Development of an Online Adaptive Parameter Tuning vSLAM Algorithm for UAVs in GPS-Denied Environments

Area of Science:

• Robotics
• Computer Vision
• Navigation Systems

Background:

• Unmanned Aerial Vehicles (UAVs) heavily rely on Global Navigation Satellite Systems (GNSS) for positioning.
• GNSS is susceptible to environmental interference and ineffective indoors, limiting UAV applications.
• Existing visual simultaneous localization and mapping (vSLAM) systems face challenges with parameter tuning, impacting accuracy.

Purpose of the Study:

• To develop an enhanced vSLAM system for UAVs that overcomes GNSS limitations.
• To improve the accuracy and robustness of UAV localization in unknown environments.
• To address the challenge of manual threshold setting in feature matching for vSLAM.

Main Methods:

• Implemented a stereo-based vSLAM algorithm fused with onboard Inertial Measurement Unit (IMU) data.
• Developed an online adaptive matching threshold based on keyframe poses to replace manual Hamming distance thresholds.
• Integrated an adaptive gain tuning for the Mahony complementary filter and a static state detection algorithm for IMU-vSLAM integration.

Main Results:

• The developed online adaptive matching threshold significantly improved vSLAM positioning accuracy.
• Dynamic tuning of the IMU's Mahony complementary filter enhanced attitude estimation during aggressive motions.
• The static state detection algorithm improved the initial guess for the bundle adjustment algorithm, boosting overall performance.

Conclusions:

• The proposed online adaptive parameter tuning algorithm effectively enhances vSLAM accuracy and robustness for UAVs.
• The integrated system provides reliable localization information without reliance on GNSS signals.
• This research expands the potential applications of UAVs in GNSS-denied or indoor environments.