Optimization Based Multi-View Coverage Path Planning for Autonomous Structure From Motion Recordings

Abstract

The area of autonomous mapping and autonomous maintenance using unmanned robots has received much attention in recent years. However, usually only single-view coverage is ensured, which is not sufficient if a camera-based three-dimensional reconstruction is desired. If high precision is required for this reconstruction, computations are expensive and not real-time feasible, such that recording and reconstruction procedure must be executed consecutively. This calls for a reliable method for collecting a sufficient set of images during the recording procedure. In this letter, we present the first approach for multi-view coverage path planning, specifically targeted at an autonomous structure from motion data acquisition. For that purpose, we introduce an incremental, computational efficient update formula, that gives an upper bound on the expected reconstruction quality of observed points on the boundary of a structure. Using this quality estimate, we iteratively solve the coverage path-planning problem by greedily selecting a next-best-view as a solution to an optimal experimental design problem. Additional geometric and dynamic constraints ensure the admissibility of the solution. The correctness of our quality estimate and coverage path-planning approach are validated in the physics simulator Gazebo.