Towards an Open-Source Benchmark for Underwater Object Detection and Pose Estimation
- Resource Type
- Conference
- Authors
- Saksvik, Ivar Bjorgo; Weydahl, Hakon; Teigland, Hakon; Alcocer, Alex; Hassani, Vahid
- Source
- 2023 IEEE Underwater Technology (UT) Underwater Technology (UT), 2023 IEEE. :1-5 Mar, 2023
- Subject
- Fields, Waves and Electromagnetics
Power, Energy and Industry Applications
Robotics and Control Systems
Autonomous underwater vehicles
Pose estimation
Object detection
Benchmark testing
Maintenance engineering
Inspection
Fiducial markers
Benchmark dataset
fiducial markers
object detection
pose estimation
- Language
Unmanned underwater vehicles (UUVs) are steadily expanding their merit in underwater inspection, maintenance, and repair tasks. A key step towards increasing their autonomy is to determine the pose of objects they interact with. This paper presents the work toward a benchmark dataset containing annotated poses and segmentation masks of industrial subsea handles and shackles. The latter are populated on a poster surrounded by fiducial markers to alleviate the object detection and pose estimation task. Video footage of the submerged poster is captured in a harbor by a remotely operating vehicle at varying depths. Detection and pose estimation of the handles and shackles are estimated by a region-based object-tracking approach. The dataset along with the tools to create it is publicly available at: https://github.com/OsloMet-OceanLab/UB_POSE