Abstract

In this paper, we propose an obstacle detection approach that uses a facet-based obstacle representation. The approach has three main steps: ground point detection, clustering of obstacle points, and facet extraction. Measurements from a 64-layer LiDAR are used as input. First, ground points are detected and eliminated in order to select obstacle points and create object instances. To determine the objects, obstacle points are grouped using a channel-based clustering approach. For each object instance, its contour is extracted and, using an RANSAC-based approach, the obstacle facets are selected. For each processing stage, optimizations are proposed in order to obtain a better runtime. For the evaluation, we compare our proposed approach with an existing approach, using the KITTI benchmark dataset. The proposed approach has similar or better results for some obstacle categories but a lower computational complexity.

Highlights

  • Autonomous vehicles use sensors for environment perception in order to detect traffic participants and other entities

  • In a 3-D point cloud obtained with a LiDAR sensor for autonomous vehicles, objects rise perpendicularly to the road surface, so the points are classified as road or non-road points

  • The LiDAR data is organized as images representing the top-view of the scene and the information encapsulated is limited to 40 m in front and 10 m on each lateral side of the car

Read more

Summary

Introduction

Autonomous vehicles use sensors for environment perception in order to detect traffic participants (pedestrians, cyclists, vehicles) and other entities (road, curbs, poles, buildings). The facet/polygonal representation provides a better localization for the boundaries of non-cuboidal shaped obstacles. This allows a more accurate environment representation, improving potential driving assistance functions. Neural network-based methods for road detection were presented in [5,6,7,8,9,10,11] Most of these methods typically represent the 3-D point cloud as images, which are provided as input to the neural networks that detect the drivable area. Each point is labeled as ground point or obstacle Another way to detect the road points was presented in [4], where Asvadi proposed an algorithm that determines the surface of the road in four stages: slicing, gating, RANSAC plane fitting, and validation. The LiDAR data is organized as images representing the top-view of the scene and the information encapsulated is limited to 40 m in front and 10 m on each lateral side of the car

Object Detection
Facet Detection
Proposed Approach for Obstacle Facet Detection
Preprocessing
Evaluation and Results
System Parameters
Ground Point Detection
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call