EmPointMovSeg: Sparse Tensor-Based Moving-Object Segmentation in 3-D LiDAR Point Clouds for Autonomous Driving-Embedded System

Object segmentation is a per-pixel label prediction task that targets at providing context analysis for autonomous driving. Moving-object segmentation (MOS) serves as a subbranch of object segmentation, targeting to separating the surrounding objects into binary options: dynamic and static. MOS is vital for the safety-critical task in autonomous driving because dynamic objects are often a true potential threat to self-driving cars compared to static ones. Current methods typically address the MOS problem as a category feature to label the mapping task, which is not rational in reality. For example, a parking car should be considered as static instead of a moving-object category. There is a little systematic theory to differentiate object moving characteristics from nonmoving characteristics in MOS. Furthermore, restricted by limited resources in the embedded system, MOS is often in an offline manner due to huge computational requirements. An online and low computational cost MOS is an urgent demand for the practical safety-critical mission which takes immediate reaction as compulsory. In this article, we propose EmPointMovSeg, an efficient and practical 3-D LiDAR MOS solution for autonomous driving. Leveraging the power of the well-adapted autoregressive system identification (AR-SI) theory, EmPointMovSeg theoretically explains the moving-object feature in large-scale 3-D LiDAR semantic segmentation. An end-to-end sparse tensor-based CNN which balances segmentation accuracy and online process ability is proposed. We construct our experiment on both representative dataset benchmarks and practical embedded systems. The evaluation result shows the effectiveness and accuracy of our proposed solution, conquering the bottleneck in the online large-scale 3-D LiDAR semantic segmentation.

I. Introduction

Object segmentation serves as an important role in autonomous driving safety. Via predicting the per-pixel label in a driving scene, the object segmentation algorithm is able to provide comprehensive context information to the system, which benefits the perception result so that path planning can be more accurate. For example, by detecting lane, traffic sign, etc., the autonomous driving system is able to get an excellent description of the environment so that smart planning decision is obtained. However, for the safety-critical mission in autonomous driving, the potential threat mainly lies on pedestrian, cars, or trucks which have moving characteristics in common. Less danger is caused by static objects, such as fence, trees, traffic signs, etc. This phenomenon generates a subresearch topic in object segmentation, the so-called moving-object segmentation. Although many solutions exist in multicategory object segmentation, the moving-object segmentation (MOS) domain, which requires the heavy computational expense and the knowledge of distinguishing dynamic objects from static ones [1]–[4], is still a bottleneck in autonomous driving safety object segmentation.