Simultaneous Segmentation and Tracking in 3D Point Cloud Data

Mehmet Ali Çağrı, Dirk Schulz

2016

Abstract

This chapter presents a simultaneous object segmentation and tracking method in 3D point cloud data which is particularly useful for the traffic scene analysis of autonomous vehicles. It replaces the conventional processing pipeline of object tracking methods that performs the segmentation and tracking modules consecutively with an integrated tracking and segmentation framework. The presented approach gains robustness against the under-segmentation issue, i.e., assigning several objects to one segment, by jointly using spatial features and motion field information to discriminate nearby objects in the data. After a pre-processing step which maps the measurements to a grid representation and removes the data points belonging to the ground, the presented method estimates the motion field of the point cloud data using Kalman filter based tracking of grid cells. A smoothing algorithm is applied to the estimated grid cell velocities for a better motion consistency of neighboring cells. A non-parametric Bayesian clustering algorithm based on a sequential variant of the distance dependent Chinese Restaurant Process exploits the spatial and motion information in an integrated way to sample possible data segmentation hypotheses and decide on the most probable one. Experiments carried out on data obtained with a Velodyne laser scanner in real traffic scenarios illustrate that the presented approach has a satisfactory detection performance and good motion consistency between consecutive time frames.

Paper Citation

in Harvard Style

Çağrı M. and Schulz D. (2015). Simultaneous Segmentation and Tracking in 3D Point Cloud Data.In European Projects in Knowledge Applications and Intelligent Systems - Volume 1: EPS Lisbon 2016, ISBN 978-989-758-356-8, pages 14-34. DOI: 10.5220/0007900800140034

in Bibtex Style

@conference{eps lisbon 201615,
author={Mehmet Ali Çağrı and Dirk Schulz},
title={Simultaneous Segmentation and Tracking in 3D Point Cloud Data},
booktitle={European Projects in Knowledge Applications and Intelligent Systems - Volume 1: EPS Lisbon 2016,},
year={2015},
pages={14-34},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0007900800140034},
isbn={978-989-758-356-8},
}

in EndNote Style

TY - CONF

JO - European Projects in Knowledge Applications and Intelligent Systems - Volume 1: EPS Lisbon 2016,
TI - Simultaneous Segmentation and Tracking in 3D Point Cloud Data
SN - 978-989-758-356-8
AU - Çağrı M.
AU - Schulz D.
PY - 2015
SP - 14
EP - 34
DO - 10.5220/0007900800140034