PALVO: visual odometry based on panoramic annular lens.

Area of Science:

• Robotics
• Computer Vision
• Sensor Fusion

Background:

• Visual odometry (VO) is crucial for autonomous systems but struggles with rapid motion and dynamic scenes.
• Existing VO methods lack robustness, limiting their real-world applicability.
• Panoramic annular lenses (PAL) offer a wider field of view, potentially improving VO robustness.

Purpose of the Study:

• To develop a robust visual odometry system using a panoramic annular lens (PAL).
• To enhance the performance of visual odometry in challenging scenarios like rapid motion and dynamic environments.
• To achieve state-of-the-art accuracy while significantly improving robustness.

Main Methods:

• Applying a panoramic annular lens (PAL) to visual odometry.
• Modifying the camera model for PAL and designing a specialized initialization using the essential matrix.
• Implementing a two-stage tracking process and probabilistic mapping based on Bayesian inference and epipolar curve feature matching.

Main Results:

• PALVO demonstrates significantly improved robustness to rapid motion.
• The system shows enhanced performance in dynamic scenarios compared to traditional VO.
• PALVO achieves accuracy comparable to existing state-of-the-art visual odometry techniques.

Conclusions:

• PALVO, utilizing a panoramic annular lens, offers a highly robust solution for visual odometry.
• The proposed method effectively addresses the fragility of VO in challenging real-world conditions.
• This approach presents a competitive and accurate alternative for visual odometry applications.