3D object categorization using 2D reflectance image for robot map construction

在这项研究中，我们开发了一种利用激光扫描仪获取的2D反射图像和3D形状信息进行精确目标分类的技术。激光扫描仪在许多应用中被广泛用于捕捉物体的形状。从激光扫描仪上，我们不仅可以获得距离数据，还可以获得激光反射率，它将反射激光的功率作为距离测量的副产品来表示。所开发的技术同时利用这两种信息进行对象分类。为了开发该系统，我们进行了1)车辆激光扫描系统的开发和激光反射率的有效性验证，2)小型数据库的开发，3)小型数据库的分类实验，4)城市规模数据库的开发和分类实验。此外，将从反射率信息中提取的HOG与从距离信息中提取的HOV相结合，开发了3D HOG，并通过室外实验验证了3D HOG的性能。

In this research, we developed an accurate object categorization technique using 2D reflectance image and 3D shape information taken by a laser scanner. A laser scanner is widely used to capture shapes of objects for many applications. From a laser scanner, we can obtain not only range data but also laser reflectance which shows the power of reflected laser as a side product of range measurement. The developed technique utilizes both information for object categorization simultaneously. To develop this system, we conducted 1) development of vehicle laser scanning system and verification of usefulness of laser reflectance, 2) development of small-size database, 3) categorization experiment using small-size database, and 4) development of city-size database and categorization experiment. In addition, we developed 3D HOG by combining HOG extracted from reflectance information and HONV obtained from range information and verified the performance of 3D HOG by outdoor experiments.