Semantic Point Cloud Segmentation Using Fast Deep Neural Network and DCRF.

Yunbo Rao,Jiansu Pu,Menghan Zhang,Junmin Xue,Zairong Wang,Zhanglin Cheng

doi:10.3390/s21082731

Abstract

Accurate segmentation of entity categories is the critical step for 3D scene understanding. This paper presents a fast deep neural network model with Dense Conditional Random Field (DCRF) as a post-processing method, which can perform accurate semantic segmentation for 3D point cloud scene. On this basis, a compact but flexible framework is introduced for performing segmentation to the semantics of point clouds concurrently, contribute to more precise segmentation. Moreover, based on semantics labels, a novel DCRF model is elaborated to refine the result of segmentation. Besides, without any sacrifice to accuracy, we apply optimization to the original data of the point cloud, allowing the network to handle fewer data. In the experiment, our proposed method is conducted comprehensively through four evaluation indicators, proving the superiority of our method.

Highlights

In recent autonomous driving and augmented reality applications, sensors that can directly capture 3D data are becoming more common
Scene segmentation and motivates us to use the 3D point cloud to deal with the task of semantic segmentation
We have established a novel 3D point cloud segmentation network consisting of four multi-layer perceptrons and two small regularized networks

Summary

Introduction

In recent autonomous driving and augmented reality applications, sensors that can directly capture 3D data are becoming more common. Extensive learning using 3D data has been extensively studied, and significant progress has been made in typical applications such as scene understanding, indoor segmentation, urban features, natural environments, shape complementation, and shape matching. The sparseness of point clouds in Thanks to the introduction of the PointNet [1] network, we can design deep networks using point cloud data directly and end-to-end, and handle the disorder and unstructured between point clouds. Qi et al [2,3] proposed an efficient and robust deep architecture to handle point clouds directly, which opens up new opportunities for 3D scene segmentation and motivates us to use the 3D point cloud to deal with the task of semantic segmentation.

Methods

Results

Conclusion