Atriva Atriva
Documentation

Atriva Edge AI Platform

Guides, tutorials, and API references to help you build and deploy Edge AI applications.

Quick Start GitHub

Atriva Video Pipeline API Service - FFMPEG x86

Overview

The Atriva Video Pipeline API Service is a containerized microservice that provides FFmpeg-based video processing capabilities for the Atriva AI platform. This service is specifically built for x86 architectures and serves as a critical component in the video processing pipeline, decoding video streams and extracting frames for downstream AI inference services.

Key Purpose

This service acts as a video decoder and frame extractor that:

  • Receives video input from files or streaming URLs (including RTSP)
  • Decodes video streams using hardware-accelerated FFmpeg
  • Extracts frames as JPEG images at configurable frame rates
  • Stores decoded frames in a shared volume structure accessible by AI inference containers
  • Provides video metadata and processing status information

Platform-Specific Note

This implementation targets x86/x86_64 architectures. Similar containers may exist for other platforms (ARM, etc.) with platform-specific optimizations.

Features

  • 🚀 Video Upload & Processing - Support for file uploads and URL-based video sources
  • 🎥 RTSP Stream Support - Real-time streaming with auto-reconnection
  • ⚡ Hardware Acceleration - CUDA, QSV, and VAAPI support for x86 platforms
  • 📸 Frame Extraction - Extract frames as JPEGs at configurable FPS
  • 📊 Video Metadata - Retrieve video information and codec details
  • 🔄 Multi-Camera Support - Concurrent processing for multiple camera streams
  • 🐳 Dockerized - Containerized for easy deployment and scaling
  • 📁 Shared Storage - Decoded frames stored in shared volumes for AI inference

Quick Start

Prerequisites

  • Docker (with support for hardware acceleration if needed)
  • NVIDIA/Intel GPU drivers (for hardware acceleration)
  • Shared volume mount for frame storage

Build & Run

# Build the Docker image
docker build -t atriva-video-pipeline-x86 .

# Run the container
docker run --rm -p 8002:8002 \
  --device /dev/dri:/dev/dri \  # For VAAPI acceleration
  -v /path/to/shared/frames:/app/frames \  # Shared frame storage
  -v /path/to/videos:/app/videos \  # Video storage
  atriva-video-pipeline-x86

Documentation

For detailed documentation, see the docs/ directory:

  • Architecture - System architecture and shared storage structure
  • Installation - Detailed installation and deployment guide
  • Development - Development setup and contribution guidelines
  • Testing - Testing procedures and examples
  • Integration - Integration guide for other services

API Endpoints

Core Endpoints

  • POST /api/v1/video-pipeline/decode/ - Start decoding video and extracting frames
  • GET /api/v1/video-pipeline/decode/status/ - Get decoding status for a camera
  • POST /api/v1/video-pipeline/decode/stop/ - Stop decoding for a camera
  • GET /api/v1/video-pipeline/latest-frame/ - Get the latest decoded frame
  • POST /api/v1/video-pipeline/video-info/ - Get video metadata
  • GET /api/v1/video-pipeline/hw-accel-cap/ - Check hardware acceleration capabilities
  • GET /api/v1/video-pipeline/health/ - Health check endpoint

See Integration Guide for detailed API documentation.

Shared Frame Storage

Decoded frames are stored in a shared volume structure:

/app/frames/
├── {camera_id_1}/
│   ├── frame_0001.jpg
│   ├── frame_0002.jpg
│   └── ...
├── {camera_id_2}/
│   ├── frame_0001.jpg
│   └── ...
└── ...

This structure allows AI inference containers to access frames directly from the shared volume without requiring API calls.

License

MIT License

Contributing

Pull requests are welcome! Please see the Development Guide for contribution guidelines.