利用 3D LiDAR SLAM技術提升車載移動式製圖系統定位精度

Using 3D LiDAR SLAM to Improve the Positioning Accuracy of Mobile Mapping System

"目前移動式製圖系統已被廣泛運用在空間資訊工程領域，藉由高精度的定位定向系統配合各式各樣的感測器，例如光達、相機等，來獲取空間中精確的三維資訊。光達式車載移動測繪系統整合了精密定位定向系統及光達系統等空間資訊蒐集技術，除了製作各種精度地圖外，其他應用領域亦相當廣泛，是獲取空間資訊出色的利器之一。然而測繪車於全球導航衛星系統（Global Navigation Satellite System, GNSS）訊號嚴重遮蔽地區，特別是高樓都市地區或隧道處，使用慣性導航系統（Inertial Navigation System, INS）和全球導航衛星系統之整合方式，其位置與姿態估計誤差會隨時間漂移和累積，使其定位定向精度大幅降低。為了提升光達測繪車於GNSS訊號遮蔽區域之定位精度，本研究以同步定位與製圖技術（Simultaneous Localization And Mapping, SLAM）理論為基礎，結合車載光達觀測量、輪速計等輔助資訊，研發提高光達測繪車定位精度之技術，進而提升光達測繪車產製成果之品質和增加成果可靠度。本研究提出基於分布之常態分布轉換法（Normal Distribution Transform, NDT），進行不同時刻點雲間之匹配與轉換，並以光達里程計之概念提供速度及航向角觀測量反饋擴增式卡曼濾波器（Extended Kalman Filter, EKF）用於與原有之INS/GNSS觀測量整合。 為了分析本研究所提方法之效能，於兩個不同環境（弱遮蔽及半遮蔽）執行兩次實驗。其結果顯示，NDT演算法能提供精確且穩定的速度及航向角觀測量。而從整體定位成果分析中，於半遮蔽環境（GNSS訊號遮蔽環境）中加入NDT及輪速計資訊能顯著地提升定位精度並達到高精地圖之定位需求（平面：20 cm；三維：30 cm）。最後，由本研究的成果分析可歸納出，透過不同感測器的組合是增強測繪車測繪能力的有效方式。"

Recently, Mobile Mapping System (MMT) has been widely applied in the field of geomatics engineering. By combining high-precision Positioning and Orientation Systems (POSs) with multi-mapping sensors such as Light Detection and Ranging (LiDAR), cameras, and others, 3D geospatial information can be then obtained precisely. The LiDAR-mounted Mobile Mapping Systems (MMS), one of the best tools for collecting geospatial information, integrates several technologies, such as POSs and LiDAR systems, which can be used for producing maps with specific accuracy and it is also efficiently exploited in different domain applications. However, the positioning accuracy will significantly decrease in the areas, like urban areas or tunnels, where Global Navigation Satellite System (GNSS) signals are blocked or contaminated by multipath effects. In order to improve the positioning accuracy and ensure the reliability of the LiDAR-mounted MMS in the GNSS-denied areas, this research aims to develop a technique based on Simultaneous Localization And Mapping (SLAM), combining with LiDAR measurements and the additional information from odometer to enhance the performance. In this study, Normal Distribution Transform (NDT) will be tested. Further, NDT-based LiDAR Odometry (LO) techniques have been introduced and developed by aiming to perform and feed the LO-derived velocity and heading measurement update into Extended Kalman Filter (EKF) model of INS/GNSS. To analyze the performance of the proposed methodology, two experiments with different environments (i.e. weak and semi-occulusion) are conducted. The obtained results show that NDT algorithm can provide stable and accurate velocity and heading measurements. From the overall positioning results, the statistic shows that adding NDT-derived observations and odometer information can significantly improve the positioning accuracy and achieve the requirement of high definition map (2D: 20 cm; 3D: 30 cm) in semi-occlusion environment. Finally, it can be concluded that the combination of different sensors is an effective way for enhancing the capability of the MMS.