My control application implements human face detection and recognition with using of CV library. My device uses usual VGA USB video camera.
- Good and uniform light illumination allows detect and recognize face successfully.
- Poor and uniform light (semidarkness), also allows detect and recognize face successfully.
- Side lighting makes it difficult for detection
I've tried to create haar cascade xml of half-lit faces and even use haar cascade of human face profile, but in both cases I have same result, detection can not manage this data correctly. The first way provides multiple extra detection of similar shadow relief, because half-lit face is too amorphous when moving and it is very difficult to determine appropriate data for it. Profile of the human face does not gives any results, except of detection of profile, for what it is intended, also with good illumination, but of course can not be detected as a half-lit face.
Recognition process itself not involved in the problem, because with detection I can process any object if it is matches example. In this question, I need to solve the problem of facilitating the camera to get a good quality video image with uniform light for successful detection.
I'm not sure, how it work with other devices or methods, but my phone with Android, Windows phone and iPhone all has a same problem with detection of half-lit face.
I've tried attache some kind of light source to my device to provide good direct light, but two problems arise, first, even if it is soft, not a very bright light, it is still directed to the eyes, what causes discomfort, and also getting of external light into the field of view causes overexposure, what affects detection even more than the lack of light.
Any advice about proper approach to this problem, for example, some suitable device, method, light source, maybe possibility of infrared light with CV etc., would be very helpful