高效移動製圖系統於高精度電子地圖之實現

Abstract

現今製作高精度地圖主要乃利用配備光達之移動製圖系統，透過所收集而成的三維點雲資訊做為繪製的基礎。由於點雲資訊僅包含環境的三維座標，因此現今製作地圖的流程中尚需大量的人工，在點雲的基礎上繪製並給予物件的屬性。因此如何提升製圖效益、降低勞力需求並能導入到現有的產製地圖流程的技術研究為地圖供應商高度期盼的，因此，本研究整合了一個高效益移動製圖系統，其中硬體部分搭載非測量等級光達、消費等級相機、以及入門等級的定位暨慣性導航系統，結合提出的點雲後製流程，其中包含匯入控制點(GCPs)以及採用同時定位與地圖構建(SLAM)等有助於軌跡修正之技術重建出高精度的點雲成果。最後在後製流程中，本論文導入一個深度學習網路進行標牌的偵測，透過相機與光達間旋轉矩陣以及轉移矩陣的轉換，影像中的標牌中心點之大地座標即可被自動地萃取。實驗的結果顯示，所產製的點雲三維精度絕對均方根誤差(RMSE)可控制在10公分內，自動標牌萃取之絕對位置精度也可達到數十公分，因此本研究成果可顯著地利用可接受的硬體成本建置出高精度的點雲資訊，更進一步驗證自動萃取屬性導入製作高精度流程的可行性。Lidar sensors are commonly equipped for HD map establishment on a mobile mapping system (MMS). However, the point clouds themselves do not contain object attributes. Therefore, human operators must manually obtain objects' positions to assign attributes for further high-definition map (HD Map) conversion, inevitably resulting in time-consuming processes and high labor costs. This dissertation presents a cost-effective MMS. The system comprises a non-survey grade Lidar, a commercial grade camera, and entry-level Global Navigation Satellite System/Inertial Navigation System (GNSS/INS). By incorporating ground control points (GCPs) with a Normal Distribution Transform Simultaneously Localization and Mapping (NDT SLAM) refinement and fluctuation refinement, both absolute position accuracy and relative position accuracy of the reconstructed point cloud can be secured. Meanwhile, a deep neural network for image detection is employed to obtain the bounding box of traffic signs. By applying the translation and rotation transformation between Lidar points and camera pixels, the intersection of the detected object in the image and Lidar scan points can be found. Experimental results show that point clouds can be reconstructed with an average 3D RMSE of less than 10cm. The center geodetic coordinates of traffic signs can be further extracted in sub-meter accuracy to reduce labor work in HD map establishment.