Better performance in lower light conditions (OpenCV)?

I am trying to detect hands, and my algorithm works perfectly during day, but during the night it completely fails- it shows no signs of working at all.

I have come to the conclusion this is because of lower light conditions.

Could someone please give me tips for better performance in lower light conditions?

My algorithm just uses cvInRangeS to find the skin-coloured pixels in an HS开发者_如何学JAVAV image.

Any tip would do, regardless of how little it would help.

Thanks

Have you normalized your input first? What's the average V value by day and by night? That said, your sensor will be RGB, and the transformation from RGB to HSV loses quite a bit of precision in the H and S components when R,G and B are low. In extrema: {0,1,0} is quite close to {1,0,0} but the hue is entirely different.

cv::equalizeHist is probably what you want for normalization/equalization. As for the colour matching - you are dropping the V component of your image to do the colour matching right? You could also try YCbCr, which has been shown to be even better than HSV in terms of lightness variance.