On the segmentation of 3D LIDAR point clouds

Douillard, Bertrand; Underwood, James; Kuntz, Noah; Vlaskine, Vsevolod; Quadros, A.; Morton, P.; Frenkel, Alex

doi:10.1109/icra.2011.5979818

Cited by 415 publications

(269 citation statements)

References 12 publications

Supporting

Mentioning

266

Contrasting

Unclassified

Order By: Relevance

“…Classic approaches for object detection in lidar point clouds use clustering algorithms to segment the data, assigning the resulting groups to different classes [2,27,6,18]. Other strategies, such as the one used as baseline method in this paper, benefit from prior knowledge of the environment structure to ease the object segmentation and clustering [20,24].…”

Section: Related Workmentioning

confidence: 99%

Low Resolution Lidar-Based Multi-Object Tracking for Driving Applications

Pino

Vaquero

Masini

et al. 2017

ROBOT 2017: Third Iberian Robotics Conference

View full text Add to dashboard Cite

Abstract. Vehicle detection and tracking in real scenarios are key components to develop assisted and autonomous driving systems. Lidar sensors are specially suitable for this task, as they bring robustness to harsh weather conditions while providing accurate spatial information. However, the resolution provided by point cloud data is very scarce in comparison to camera images. In this work we explore the possibilities of Deep Learning (DL) methodologies applied to low resolution 3D lidar sensors such as the Velodyne VLP-16 (PUCK), in the context of vehicle detection and tracking. For this purpose we developed a lidar-based system that uses a Convolutional Neural Network (CNN), to perform pointwise vehicle detection using PUCK data, and Multi-Hypothesis Extended Kalman Filters (MH-EKF), to estimate the actual position and velocities of the detected vehicles. Comparative studies between the proposed lower resolution (VLP-16) tracking system and a high-end system, using Velodyne HDL-64, were carried out on the Kitti Tracking Benchmark dataset. Moreover, to analyze the influence of the CNN-based vehicle detection approach, comparisons were also performed with respect to the geometric-only detector. The results demonstrate that the proposed low resolution Deep Learning architecture is able to successfully accomplish the vehicle detection task, outperforming the geometric baseline approach. Moreover, it has been observed that our system achieves a similar tracking performance to the high-end HDL-64 sensor at close range. On the other hand, at long range, detection is limited to half the distance of the higher-end sensor.

show abstract

Section: Related Workmentioning

confidence: 99%

Low Resolution Lidar-Based Multi-Object Tracking for Driving Applications

Pino

Vaquero

Masini

et al. 2017

ROBOT 2017: Third Iberian Robotics Conference

View full text Add to dashboard Cite

show abstract

“…Some works have tested the benefit of using the ground as a separator between objects [10]. Generally, ground segmentation methods assume the whole ground is continuous, almost flat, or the biggest object in the scene.…”

Section: B Ground Segmentationmentioning

confidence: 99%

“…Once the ground has been extracted, the object partition is found by clustering together adjacent voxels [10], see Figure 5. Each object is also split into regions.…”

Section: B Ground Segmentationmentioning

confidence: 99%

“…Usually the scene is discretized in 3D cubes with constant size as a 3D grid, but these voxels can also be dynamically positioned around groups of points [9]. In [10] several algorithms have been developed for both dense and sparse pointclouds, which depends on the scan device. The work concludes that algorithms which extract the ground offer more accurate results in the subsequent segmentation process .…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

3D segmentation method for natural environments based on a Geometric-Featured Voxel map

Plaza

Ababsa

García-Cerezo

et al. 2015

2015 IEEE International Conference on Industrial Technology (ICIT)

View full text Add to dashboard Cite

-This work proposes a new segmentation algorithm for three-dimensional dense point clouds and has been specially designed for natural environments where the ground is unstructured and may include big slopes, non-flat areas and isolated areas. This technique is based on a Geometric-Featured Voxel map (GFV) where the scene is discretized in constant size cubes or voxels which are classified in flat surface, linear or tubular structures and scattered or undefined shapes, usually corresponding to vegetation. Since this is not a point-based technique the computational cost is significantly reduced, hence it may be compatible with Real-Time applications. The ground is extracted in order to obtain more accurate results in the posterior segmentation process. The scene is split into objects and a second segmentation in regions inside each object is performed based on the voxel's geometric class. The work here evaluates the proposed algorithm in various versions and several voxel sizes and compares the results with other methods from the literature. For the segmentation evaluation the algorithms are tested on several differently challenging hand-labeled data sets using two metrics, one of which is novel. 3D Segmentation Method for Natural Environments based on a Geometric-Featured Voxel MapVictoria Plaza * , Fakhr-Eddine Ababsa † , Alfonso J. Garcia-Cerezo * and Jose Antonio Gomez-Ruiz * * University of Malaga, 29071 Malaga, Spain. Email: victoriaplaza, ajgarcia, janto@uma.es † Evry University, IBISC EA 4526 France Email: fakhr-eddine.ababsa@univ-evry.frAbstract-This work proposes a new segmentation algorithm for three-dimensional dense point clouds and has been specially designed for natural environments where the ground is unstructured and may include big slopes, non-flat areas and isolated areas. This technique is based on a Geometric-Featured Voxel map (GFV) where the scene is discretized in constant size cubes or voxels which are classified in flat surface, linear or tubular structures and scattered or undefined shapes, usually corresponding to vegetation. Since this is not a point-based technique the computational cost is significantly reduced, hence it may be compatible with Real-Time applications. The ground is extracted in order to obtain more accurate results in the posterior segmentation process. The scene is split into objects and a second segmentation in regions inside each object is performed based on the voxel's geometric class. The work here evaluates the proposed algorithm in various versions and several voxel sizes and compares the results with other methods from the literature. For the segmentation evaluation the algorithms are tested on several differently challenging hand-labeled data sets using two metrics, one of which is novel.

show abstract

“…In [2] a solution has been proposed for efficiently handling data that is continuously streamed from a sensor on a mobile robot, and for separating different semantic regions in the point cloud. [3] presents a set of clustering methods for various types of 3D point clouds, including dense 3D data (e.g. Riegl scans) and sparse point sets (e.g.…”

Section: Introductionmentioning

confidence: 99%

Object extraction in urban environments from large-scale dynamic point cloud datasets

Börcs

Jozsa

Benedek

2013

2013 11th International Workshop on Content-Based Multimedia Indexing (CBMI)

View full text Add to dashboard Cite

Abstract-In this paper, we introduce a system framework which can automatically interpret large point cloud datasets collected from dense urban areas by moving aerial or terrestrial Lidar platforms. We propose novel algorithms for region segmentation, motion analysis, object identification and population level scene analysis which steps can highly contribute to organize the data into a semantically indexed structure, enabling quick responses for content based user queries about the environment. The system is tested on real Lidar data, and for demonstration quantitative evaluation is given on vehicle detection.

show abstract

On the segmentation of 3D LIDAR point clouds

Cited by 415 publications

References 12 publications

Low Resolution Lidar-Based Multi-Object Tracking for Driving Applications

Low Resolution Lidar-Based Multi-Object Tracking for Driving Applications

3D segmentation method for natural environments based on a Geometric-Featured Voxel map

Object extraction in urban environments from large-scale dynamic point cloud datasets

Contact Info

Product

Resources

About