Nonlinear Complementary Filter for Attitude Estimation by Fusing Inertial Sensors and a Camera.

Area of Science:

• Computer Vision
• Robotics
• Sensor Fusion

Background:

• Pose estimation typically requires at least four feature points for camera-based algorithms.
• Limited feature points pose a significant challenge for traditional point correspondence methods.

Purpose of the Study:

• To develop an accurate attitude estimation algorithm using only two visual feature points.
• To fuse camera data with low-cost inertial sensors for robust pose estimation.

Main Methods:

• A nonlinear complementary filter designed on the special orthogonal group SO(3).
• Derivation of an implicit geometry measurement model from two feature points.
• Fusion of visual geometry, angular rate, and vector measurements from inertial sensors.

Main Results:

• The proposed algorithm demonstrates superior accuracy compared to existing methods.
• The filter ensures locally asymptotic stability, validated by nonlinear system analysis.
• A quaternion-based implementation offers computational efficiency.

Conclusions:

• The novel approach effectively addresses the challenge of pose estimation with minimal visual features.
• Fusion of camera and inertial sensors via a nonlinear complementary filter provides a robust solution.
• The algorithm is practical for real-world applications, as shown by smartphone validation.