WATonomous / wato_monorepo

Dockerized ROS2 stack for the WATonomous Autonomous Driving Software Pipeline
Apache License 2.0
89 stars 1 forks source link

Association #152

Open Kishore-Yogaraj opened 1 month ago

Kishore-Yogaraj commented 1 month ago

Objective: To develop a system that associates objects detected by multiple overlapping cameras, ensuring consistent 3D tracking of the same object across different camera views for autonomous driving.

Key Details: Data Sources: Leverage multiple cameras providing different perspectives of the same scene to detect and track objects.

Goal: Correctly associate object detections from different cameras, ensuring they represent the same object in the environment and maintain consistent tracking across camera views.

Approach:

Align the 2D object detections from multiple overlapping cameras. Use projection techniques and camera calibration data to transform detections into a common 3D coordinate system. Implement a matching algorithm (e.g., epipolar geometry, nearest neighbor, IoU) to associate objects between the cameras. Handle occlusions and partial views to ensure robust object tracking. Expected Output: A reliable 3D tracking system where object detections from overlapping cameras are consistently associated, improving the continuity and accuracy of 3D object tracking in multi-camera setups.

This should help establish proper tracking between different camera views in a multi-camera setup!