Improving Lidar-Based Semantic Segmentation of Top-View Grid Maps by Learning Features in Complementary Representations

TL;DR

This paper introduces a novel deep learning approach that fuses features from complementary 2D representations of LiDAR data to improve semantic segmentation of top-view grid maps in autonomous driving.

Contribution

It presents a new neural network architecture that effectively combines features from orthogonal LiDAR projections, enhancing segmentation accuracy over previous methods.

Findings

Improved segmentation accuracy on SemanticKITTI dataset.

Effective fusion of features from multiple representations.

Insights into the impact of different feature embeddings.

Abstract

In this paper we introduce a novel way to predict semantic information from sparse, single-shot LiDAR measurements in the context of autonomous driving. In particular, we fuse learned features from complementary representations. The approach is aimed specifically at improving the semantic segmentation of top-view grid maps. Towards this goal the 3D LiDAR point cloud is projected onto two orthogonal 2D representations. For each representation a tailored deep learning architecture is developed to effectively extract semantic information which are fused by a superordinate deep neural network. The contribution of this work is threefold: (1) We examine different stages within the segmentation network for fusion. (2) We quantify the impact of embedding different features. (3) We use the findings of this survey to design a tailored deep neural network architecture leveraging respective advantages of different representations. Our method is evaluated using the SemanticKITTI dataset which provides a point-wise semantic annotation of more than 23.000 LiDAR measurements.