一種里程估測演算法基於非一致性多率取樣含限制條件之卡曼濾波器使用LiDAR及IMU

Abstract

本論文提出一個里程估測的演算法，使用EM-GICP當作LiDAR點雲縫合的方法，結合拓撲的邊緣、平面特徵點截取方法以及使用重力投影法來估測線性加速度。最後使用非一致性取樣多取樣率之卡爾曼濾波器來串連LiDAR以及IMU。 本論文旨在解決點雲縫合時會因環境特徵不足等原因造成最佳化的局部最小值，和IMU當作里程計時會有的偏移現象。使用的主要方法是結合兩個sensor的特性，IMU輔助動態的狀態估測，LiDAR強調穩態的收斂。更使用最大期望算法增強點雲縫合的最佳化強健性

The proposed algorithm uses EM-GICP as LiDAR registration algorithm combined with topology approach plane and edge feature extraction, gravity projection method to estimate the linear acceleration and complete the odometry algorithm with nonlinear multi-rate Kalman filter does fusion of data from noisy sensors, LiDAR and IMU. The proposed method solves the problem of local minimum in common LiDAR base algorithm, and drift problem while using IMU. Moreover, this thesis combines the General-ICP with EM algorithm to get more robust optimization outcome in each iteration, use topology approach feature extraction method to fit the property of sensor for getting more physical meaning feature points.