Point cloud-based 3D local feature descriptors are applied usually in 3D object recognition and categorization. In the context of navigation on asteroid surfaces, such descriptors are useful for place recognition, point cloud registration and semantic segmentation. We propose in this paper a projected contour histogram descriptor that encodes geometric contours of rock point clouds viewed from different angles. An extraction method is proposed as well that extracts the rock point clouds from point clouds of asteroid terrain surface. This method is efficient and convenient utilizing inherent flash LiDAR sensor specificities. To complement the descriptor, a matching method is introduced. Validation of our methods in a high-fidelity simulator of the asteroid surface environment shows that our extraction method is capable of filtering out robustly rock points from surface point clouds. The proposed descriptor achieves a unique description of every rock point cloud and can thus provide distinguishable natural landmarks.