Proposal: LiDAR Point Cloud Clustering using DBSCAN

[rai-sukant]

Description :

DBSCAN (Density-Based Spatial Clustering of Applications with Noise) is a density-based clustering algorithm that groups LiDAR points into objects based on local point density rather than fixed distance thresholds. It identifies “core” points with sufficient nearby neighbors, expands clusters from them, and treats isolated points as noise, making it intuitive for explaining how objects can be segmented from raw point clouds in a 2D LiDAR setting.

Proposed Tasks :

• Implement a DBSCANDetector module in src/components/detection/dbscan/ that clusters point clouds using KdTree from the existing repo.
• Visualize the clustering process step-by-step (e.g., showing visited points, growing clusters, and final cluster boundaries).
• Integrate the new detector with the existing OmniDirectionalLidar, Vehicle, and GlobalXYVisualizer framework so it can run similarly to the current rectangle fitting example.
• Add a new sample program under src/simulations/perception/ demonstrating DBSCAN-based LiDAR point cloud clustering with a simple matplotlib animation.
• Document the algorithm conceptually in a short comment/header explaining eps, min_samples, core points, border points, and noise.

Reference Code : https://github.com/ShisatoYano/AutonomousVehicleControlBeginnersGuide/tree/main/src/simulations/perception/point_cloud_rectangle_fitting

[ShisatoYano]

Hi @rai-sukant , Thank you for proposing such a good issue. I understood what the algorithm is and what you will implement. You can start implementing the code. I`m looking forward to seeing your PR.

[rai-sukant]

@ShisatoYano Excuse me for the delay, will be opening PR with updates soon.