Live ROS 2 Object Tracking: RANSAC Ground Removal & persistent DBSCAN Clustering

A live RViz demonstration of our autonomous point cloud filtering and cluster isolation pipeline, integrated with RTAB-Map SLAM and simulated in Gazebo/PX4 (part of our IROS 2026 submission).

The video is divided into four synchronized panels showing the real-time processing of the drone's survey scan:

Top-Left (Raw SLAM): The incoming dense point cloud directly from RTAB-Map.
Top-Right (Ground Removal): The terrain surface after real-time RANSAC-based ground plane extraction.
Bottom-Left (DBSCAN Clustering): Raw density-based candidate clusters, color-coded. Notice the spurious transient noise scattered around the edges!
Bottom-Right (Geometrically Validated Target Clusters): The final output of the two-phase tracking pipeline. Only stable rock clusters (with consistent centroids, bounding boxes, and PCA eigenvectors) that persist across loop closures are retained, eliminating all SLAM artifacts and ground leakage.

Видео Live ROS 2 Object Tracking: RANSAC Ground Removal & persistent DBSCAN Clustering канала Kanav Prashar