Towards Accurate Ground Plane Normal Estimation from Ego-Motion.

Area of Science:

• Robotics
• Computer Vision
• Autonomous Systems

Background:

• Dynamic changes in ground planes (e.g., braking, uneven surfaces) cause vehicle pose oscillations, particularly affecting pitch angle.
• Accurate ground plane normal estimation is crucial for improving the reliability of various autonomous driving tasks.
• Existing methods may struggle with real-time performance or dynamic environments.

Purpose of the Study:

• To introduce a novel, real-time method for estimating ground plane normals for wheeled vehicles.
• To enhance the robustness of autonomous driving systems by leveraging accurate ground plane information.
• To provide a method compatible with both camera- and inertial-based odometry.

Main Methods:

• The proposed method utilizes odometry (ego-motion) to infer ground plane normals.
• An Invariant Extended Kalman Filter (IEKF) is designed to estimate the normal vector in the sensor's coordinate frame.
• The approach exploits the relationship between ego-pose odometry and the local ground plane.

Main Results:

• The method achieves accurate ground plane normal estimation in real time.
• It demonstrates improved robustness for autonomous driving tasks.
• State-of-the-art accuracy was achieved on the KITTI dataset with an estimated vector error of 0.39°.

Conclusions:

• The proposed method is simple, efficient, and effective for real-time ground plane normal estimation.
• It significantly improves the robustness of autonomous driving systems.
• The approach is versatile, supporting various odometry inputs.