[Task] Centerline Generation in 2D

[mjorgecruz]

Description

Averaging Boundaries:

• Compute the 2D centerline by taking the midpoint between corresponding left and right boundary points.
• This centerline gives a preliminary trajectory within the open lane area.
  Refinement:
• Optionally, smooth the centerline using a filter (e.g., a moving average or spline fitting) to remove noise.

Priority

Medium

[luis-ffe]

Summary

The current pipeline:

• Segmentation: Converting video frames into a binary mask via an ONNX neural network.
• Lane Extraction: Splitting the mask into left/right halves and extracting lane points from the lower half.
• Clustering: Grouping lane points to handle multiple candidate lanes.
• Robust Polynomial Fitting: Using RANSAC to fit quadratic polynomials with a confidence score.
• Temporal Smoothing & Kalman Filtering: Integrating a Kalman filter and exponential smoothing to stabilize lane estimates.
• Candidate Selection: Choosing the two lanes closest to the vehicle (by evaluating the bottom of the image).
• Trajectory Computation: Averaging the left and right lanes to compute a mid-lane or math out when only 1 valid using previous midlane as weight.
• Visualization: Drawing the detected lanes on an overlay blended with the original frame.