反射強度を用いた三次元点群のテクスチャ分割に関する研究

摘要

This paper presents a texture-level segmentation method for 3D point cloud based on the changes of laser's reflectance intensity. Existing segmentation methods for the point cloud are usually based on the spatial information, which is impossible to distinguish the texture level difference. In this work we propose a texture-level segmentation for the point cloud with intensity. We first segment the point cloud with a spatial-based method. Then, we perform the texture-level segmentation on each individual segment of segmentation results with the spatial-based method. For each segment, we first estimate the probability density of the intensity distribution with a Gaussian kernel estimator. Moreover, we introduce a method for adaptively determining the bandwidth for Gaussian kernel. Finally, we segment the point cloud based on the number and value of the detected peaks of the estimated probability density function. To evaluate the proposed method, we applied the proposed method to the point cloud data acquired by a Velodyne HDL-32 LiDAR sensor and compare with the non-texture-level segmentation method qualitatively using the corresponding image for reference. The comparison results verify the feasibility of the prosed method for texture-level segmentation of the point cloud.%近年，三次元センサ，特にLIDAR(Light Detection And Ranging)センサが自動運転，ロボットや三次元復元の様々な分野に広く使われている.LIDARセンサは照射したレーザと反射したレーザの時間差により距離（図1b)を計測し，多くのLIDARでは反射強度(図1c)も得られる.さらに，3D LIDARセンサは複数本のレーザが並んでおり，任意の軸で回すことで，LiDARセンサの座標系における三次元点群とその反射強度が得られる.測定した点群から環境認識を行うためには，分割，位置合わせ，意味的な抽出やモデリングの処理が必要となる.特に，点群の分割は意味的な抽出やモデリングの前処理として非常に重要である.