A central challenge to autonomous off-road navigation is discriminating between obstacles that are safe to drive over and those that pose a hazard to navigation and so must be circumnavigated. Foliage, which can often be safely driven over, presents two important perception problems. First, foliage can appear as a large impenetrable obstacle, and so must be discriminated from other objects. Second, real obstacles are much harder to detect when adjacent to or occluded by foliage and many detection methods fail to detect them due to additional clutter and partial occlusions from foliage. This paper addresses both the discrimination of foliage, and the detection of obstacles in and near foliage using Lidar. Our approach uses neighboring pixels in a depth image to construct features at each pixel that provide local surface properites. A generative model for obstacles is used to accumulate probabilistic evidence for obstacles and foliage in the vicinity of a moving platform. Detection of obstacles is then based on evidence within overlapping cells of a map without the need to segment segment obstacles and foliage. High accuracy obstacle and foliage discrimination is obtained and compared with the use of a point scatter measure.