- 學位論文
自適性張量分析應用於光達點雲特徵萃取
Feature Extraction of LiDAR Point Clouds based on Adaptive Tensor Analysis
摘要
光達(Light Detection and Ranging, LiDAR)系統具有快速獲取高密度三維點雲之特性，然而光達點雲為離散型態，所隱含之物空間場景資訊必須藉助特徵萃取方能具體顯性化，供後續使用。 本研究利用張量分析法搭配區域成長法進行光達點雲的特徵萃取，並考量點雲密度與誤差特性，透過誤差傳播給予合理特徵分類與分群門檻，進行三維點雲中的點、線、面幾何特徵萃取，以及利用區域成長法進行叢聚分析建構符合掃描標的物之幾何特徵描述。除此之外，藉由特徵分類門檻與特徵萃取品質關係之探討，驗證本研究之張量特徵分類門檻設定機制之有效性，從而可免去人為試誤搜尋最佳值域的程序，進而突顯本研究方法之效益。
並列摘要
LiDAR system is a widely adopted assignment, as a result of it has a large number of benefits for quickly obtaining the high-density 3-D point clouds. However, LiDAR point clouds have discrete patterns, thereby the implied object space information must be explicit by feature extraction for subsequent use. This study applies two affected parameters, including point cloud density and random error, to combine tensor analysis and region growing. Then, theoretical thresholds have been set up by error propagation to classify and extract corner, straight line, and planar geometry features. The geometry orientation relationship must be constructed by region growing method. Furthermore, it is found that the self- adaptive thresholds are effective on feature extraction quality.