating over-exposure, bloom, or lens flare with clever preprocessing, specialized sensors, HDR pipelines, or training tricks such as illumination-invariant embeddings, please let me know. OpenCV, TensorFlow, PyTorch, or any proven commercial engine is acceptable as long as the final stack can be handed over with clear build/run instructions.

I’ll provide access to camera streams for testing and a small labelled dataset; you are free to augment it as needed. Success will be measured by recognition precision/recall in direct sunlight during a live demo and the stability of the frame-rate under those conditions.