Improved semantic segmentation network using normal vector guidance for LiDAR point clouds

Abstract

As Light Detection and Ranging (LiDAR) sensors become incr easingl y pr ev alent in the field of autonomous driving, the need for ac- cur ate semantic se gmentation of three-dimensional points gro ws accor dingl y. To addr ess this c hallenge , w e propose a novel network model that enhances segmentation performance by utilizing normal v ector information. Firstl y, we pr esent a method to impr ov e the accuracy of normal estimation by using the intensity and reflection angles of the light emitted from the LiD AR sensor . Secondly, we introduce a novel local feature aggregation module that integrates normal vector information into the network to impr ov e the performance of local feature extraction. The normal information is closely related to the local structure of the shape of an object, which helps the network to associate unique features with corresponding objects. We propose four different structures for local fea- tur e a ggr egation, ev aluate them, and c hoose the one that shows the best performance . Experiments using the SemanticKITTI dataset demonstrate that the proposed architecture outperforms both the baseline models, RandLA-Net, and other existing methods, achiev- ing mean intersection over union of 57.9%. Furthermore, it shows highly competitive performance compared with RandLA-Net for small and dynamic objects in a real road en vironment. F or example, it yielded 95.2% for cars, 47.4% for bicycles, 41.0% for motorcycles, 57.4% for bicycles, and 53.2% for pedestrians. © The Author(s) 2023.