Advanced Monocular Outdoor Pose Estimation in Autonomous Systems: Leveraging Optical Flow, Depth Estimation, and...

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Artificial Intelligence

Background:

• Global Positioning System (GPS) limitations in obstructed environments necessitate GPS-independent localization for autonomous vehicles (AVs).
• Existing solutions using LiDAR or stereo cameras are often costly and complex.
• Monocular vision offers a practical, cost-effective alternative but lacks robust pose estimation models.

Purpose of the Study:

• To develop a novel adaptive framework for outdoor pose estimation and safe navigation using enhanced visual odometry (VO) with monocular cameras.
• To address the need for robust, GPS-independent localization solutions adaptable to various platforms and cost constraints.
• To ensure safety and real-time decision-making for AVs in GPS-denied areas.

Main Methods:

• Utilized visual odometry (VO) to estimate camera pose from image sequences in GPS-denied environments.
• Developed an adaptive framework leveraging monocular vision, advanced control theory, and machine learning.
• Integrated AI-driven models to meet multi-sensor system performance standards.

Main Results:

• Achieved significant improvements in pose estimation accuracy on the KITTI odometry dataset.
• Demonstrated a cost-effective and robust solution for real-world AV applications.
• Enhanced AV safety and performance in complex traffic scenarios through integrated control theory.

Conclusions:

• The proposed adaptive visual odometry framework provides a reliable and cost-effective GPS-independent localization solution for autonomous vehicles.
• This approach enhances navigation safety and adaptability in challenging environments where traditional GPS is unavailable.
• The research advances monocular vision capabilities for autonomous systems, paving the way for wider adoption.