Plane-Aided Visual-Inertial Odometry for 6-DOF Pose Estimation of a Robotic Navigation Aid.

Area of Science:

• Robotics
• Computer Vision
• Sensor Fusion

Background:

• Visual-inertial odometry (VIO) fuses camera and inertial measurement unit (IMU) data for 6-DOF pose estimation.
• Current VIO methods struggle with online evaluation of visual odometry (VO) accuracy, limiting IMU bias estimation.
• Accurate VO output is crucial for reliable VIO performance.

Purpose of the Study:

• To introduce a novel VIO method, PAVIO, for enhanced pose estimation in robotic navigation aids.
• To develop a reliable online method for evaluating VO accuracy within a VIO system.
• To improve IMU bias estimation and reduce camera pose error.

Main Methods:

• PAVIO utilizes a 3D time-of-flight camera and extracts planes from 3D point clouds.
• Planes are tracked across camera views using IMU measurements, with covariance computed for parameter estimation.
• A chi-square test validates VO output accuracy; accepted data refines a factor graph for pose and bias optimization.

Main Results:

• PAVIO effectively evaluates VO output accuracy online.
• The method improves IMU bias estimation accuracy.
• Experimental results demonstrate reduced camera pose error for robotic navigation aids.

Conclusions:

• PAVIO offers a robust solution for accurate VIO pose estimation.
• The plane-consistent check enhances the reliability of VIO systems.
• PAVIO is applicable to any 3D-camera-based visual-inertial navigation system.