Improved Road Detection Algorithm Based on Illuminant Invariant

Aiming at the problems that most road detection methods are sensitive to variation of illumination and shadow, which lead to false detection or leak detection, improved road detection algorithm based on illumination invariant is proposed. First, the thesis transformed RGB space of road images into log-chromaticity space by geometric mean. And then, according to Shannon entropy, camera angle θ of axis calibration is determined. Using Chebyshev's theory, it removed singular value of θ and got illumination invariant images I_θ. Besides, some sampling points of road are extracted by a random sampling, which include standard sample points and referenced sample points. Finally, a confidence interval classifier of road is established, which could detect road area. The experimental results show that the proposed algorithm not only can effectively eliminate the influence of illuminant variance and shadows on road detection, but also can guarantee high detection precision and real-time requirements.