Disparity-Based Robust Unstructured Terrain Segmentation
Autonomous robot navigation in unstructured outdoor environment is still a challenging problem,and the terrain segmentation is one of the key tasks in robot navigation.Previous methods work well on common terrains like urban roads,but tend to fail in wild conditions due to different illumination,weather and road variations.In this paper,we propose a novel two branches terrain segmentation network based on disparity map and ground plane fitting,introducing geometric characteristics into the network.The terrain segmentation main branch uses convolutional feature layers with multiple sampling rates filters,which effectively considers local and global context information and smooths the holey information in the disparity map.The enhancement branch exploits plane geometry property of the ground plane deviation map calculated from the disparity map,which adaptively generates reference feature maps for improving the robustness of identifying traversable areas under conditions of unseen terrains.Experimental results demonstrate excellent performance of the proposed method on terrain segmentation both qualitatively and quantitatively.
Terrain segmentation Disparity-based Plane geometry property Convolutional neural network
国际会议
广州
英文
421-431
2018-11-23（万方平台首次上网日期，不代表论文的发表时间）