Laser-based modeling

Study on CPS SLAM - 3D laser measurement system for large scale architectures using multiple mobile robots

To construct three-dimensional models of large-scale architectural structures using a laser range finder, multiple range images must be captured from different viewpoints and aligned through post-processing techniques such as the Iterative Closest Point (ICP) algorithm. Typically, to ensure precise alignment, these range images must first be manually registered to positions close to the correct ones prior to applying ICP. Furthermore, sufficient overlap between range images must be ensured by capturing them from densely spaced viewpoints. We propose a novel measurement system for large-scale architectural modeling using a team of mobile robots equipped with onboard laser range finders. Each measurement position is accurately determined using a high-precision Cooperative Positioning System (CPS), which takes advantage of the coordinated operation of multiple robots. This approach enables the construction of 3D models of large-scale structures without the need for post-processing procedures such as the ICP algorithm or dense viewpoint sampling.

Photo
A group of robots equipped with laser range finder, CPS-VIII		 3D models by automatic modeling
Photo
A group of robots equipped with laser range finder, CPS-V		 PhotoPhoto 3D map of large scale architecture
Introduction (19MB) Introduction (14MB)
Photo Dazaifu Tenmangu shrine Photo Dazaifu Tenmangu shrine
Seibuen Park Kyushu University

Papers

Laser measurement system of precise tunnel shape

A three-dimensional shape measurement platform for urban structures, such as tunnels under construction, has been developed. The system consists of a mobile platform equipped with a two-dimensional laser scanner, a three-axis fiber optic gyroscope, corner cubes, integrated software, and a total station. The total station is placed at reference points within the tunnel and is used to measure the positions of the corner cubes mounted on the mobile platform. As the mobile platform moves, the two-dimensional laser scanner continuously captures cross-sectional profiles of the tunnel. These profiles are then reconstructed into a three-dimensional tunnel shape using positional data obtained from the total station and orientation data from the fiber optic gyroscope. The integrated software enables real-time, on-site comparison between the measured tunnel geometry and the design geometry from a 3D CAD model, imported via the standard LandXML format. Error distributions between the actual and design shapes are visualized through a graphical user interface, supporting effective construction process management.

Photo Laser measurement system Multiple robot system
Real-time tunnel shape measurement Measured 3D tunnel shape

Smoothing and impainting technique using laser reflectivity

We developed two denoising techniques for noisy range images using reflectivity information: range image smoothing with a trilateral filter and range image inpainting via belief propagation. The trilateral filter incorporates reflectivity, spatial, and intensity information to preserve geometric features such as jump edges and roof edges while effectively smoothing noise. The inpainting technique based on belief propagation restores deteriorated range images by leveraging not only neighboring range values but also the continuity of the reflectance image. Simulations and experiments using both synthetic and real range images acquired by a laser scanner demonstrate that the proposed techniques effectively suppress noise, preserve critical geometric features, and repair degraded range data.

Photo Raw range image Photo Bilateral filter Photo Proposed filter
Photo Shape recognition by slit laser using reflectivity Photo Reflectance image Photo Recognition of reflectance image

Papers

6D localization using range and reflectance data

We developed a 6D global localization technique that combines range and reflectance data from a laser scanner with the bag-of-features (BoF) approach. Global localization is essential for determining the accurate global position of a mobile robot when revisiting a known environment. Map-based global localization achieves high positional accuracy by computing precise transformations between the current observation and the pre-built map. However, it involves comparing large 3D datasets, which is computationally intensive. In contrast, appearance-based global localization uses image retrieval techniques to estimate the global position based on structural similarity. While it operates in real time, it typically relies on external illumination and performs poorly in low-light or dark environments. To address these limitations, we propose a hybrid approach that integrates map-based and appearance-based localization. Rather than using conventional camera images for appearance-based localization, we utilize reflectance images, which are captured as a byproduct of range sensing by the laser scanner. This enables robust global localization even in dark environments while maintaining real-time performance and high accuracy.

Photo Proposed technique Photo Pairs in reflectance images using BoF

Papers

Geometrical and photometrical modeling of real objects using laser range finders and digital cameras

Tokyo University, Ikeuchi Lab.

Textured image of Kamakura Great Buddha

Papers

  • Katsushi Ikeuchi, Takeshi Oishi, Jun Takamatsu, Ryusuke Sagawa, Atsushi Nakazawa, Ryo Kurazume, No Nishino, Mawo Kamakura and Yasuhide Okamoto, The Great Buddha Project: Digitally Archiving, Restoring, and Analyzing Cultural Heritage Objects, International Journal of Computer Vision, Vol.75, No.1, pp.189-208, (2007.2)
  • Ryo Kurazume, Ko Nishino, Mark D. Wheeler, Katsushi Ikeuchi, Mapping textures on 3D geometric model using reflectance image, Systems and Computers in Japan, Vol. 36, Issue 13, pp.92-101, (2005.11)
  • Takeshi Oishi, Atsushi Nakazawa, Ryo Kurazume, and Katsushi Ikeuchi, Fast Simultaneous Alignment of Multiple Range Images Using Index Images, in Proc. The 5th International Conference on 3-D Digital Imaging and Modeling (3DIM), pp.476-483,2005.
  • Yuichiro Hirota, Tomohito Masuda, Ryo Kurazume, Kohichi Ogawara, Kazuhide Hasegawa, Katsushi Ikeuchi, Flying Laser Range Finder and its data registration, Proc. IEEE Int. Conf. on Robotics and Automation, pp. 3155-3160, 2004.
  • Yuichiro Hirota, Tomohito Masuda, Ryo Kurazume, Koichi Ogawara, Kazuhide Hasegawa, Katsushi Ikeuchi, Designing a laser range finder which is suspended beneath a balloon, Proc. of Sixth Asian Conference on Computer Vision (ACCV), 2004.
  • Takeshi Oishi, Ryusuke Sagawa, Atsushi Nakazawa, Ryo Kurazume, Katsushi Ikeuchi, Parallel Alignment of a Large Number of Range Images, Proc. The 4th International Conference on 3D Digital Imaging and Modeling (3DIM 2003) , pp.195-202, 2003
  • Ryusuke Sagawa, Takeshi Oishi, Atsushi Nakazawa, Ryo Kurazume, Katsushi Ikeuchi, Iterative Refinement of Range Images with Anisotropic Error Distribution,Proceedings of the 2002 IEEE/RSJ Intl. Conference on Intelligent Robots and Systems, October 2002
  • Ryo Kurazume, Ko Noshino, Zhengyou Zhang, and Katushi Ikeuchi, Simultaneous 2D images and 3D geometric model registration for texture mapping utilizing reflectance attribute, Proc. of Fifth Asian Conference on Computer Vision (ACCV), vol.1, pp.99-106, 2002.
  • Katsushi Ikeuchi, Yutaka Takase, Ryo Kurazume, Takeshi Oishi, Ryusuke Sagawa, Ko Nishino, Modeling Cultural Heritage Through Observation, Int. Sympo. on Artificial Intelligence, Robotics and Human Centered Technology for Nuclear Applications, pp.26-32, 2002
  • Ryo Kurazume, Mark D. Wheeler, and Katushi Ikeuchi, Mapping textures on 3D geometric model using reflectance image, Workshop for Data Fusion at IEEE Int. Conf. on Robotics and Automation, 2001.