Visual Servoing

Problem Statement

Visual servoing controls UAV motion directly from image-space error signals. It is effective for target following and precision alignment when full 3D reconstruction is unavailable.

Model and Formulation

Given image feature error e = s - s^*, the control law is:

$$ \dot{q} = -\lambda L_s^+ e $$

where L_s is the interaction matrix and L_s^+ its pseudo-inverse. In bounding-box tracking, feature vectors include center and area terms.

Algorithm Procedure

1. Extract target feature in image frame.
2. Compute feature error to desired setpoint.
3. Convert image-space error to body-frame commands.
4. Apply velocity/attitude commands with saturation limits.

Tuning and Failure Modes

• Gain \lambda too high causes oscillatory camera motion.
• Target occlusion can destabilize command generation without fallback logic.
• Camera latency and rolling shutter distort high-speed tracking.

Implementation and Execution

python -m uav_sim.simulations.perception.visual_servoing

Evidence

References

• Chaumette and Hutchinson, Visual Servo Control Part I (2006)
• Chaumette and Hutchinson, Visual Servo Control Part II (2007)

Related Algorithms

• Gimbal Tracking
• Gimbal BBox Tracking