The below is in response to a query on facial recognition in arenas and stadiums.
Looking into the use of facial recognition and its uses in stadiums; the standard required requires the recognition of faces from pictures in crowds with differing light levels.
The major factors that impact accuracy of facial recognition are:
- Sufficient resolution across the face (optical not digital zoom)
- Sufficient illumination across the face
- Uniform illumination across the face
- Minimal angle on the face
… on both the reference and search images.
The challenge you describe (crowds, distant cameras, lighting, angles) sounds like one of cameras producing decent imagery rather than an FR problem. Complex environments will require expensive camera systems. Simple environments can use much simpler (cheaper) cameras.
You will always get the best result when and where you can control the sensor and the sensor’s environment (lighting, glare, minimal distance from and angle on subject etc).
This is why unidirectional (everyone facing the same) way pinch points (such as entrances, turnstiles, hallways) will always yield the best results. Your challenge is not insurmountable, but it will possibly be prone to False Rejects (missed matches) and Fales Accepts (mismatches) unless thought and investment is made into the camera systems.
Be wary of anybody that tells you that their FR solution can “easily” deliver highly accurate solutions in this environment.
Also of interest, may be: