Raspberry Pi 5 Camera Streaming Project

A professional, ultra-low-latency camera streaming solution for Raspberry Pi 5 with Camera Module 3 (IMX708), featuring OpenVINO FP16 Accelerated YOLO AI Object Detection, WebRTC support, adaptive CPU resource management, Docker containerization, and a modern web viewer.

Created by CiscoPonce

---

🏆 Key Achievements & Verified System Status

[SUCCESS] 7/7 System Diagnostic Tests Passed!
--------------------------------------------------
[✓] Camera Hardware: Sony IMX708 (Module 3) @ 30 FPS (PiSP RP1 ISP)
[✓] Docker Engine & Container Orchestration: Active
[✓] SRS Realtime Media Server: Active (WebRTC / RTMP / FLV)
[✓] Web Viewer Interface (Nginx): Accessible on Port 80
[✓] OpenVINO AI WebSocket Service: Active on Port 8765
[✓] Network Port Access: All 5 required ports open
[✓] System Health: CPU 56.75°C | RAM 2.0GB/15GB (13.0GB free) | Load 4.33

---

⚡ Slimmed-Down Resource Performance

Component / Service	Idle / Unticked	Active Mode	Performance Optimization
YOLO OpenVINO AI Service	0.0% CPU	45.0% CPU	Adaptive standby loop (pauses 100% when unticked in UI)
FFmpeg H.264 Encoder	36.7% CPU	36.7% CPU	Tuned 30 FPS pipeline (-60% CPU savings vs 60 FPS)
PiSP Camera Capture	8.3% CPU	8.3% CPU	Native Pi 5 RP1 hardware ISP capture (/dev/media0)
SRS Media Server	8.4% CPU	8.4% CPU	Low-overhead Docker RTMP/WebRTC container
Total System Load	1.4	4.33	Leaves >70% CPU and 13GB RAM free for other projects

---

🎯 Project Showcase

This project demonstrates advanced engineering skills in:

• Embedded Systems Development with Raspberry Pi 5 & Camera Module 3
• Edge Computer Vision & AI using OpenVINO FP16 & YOLO11 Small
• Real-time Video Streaming using WebRTC, RTMP, and HTTP-FLV
• Adaptive CPU Resource Management (Client-aware inference pausing)
• Docker Containerization for scalable deployment
• Performance Optimization for ARM NEON SIMD vector hardware

---

🛠️ Technologies Used

• Hardware: Raspberry Pi 5 (16GB), Camera Module 3 (Sony IMX708)
• AI / Computer Vision: OpenVINO 2026 FP16 Engine, Ultralytics YOLO11 Small, OpenCV, WebSockets
• Streaming Engine: SRS (Simple Realtime Server v5), WebRTC, RTMP, HTTP-FLV
• Containerization: Docker, Docker Compose
• Web Server: Nginx with optimized reverse proxy
• Camera Pipeline: Native Pi 5 PiSP libcamera v0.7.1, rpicam-vid
• Video Encoding: Hardware-accelerated H.264 / FFmpeg
• Frontend UI: HTML5, Modern Vanilla CSS, WebRTC API, Canvas Bounding Box Renderer

---

✨ Features

• 🚀 Sub-500ms WebRTC Streaming: Direct WebRTC peer connections with instant HTTP-FLV fallback.
• 🤖 Real-time OpenVINO YOLO11 Overlay: Live target detection overlay (person, car, bicycle, etc.) toggleable directly from the web browser.
• 🌱 Adaptive CPU Saver: Unticking YOLO on the web UI pauses inference, dropping AI CPU usage to 0%.
• 🍓 Raspberry Pi 5 PiSP Hardware Native: Custom compiled rpi/pisp driver for the Pi 5 RP1 controller.
• 🐳 Docker Containerized: Isolated SRS and Nginx web services.
• 🌐 Modern Web Viewer: Responsive dark-mode UI with live status indicators, mute toggles, and fullscreen mode.
• 📊 100% Automated Testing: Diagnostic script (./scripts/test-setup.sh) verifying end-to-end component health.

---

📋 Network Ports & Service Architecture

Port	Service	Protocol	Description
80	Web Viewer	HTTP	Modern HTML5 Pi Camera Viewer UI
8765	YOLO AI Service	WebSocket	Real-time OpenVINO YOLO detection JSON stream
1935	RTMP Ingest	TCP	Live video stream ingest
1985	SRS API	HTTP	WebRTC signaling & server management API
8081	HTTP Console	HTTP	FLV playback endpoint (/live/cam.flv) & console
8000-8100	WebRTC ICE	UDP	Interactive connectivity establishment

---

🚀 Quick Start

1. Clone & Setup Dependencies

git clone https://github.com/CiscoPonce/pi-camera-streaming.git
cd pi-camera-streaming

# Install required build tools & Python packages
sudo apt update && sudo apt install -y \
  build-essential git cmake meson ninja-build libboost-dev \
  libgnutls28-dev openssl libtiff-dev pybind11-dev python3-yaml python3-ply \
  libglib2.0-dev libgstreamer-plugins-base1.0-dev libdrm-dev libexif-dev \
  libepoxy-dev libjpeg-dev libpng-dev libpisp-common libpisp1 \
  python3-pip python3-opencv python3-websockets python3-numpy

pip3 install --break-system-packages ultralytics openvino onnxruntime

2. Build Raspberry Pi 5 PiSP libcamera & rpicam-apps

# Build Raspberry Pi libcamera with Pi 5 PiSP support
git clone https://github.com/raspberrypi/libcamera.git /tmp/libcamera
cd /tmp/libcamera
meson setup build --buildtype=release -Dpipelines=rpi/vc4,rpi/pisp -Dipas=rpi/vc4,rpi/pisp -Dv4l2=true -Dpycamera=disabled -Dgstreamer=disabled -Ddocumentation=disabled
ninja -C build && sudo ninja -C build install && sudo ldconfig

# Build rpicam-apps
git clone https://github.com/raspberrypi/rpicam-apps.git /tmp/rpicam-apps
cd /tmp/rpicam-apps
meson setup build --buildtype=release -Denable_libav=disabled -Denable_qt=disabled -Denable_opencv=disabled -Denable_tflite=disabled
ninja -C build && sudo ninja -C build install && sudo ldconfig

3. Start Streaming & AI Services

cd pi-camera-streaming

# Start Docker containers (SRS + Nginx)
docker compose up -d

# Launch 30 FPS Hardware Camera Stream
./scripts/start-camera.sh --fps 30 &

# Launch OpenVINO YOLO AI WebSocket Service
python3 ai_vision/yolo_service.py &

4. Run System Diagnostic Test

./scripts/test-setup.sh

---

🌐 Accessing the Stream

Open your browser from any device on your local network:

• Web UI (Pi Camera Viewer): http://<PI_IP>/
• Direct FLV Stream: http://<PI_IP>:8081/live/cam.flv
• SRS Console: http://<PI_IP>:8081/

---

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

👨‍💻 Author

CiscoPonce

• GitHub: @CiscoPonce
• Repository: https://github.com/CiscoPonce/pi-camera-streaming