3D Digital Map Data Fusion Enabled Real-time Precision Positioning For the Self-driving System Using the Unscented Kalman Filter and Interactive Multiple Model Based Vehicle Motion Detection Techniques

碩士 === 國立臺灣大學 === 機械工程學研究所 === 105 === This research proposes an approach that is able to locate vehicle position with lane level precision using low-cost multi-sensor fusion including commercial GNSS, IMU and digital maps. The approach is based on interactive multiple models (IMM), data fusion, and...

Full description

Bibliographic Details
Main Authors: Po-Fu Wu, 吳柏富
Other Authors: Kang-Li
Format: Others
Language:zh-TW
Published: 2016
Online Access:http://ndltd.ncl.edu.tw/handle/94351788298999523598
Description
Summary:碩士 === 國立臺灣大學 === 機械工程學研究所 === 105 === This research proposes an approach that is able to locate vehicle position with lane level precision using low-cost multi-sensor fusion including commercial GNSS, IMU and digital maps. The approach is based on interactive multiple models (IMM), data fusion, and unscented Kalman filter techniques. The unscented Kalman filter (UKF) technique is used to design the estimator of the vehicle position, as well as executing data fusion which integrates multiple sensor data. The sigma points around the position center will be calculated by unscented transform, representing the probability of vehicle position. In this research, the probability of vehicle motion is also estimated by the motion sensor through IMM, including longitudinal motion, lateral motion and slope motion. For the estimation result, digital maps will be used to increase the precision of the vehicle position by providing road information and attributes. By utilizing the constraints such as road boundary on UKF, the sigma points positions can be realigned according to the position reference, increasing the precision of vehicle position. The algorithms proposed in this research uses road and vehicle information obtained from vehicle dynamics simulation software CarSim to validate positioning precision with different vehicle velocity and motion. The results when compared with general cases demonstrated significant enhancement on vehicle positioning, with the proposed algorithm able to gather more road and vehicle motion related data for the driver. Finally, the proposed system has been validated using experimental vehicle driven around the NTU campus and Shue-Yuan expressway, with results showing consistent positioning precision elevation down to lane level.