Vision-Based Obstacle Avoidance Strategies for MAVs Using Optical Flows in 3-D Textured Environments

Area of Science:

• Robotics
• Computer Vision
• Aerospace Engineering

Background:

• Micro aerial vehicles (MAVs) face payload limitations, driving research into lightweight, cost-effective vision-based navigation.
• Optical flow methods offer efficient obstacle avoidance but struggle with 3D environments and frontal, wall-like obstacles.

Purpose of the Study:

• To develop a vision-based obstacle avoidance algorithm for MAVs capable of handling 3D complex environments and wall-like obstacles.
• To improve MAV navigation stability and efficiency by addressing limitations of existing optical flow approaches.

Main Methods:

• A monocular camera captures images, which are split into horizontal and vertical half-planes.
• Optical flow sums in each half-plane determine heading direction and climb rate for 3D obstacle avoidance.
• Wall-like obstacle avoidance is achieved by analyzing optical flow divergence at the focus of expansion, with goal navigation using a sigmoid weighting function.

Main Results:

• The proposed algorithm successfully demonstrated obstacle avoidance in simulated and real-world indoor flight experiments.
• The method proved effective in complex 3D environments and in avoiding challenging wall-like frontal obstacles.
• Validated performance indicates improved stability and avoidance capabilities compared to existing methods.

Conclusions:

• The developed vision-based optical flow algorithm provides a robust solution for MAV obstacle avoidance in 3D textured environments.
• The approach effectively mitigates issues like jitter and unnecessary motion associated with previous methods when facing frontal obstacles.
• This research contributes to safer and more autonomous MAV operations in diverse and complex scenarios.