The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Publications Copernicus
Articles | Volume XLI-B1
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLI-B1, 675–679, 2016
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLI-B1, 675–679, 2016

  06 Jun 2016

06 Jun 2016


Chun Liu, Zhengning Li, and Yuan Zhou Chun Liu et al.
  • College of Surveying and Geo-Informatics, Tongji University, Shanghai 200092, China

Keywords: Stereo Visual Odometry, 3D Reconstruction, Parallel Structure, Underground Infrastructure

Abstract. Presently, we developed a novel robust motion estimation method for localization and mapping in underground infrastructure using a pre-calibrated rigid stereo camera rig. Localization and mapping in underground infrastructure is important to safety. Yet it’s also nontrivial since most underground infrastructures have poor lighting condition and featureless structure. Overcoming these difficulties, we discovered that parallel system is more efficient than the EKF-based SLAM approach since parallel system divides motion estimation and 3D mapping tasks into separate threads, eliminating data-association problem which is quite an issue in SLAM. Moreover, the motion estimation thread takes the advantage of state-of-art robust visual odometry algorithm which is highly functional under low illumination and provides accurate pose information. We designed and built an unmanned vehicle and used the vehicle to collect a dataset in an underground garage. The parallel system was evaluated by the actual dataset. Motion estimation results indicated a relative position error of 0.3%, and 3D mapping results showed a mean position error of 13cm. Off-line process reduced position error to 2cm. Performance evaluation by actual dataset showed that our system is capable of robust motion estimation and accurate 3D mapping in poor illumination and featureless underground environment.