Robust Curb Detection Based on the Accessible Route Analysis and Key Frames Prediction

For autonomous driving in urban environments, road curb plays a significant role in tasks such as lane-keeping, assisted localization, and path planning. A real-time robust curb detection algorithm based on 3D LiDAR is proposed in this paper. Firstly, the iterative beam model is applied to get the accessible route of the road to obtain the starting point of the search step for each scan line. Secondly, the candidate curb points are extracted according to the spatial distribution characteristics of the point cloud. To effectively combine the historical boundaries information, a Bayesian filter is used to track the road width to reduce the false detection of curb points when the boundaries are interrupted, or on-road obstacles appear. The proposed algorithm is tested in different road environments. The experimental results show that our method has strong scene adaptability. The detection accuracy is over 90%, and the average runtime is 34.62 ms.