Tracking Objects in Videos

Objective: Implement the Lucas-Kanade algorithm to track an object in a video sequence.

Key Steps:

1. Initialization:

• Draw a bounding box around the target object in the first frame (template).

2. Iterative Tracking:

• For each subsequent frame:
  • Compute the updated warping parameters between the current frame and the initial frame using lucas kanade.
  • Warp the initial bounding box using the updated parameters.
  • Draw the bounding box on the current frame and append it to the new video sequence.

3. Lucas Kanade Algorithm

• Compute the image gradients (Ix and Iy) using Sobel operators.

• Perform iterative optimization to estimate the warping parameters:

  • Update the warp matrix (W) based on the current parameters (p).

  • Warp the coordinates using the warp matrix.

  • Extract pixel intensities (T) from the template.

  • Extract pixel intensities (I) from the next frame at the warped coordinates.

  • Compute the error (E) between the intensities of the template and the frame.

  • Compute steepest descent matrix (SD) based on image gradients and coordinates of ROI.

  • Compute the Hessian matrix (H) using the steepest descent matrix.

  • Update the warping parameters (p).

  • Check convergence by evaluating the norm of the parameter update.

  • Repeat until convergence or maximum iterations are reached.

• Return the warped matrix (W) and refined warping parameters (p).

Region of Interest:

• Car

• Landing Helicopter