Acquisition of a 3D Map Reduced Blind Spots in Multi View Stereo by Re-planning Autonomous Flight of a Drone

2015 Fiscal Year Final Research Report

• Project/Area Number: 25330213
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Perceptual information processing
• Research Institution: Osaka Electro-Communication University
• Principal Investigator: Echigo Tomio 大阪電気通信大学, 情報通信工学部, 教授 (80434801)
• Project Period (FY): 2013-04-01 – 2016-03-31
• Keywords: structure from motion / multi view stereo / occupancy voxel / SLAM / blind spot / 観測不十分領域 / 自律飛行

Outline of Final Research Achievements

This paper describes a method to acquire a 3D map reduced blind spots in multi view stereo from a drone. It is convinced that a drone becomes a significant tool for emergency investigation of disaster conditions. However, blind spots are happened in a image sequence captured from a flight along a prior planning route, because unexpected obstacles hide some regions from a drone, where a rescuer may exist.
The proposed approach employs a multi view stereo for 3D reconstruction. The moving spaces of a drone are discriminated as vacancies, obstacles, and regions occluded by obstacles. Insufficient observation regions are defined as observable ones from a few viewpoints and sightless ones from other lots of viewpoints. The new viewpoint is determined for observing the largest insufficient observation regions. Next, the route of the minimal distance between the current and the new viewpoint is investigated. Finally, autonomous flight of the drone can complement a 3D map.

Free Research Field

コンピュータビジョン