Model-based Pose Measurement Using Structured Light Vision Sensor for a Target with Reflective Surface

Zhe Wang,Jun Hou,Zhaoyang Liu,Junfeng Fan,Fengshui Jing

DOI: https://doi.org/10.1109/cac53003.2021.9727686

2021-01-01

Abstract:Pose measurement is a critical procedure for automation engineering. For a target with reflective surface, achieving effective pose measurement under the interference of light reflection while maintaining high precision remains a challenge. To address this, we propose a model-based pose measurement method using structured light vision. We equip structured light vision sensor with optical filters and use it to capture the point cloud of the target, attenuating the impact of light reflection from the input. We define the canonical model for the measurement using a partial point cloud sampled from 3D model of the target, resembling the target point cloud collected from the sensor. The canonical model is then coarsely aligned with the target point cloud to estimate its pose based on respective principal axes instead of specific features, strengthening the robustness against noises while reducing the computational cost. Then the target pose from the coarse alignment is further refined using ICP registration, and the final target pose is thus obtained. Experiments are conducted to demonstrate the superior accuracy, robustness, and efficiency of the proposed measurement method, compared with other methods.