Contrôle de Drone avec IBVS

Image-Based Visual Servoing for Drone Control

• Objective: Control a drone's movement using visual feedback (IBVS) from a camera. Align the drone with a target object or maintain defined visual parameters.

• Methods:

  • ROS (Robot Operating System): Used for communication and control.
  • Python: Used for scripting.
  • Libraries: Include PID control and geometric transformation libraries.

• Real-time Processing: The system processes camera streams to compute the desired drone motion in real time.

• Callback Method: Processes incoming target coordinates from the camera, extracting center, width, height, detection state, and bounding box area.

• Control Loop: Calculates the desired drone velocities (vx, vy, yaw_rate) based on errors and PID control.

• MAVROS Framework: Used for drone interaction.

• Node Initialization: Initializes a ROS node using rospy.init_node().

• Failsafe Mechanism: When the script terminates, the drone stops motion by publishing zero velocities.

• ROS Installation on Windows:

• Run WSL2 (Windows Subsystem for Linux).

• Use Ubuntu 20.04 and ROS Noetic.

• Install ROS Noetic using apt package management.

Homogeneous Transformations and Coordinate Frames

• Homogeneous Transformation Matrices: Used to represent transformations between coordinate frames.
• Relative Pose: A point's coordinates in one frame can be expressed in another frame using a homogeneous transformation matrix.
• Composition: Transformations can be composed to express a point in one frame relative to a third frame through an intermediate frame.
• Inverse Transformation: The inverse transformation can be calculated.
• Rotation about an Arbitrary Vector: The rotation matrix can be expressed using the Rodrigues' equation. The rotation matrix is encoded in the eigenvalues and eigenvectors.

Camera Calibration and Perspective Projection

• Calibration Matrix (𝐾): Also known as the camera parameter matrix.
• Perspective Projection: Has limitations, particularly a limited field of view. A major constraint is that Z = 0 is not allowed in this model.
• Alternative to Refractive Optics: Reflective surfaces like mirrors for image formation are explored.

Stereo Vision and Epipolar Geometry

• Stereo System: Two cameras with associated coordinate frames {1} and {2} and their image planes are considered.
• Fundamental Matrix (𝐹): A 3x3 matrix that relates corresponding points in the two images.
• Epipolar Geometry: The relationship between the corresponding points in the two images.
• Epipoles: Calculated from the null-space of the fundamental matrix (expressed in homogeneous and euclidean coordinates).
• Essential Matrix (𝐸): A function of the relative pose of camera {1} with respect to camera {2}.

Image Coordinates and Parameterizations

• Image Coordinates (𝑥, 𝑦): Image coordinates (in meters).
• (ρ, θ) Parameterization: A way to represent points in an image using radial distance and angle from the origin. The relation between image coordinates and (ρ, θ) needs to be clarified with the figure and a web search.
• Position-Based Visual Servoing: Explained visually in the provided figure.

Important Equations

• Equations for homogeneous transformations, rotations, fundamental matrix, and essential matrices.

Ce quiz explore les méthodes de contrôle de drone utilisant le Servoing Visuel Basé sur l'Image (IBVS). Les participants apprendront à utiliser ROS et Python pour aligner un drone sur un objet cible en traitant des flux caméra en temps réel. Le quiz aborde également des concepts comme le contrôle PID et les mécanismes de sécurité.