Cast Shadow Resistant Ground Plane Detection in Single Image

Abstract

Ground plane detection is useful for vision navigation to robots and autonomous vehicles. Cast shadow on the ground plane is a challenging issue that may cause the detection fail. In this paper, we present a cast shadow resistant ground plane detection approach from a single color image. We first derive an initial ground plane using geometric layout method. We then apply shadow invariant transform on the roughly detected shadow edges to get a gray-scale intrinsic image. The final shadow resistant ground plane result is obtained by employing region growth on the initial seed region in the shadow-free intrinsic image. The approach proposed here does not need priori assumption such as calibration or landmark. Experimental results show the method works for different scenes.