Visibility enhancement for robust tracking under bad weather

In this paper, we present a novel method for visibility enhancement based on atmospheric scattering imaging models. Given only a single degraded image, we firstly estimate global atmospheric light vector based on dark channel prior. Then fast bilateral filter is used to deduce atmospheric veil, which is the key contribution of this paper. Following these, the ideal scene radiance could be recovered by directly solving physics-based imaging equation finally. The main advantage of our weather removal algorithm is that, it does not require any a priori scene structure, distributions of scene reflectance, or detailed knowledge about the particular weather condition, and could achieve similar or better restoration results with only a fraction of time consumption in contrast to state-of-art techniques both for color and grey images. Experiments results demonstrate that out algorithm could significantly enhance the details of hazy images, which is very important for features extraction and robust tracking for out-door vision system.