Jetson AI Service

A lightweight and high-performance object detection service designed for NVIDIA Jetson devices.
Built on YOLOv8, this service processes real-time video streams from local cameras or RTSP sources,
providing a self-contained solution for edge AI applications.

---

Key Features

• Offline Operation: Runs completely independent of external servers or APIs.
  All detection and telemetry outputs are printed to the console.

• Plug-and-Play AI: Starts automatically by loading a local YOLOv8 model and connecting to the specified camera device.

• Real-Time Performance: Uses an intelligent, multi-threaded worker system to ensure continuous, non-blocking video stream processing.

• System Telemetry: Monitors and reports critical system metrics (CPU, GPU, memory) directly to the console.

• Dockerized Deployment: Deployed using a single Docker Compose file, ensuring a consistent and easily manageable environment.

---

Getting Started

Prerequisites

• NVIDIA Jetson device (e.g., Orin Nano, AGX Orin)
• Docker Engine and Docker Compose installed
• NVIDIA Container Runtime configured on the Jetson device

---

Project Structure

Before running, create the following directory structure: jetson-ai-service/

├── app/

│ └── main.py # The main service script

├── models/

│ └── yolov8n.pt # Your YOLOv8 model file

├── logs/ # Logs will be stored here

├── outputs/ # Detected images will be saved here

└── docker-compose.yml # The Docker orchestration file

• Place the provided main.py script into the app/ directory.
• Place your YOLOv8 model (.pt file) inside the models/ directory. (Default: yolov8n.pt)

---

Running the Service

Execute the following command in the root directory of your project:

docker compose up

This command builds and runs the container, mapping the local directories to the container and starting the AI service.

Configuration

You can customize the service by editing the environment variables in the docker-compose.yml file.

Variable	Default Value	Description
MODEL_PATH	/workspace/models/yolov8n.pt	Path to the YOLOv8 model file inside the container.
CONF	0.25	Object confidence threshold for detections.
IMG_SIZE	640	Image size for inference (e.g., 640 for 640x640).
DETECTION_INTERVAL	0.5	Time in seconds between each detection cycle.
OUTPUT_DIR	/outputs	Directory where detected images with bounding boxes are saved.

---

Usage

Viewing Detections

The service prints all detections as JSON objects directly to standard output.
You can view them by checking the container logs:

docker compose logs -f

The output will include detection details, bounding boxes, and timestamps.

Viewing System Telemetry

The service prints system metrics (CPU, GPU, memory, disk usage) to the console every 10 seconds. You can monitor the Jetson's performance in real-time within the same logs.

License

This project is licensed under the MIT License.