How ak30072002/orin-quadcam-yolo works
This repository describes a Jetson AGX Orin application for four-camera YOLOv8 detection, tracking, and live display. Its principal workflow captures four CSI streams, converts and queues frames in GPU memory, batches them for TensorRT inference, tracks detections, then composites and displays the result. The graph follows the README’s documented pipeline; source excerpts were unavailable, so implementation details beyond the documented script and inference configuration are not inferred.
Main entry points: quadcam.sh TensorRT inference · config_infer_yolov8n.txt Inference settings
Components 13
Capture and batching
- CSI capture
scripts/quadcam.sh - UYVY conversion
scripts/quadcam.sh - Bounded queues
scripts/quadcam.sh - Stream batching
scripts/quadcam.sh
Inference
- TensorRT inference
scripts/quadcam.sh - Inference settings
config/config_infer_yolov8n.txt - YOLO bbox parser
Tracking and display
- NvDCF tracking
scripts/quadcam.sh - 2×2 tiler
scripts/quadcam.sh - Detection overlay
scripts/quadcam.sh - 3D display sink
scripts/quadcam.sh
Other components
- Four CSI cameras
- Live display
Connections 12
- Four CSI cameras → CSI capture: supply frames
README.md - CSI capture → UYVY conversion: outputs UYVY
README.md - UYVY conversion → Bounded queues: outputs NV12
README.md - Bounded queues → Stream batching: feed streams
README.md - Stream batching → TensorRT inference: batch frames
README.md - TensorRT inference → Inference settings: reads settings
README.md - TensorRT inference → YOLO bbox parser: parses detections
README.md - TensorRT inference → NvDCF tracking: send detections
README.md - NvDCF tracking → 2×2 tiler: send tracked frames
README.md - 2×2 tiler → Detection overlay: send composite
README.md - Detection overlay → 3D display sink: send overlay
README.md - 3D display sink → Live display: render frames
README.md