以道路知識進行車載光達點雲之道路點雲與道路標記萃取

The Extraction of Road Regions and Road Marks from Mobile Lidar Point Clouds based on Knowledge of Road

連以諾(Yi-No Lien)

指導教授 ： 張智安

國立交通大學/工學院/土木工程系所/碩士(2013年)

https://doi.org/10.6842/NCTU.2013.00534

摘要

光達(Light Detection and Ranging, Lidar)是主動式系統，光達藉由短時間發射大量脈衝，獲得大量且密集的三維點座標，一般稱之為點雲(Point clouds)。隨著撞擊物體材質的不同，接收脈衝的回訊強弱也有所不一，因此點雲有明暗差異。透過大量三維空間的點座標以及點雲明暗程度，點雲可以描繪物體的幾何外觀，並以輻射差異表現不同材質的物體。車載光達將光達裝置於汽車，對於道路及街景進行掃描，可獲得較完整且密度較高的道路點雲。因此車載光達點雲可細緻且完整的描繪道路的外觀，提供可靠的觀測資料以獲取道路資訊。
本研究目的為發展半自動化道路資訊萃取，以道路知識從光達點雲中獲取道路區域與道路標記，並以資訊回饋方式提升道路資訊萃取成果。研究中首先以人工給定大略的路寬，對原始點雲進行資料切割，濾除道路範圍外的點雲以降低資料量。接著，將切割後的點雲網格化產生回波影像，以形狀幾何萃取白虛線與斑馬線，再藉由人工給予車道線的數量及初始白虛線的概略位置，自動偵測及連結白虛線以獲得道路車道線。萃取道路區域方面，以高程門檻概略濾除非路面點，再使用車道線推估路肩位置獲得道路邊界，並以三次多項式擬合邊界內的點雲，進一步剔除非路面點，得到道路區域內的路面點雲。道路標記使用萃取的路面點產生回波影像，以Sum of Absolute Difference (SAD)計算回波影像中道路標記與資料庫模板的相似度，辨識道路標記的種類。
研究中，使用Riegl VMX-250車載光達獲取的點雲資料進行實驗。實驗區域包含平面道路與高架道路，其地點分別是內湖區的舊宗路與台北市民權大橋。成果分析部分，道路區域針對平面與高程進行驗證。平面以地測資料驗證道路邊界與路寬，高程則以人工編修路面點評估自動偵測路面點雲之精度。道路標記以辨識種類的成果進行評估分析，並以e-GPS在道路標記上測得的角點驗證道路標記平面精度。
研究成果顯示，本研究方法可成功萃取平面道路與高架道路的道路資訊。道路區域平面精度的成果可達10公分，高程精度部分，成果與人工編修DEM相比較，其實驗成果顯示人工分類之路面點與自動萃取之路面點有很高的一致性。道路標記萃取的成功率可達70%，而誤判率低於5%或誤判個數小於5個。道路標記成果的平面精度可達10公分。

關鍵字

移動式光達系統 ； 道路知識 ； 道路資訊 ； 特徵萃取

並列摘要

Lidar (Light Detection and Ranging) is an active system to acquire three-dimensional spatial information. Lidar emits a large number of laser pulses to measure three-dimensional coordinates of the objects in a short time and these three-dimensional coordinates called point clouds. Lidar system scans different objects and the return signals have different intensities. Hence, lidar point clouds are useful information to extract the geometry and attribute of the objects.
The purpose of this research is to develop a semi-automatic method to extract road information from mobile lidar point clouds. This study includes four major parts. First, we divide the point clouds into several road parts. Then, the road parts are converted into intensity images to extract white dashed and zebra crossing, and we connect the white dashed to obtain lanes. Next, we use the road lanes to find the road boundaries. This study utilizes the cubic curve fitting and point-to-curve distance to extract road points. Finally, road marks are recognized by calculating the similarity index from intensity image and road primitives.
The test data are acquired by Riegl VMX-250 and the test area are located in Chiu-Tsung Road and Min-cyuan Bridge in Taipei city. In verification, we compare the extracted results, independent check points, and manual edited DEM.
According to the experimental results, the proposed method may extract road information in overhead road and plane road successfully. The accuracy of the road boundary is better than 10 cm, and result of the road surface is close to manual edited digital elevation model (DEM). The successful rate of the road marks extraction is better than 70%, and commission error is lower than 5%. Moreover, the accuracy of road marks is better than 10 cm.

並列關鍵字

mobile lidar system ； knowledge of the road ； road information ； feature extraction

參考文獻

Canny, J., 1986. A computational approach to edge detection. IEEE Transactions on Pattern Analysis and Machine Intelligence 8, 679-698.

Chen, L.C., Lo, C.Y., 2009. 3D road modeling via the integration of large-scale topomaps and airborne LIDAR data. Journal of the Chinese Institute of Engineers 32, 811-823.

Hinz, S., Baumgartner, A., 2003. Automatic extraction of urban road networks from multi-view aerial imagery. ISPRS Journal of Photogrammetry and Remote Sensing 58 (1-2), 83-98.

IBM, 2010. Welcome to the Decade of Smart.

Ibrahim, S., Lichti, D., 2012. Curb based street floor extraction from mobile terrestrial lidar point cloud. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences 39 (Part B5), 193-198.