Title Page
Contents
ABSTRACT 8
Ⅰ. Introduction 10
Ⅱ. Previous Studies 14
2.1. Approaches for extrinsic parameter calibration 14
2.2. Approaches for evaluation of extrinsic parameter calibration 15
2.3. Limitations of previous methods 17
Ⅲ. Calibration Performance Index of LiDAR-Motion sensor 18
3.1. Mapping of point cloud map (PCM) 21
3.1.1. Generation of motion sensor's pose using input motion data 21
3.1.2. Point cloud conversion 21
3.1.3. Generation of point cloud map by accumulation 21
3.2. Evaluation of matching error 23
3.2.1. Generation of ground-truth motion sensor's pose using input motion data 23
3.2.2. Generation of predicted motion sensor's pose using point cloud - PCM matching 23
3.2.3. Obtaining of the localization error between ground-truth motion sensor's pose and predicted pose 25
Ⅳ. Experimentally Verification based on Monte-Carlo Simulation 27
4.1. Environment of verification 27
4.1.1. Definition of variables 27
4.1.2. Generation of motion sensor's pose and calibration parameter through Monte-Carlo sampling 28
4.2. Process of verification using Monte-Carlo simulation 29
4.2.1. Generation of landmark using n motion sensor's pose and l calibration parameter 29
4.2.2. Generation of landmark using n motion sensor's pose and m calibration parameter 31
4.3. Result of verification 31
Ⅴ. Experiment 36
5.1. Experiment using simulation data 36
5.1.1. Simulation environment 36
5.1.2. Result and analysis of experiments 38
5.2. Experiment using real data 41
5.2.1. Experimental environment 41
5.2.2. Result and analysis of experiments 42
Ⅵ. Conclusions 44
References 45
Appendix 49
Abstract (in Korean) 50
〈Table 4-1〉 Calibration parameter error configuration for Monte-Carlo simulation 31
〈Table 5-1〉 Calibration parameter between 1 motion sensor and 6 LiDARs 37
〈Table 5-2〉 Various environments for experiments using simulation data. Green lines represent the trajectories of vehicles. Rectangular boxes... 38
〈Table 5-3〉 Calibration parameter error configuration 38
〈Table 5-4〉 Result of experiment using simulation data 38
〈Table 5-5〉 Calibration parameter between 1 Motion sensor (IMU) and 3 LiDARs (LiDAR1: Top-left LiDAR, LiDAR2: Top-mid LiDAR, LiDAR3:... 41
〈Table 5-6〉 Result of experiment using real data 43
〈Figure 3-1〉 Overview of an algorithm to acquire the performance index of the calibration parameter 19
〈Figure 3-2〉 Relationship of three coordinate systems: world, motion, LiDAR coordinates 19
〈Figure 3-3〉 (a) Result of point cloud map (PCM) using the good calibration parameter; (b) result of PCM using the wrong calibration parameter 22
〈Figure 3-4〉 (a) Result of matching between PCM and the point cloud using the good calibration parameter; (b) result of matching using the... 25
〈Figure 4-1〉 Relationship of the four coordinate systems: world, LiDAR, motion, landmark coordinates 28
〈Figure 4-2〉 (a) (green) Landmark for detection and (black) sampled motion sensor's poses. The direction of the arrow is the +z axis of the... 29
〈Figure 4-3〉 (a), (b) (green) Landmark for detection and (cyan) newly generated blurred landmark using equation (8) 30
〈Figure 4-4〉 (a, c, e, g, i, k) are PIdist when adding each 6-DoF error to the calibration parameter as much as in 〈Table 4-1〉. (b, d, f, h, j, l)...[이미지참조] 36
〈Figure 5-1〉 (a) The outdoor parking lot of Konkuk University used for data acquisition and green line represents trajectory of the vehicle;... 42