個別研究テーマ編集

三田誠一シニア研究スカラ
助教 ジョン ヴィジャイ コーネリウス キルバカラン

掲載年度

2019

個別研究テーマ
(日本語)

ポイントクラウドとステレオビジョンを用いる車両の自己位置推定(Vehicle Localization in Point Cloud Maps using Stereo)

個別研究テーマ
(英語)

研究者 ジョン ヴィジャイ コーネリウス キルバカラン
研究概要

In this research, we present a stereo vision based vehicle localization algorithm using 3D point cloud map and particle swarm optimization. The point cloud map used for localization contains dense 3D geometric features generated by a mobile mapping system. Typically, the Global Positioning System
(GPS) is employed to localize a vehicle. However, GPS is prone to intermittent missing signal under certain environmental and satellite conditions. To address this issue, we combine the GPSbased positioning with the stereo vision to localize the vehicle for an improved accuracy and reliability. The stereo vision based localization is formulated as a tracking problem, where point cloud depth images are generated and matched with the stereo depth images using the particle swarm optimization. Each virtual
point cloud depth image is generated from the 3D point cloud data using a series of transformation matrices. These matrices transform the 3D data points from the real-world coordinate system to the stereo coordinate system. The proposed algorithm consists of three phases: 1) offline phase; 2) bootstrapping
phase; and 3) online phase. In the offline phase, particle swarm optimization is used to estimate the fixed transformation between the vehicle and the stereo coordinate system. The bootstrapping
phase initializes the online tracker with the GPS information and uses the Kalman filter to estimate the real-world vehicle transformation parameters. In last online phase, we localize the
vehicle by estimating the transformation between the real-world and vehicle coordinates in each frame. This is achieved by a novel tracking framework based on multi-swarm particle swarm optimization.

close

ページのトップへ戻る