Incremental Mapping from Image Sequences

• Applicant: Professor Dr. Cyrill Stachniss
• Subject Area: Geodesy, Photogrammetry, Remote Sensing, Geoinformatics, Cartography
• Term: from 2011 to 2019
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 166047863

This project addresses the problem of incremental mapping based on image sequences that have been recorded with an autonomous multicopter. The obtained map is a 3D point cloud that additionally stores local normal information together with the appearance and the uncertainty associated to each point. Within the first phase, we developed a fast and robust visual odometry approach that operates with 10\,Hz on 2 stereo camera pairs in parallel on the micro aerial vehicle. The visual odometry is based on an incremental bundle adjustment that can exploit points at infinity. This approach has been developed within this project and yields a robust and accurate estimate of the angular orientation of the copter in real time. Through the integration of GNSS observations from P1, the visual odometry is directly georeferenced. The same technique is also applied on the base station to compute the resulting map from the high-resolution monocular camera of the micro aerial vehicle. Our evaluations indicate that our approach results in a highly accurate georeferenced point cloud. In the second phase of this project, the overall focus stays the same but we will developed new techniques to extend and robustify the components developed in the first phase. In addition to that, we will explicitly address the integration of multiple image sequences so that a mission can consist of multiple flights. Besides a statistically sound, tightly coupled integration of the visual information with GNSS, IMU, and laser range data, we will explicitly address the problem of matching image sequences that have been recorded at different points in time and thus under changed conditions. We will furthermore increase the robustness of the the localization and mapping system of the copter. We aim at handling blackouts of the individual sensors and will propose strategies for coping with failures to successfully continues the mission.

• DFG Programme: Research Units
• Subproject of: FOR 1505: Mapping on Demand
• Co-Investigator: Professor Dr.-Ing. Wolfgang Förstner