# Getting Started Guide - Metro Vision AI SDK ## Overview The Metro Vision AI SDK provides a comprehensive development environment for computer vision applications using Intel's optimized tools and frameworks. This guide demonstrates the installation process and provides a practical object detection implementation using DL Streamer and OpenVINO. ## Learning Objectives Upon completion of this guide, you will be able to: - Install and configure the Metro Vision AI SDK - Execute object detection inference on video content - Understand the basic pipeline architecture for computer vision workflows ## System Requirements Verify that your development environment meets the following specifications: - Operating System: Ubuntu 24.04 LTS or Ubuntu 26.04 LTS - Memory: Minimum 8GB RAM - Storage: 20GB available disk space - Network: Active internet connection for package downloads ## Installation Process Execute the automated installation script to configure the complete development environment: The installer clones [Metro AI Suite](https://github.com/open-edge-platform/metro-ai-suite) into `~/oep/metro-ai-suite`, plus Edge AI Libraries and Scenescape, all from `main`. ```bash curl -fsS https://raw.githubusercontent.com/open-edge-platform/edge-ai-libraries/refs/heads/main/tools/oep-sdk-manager/scripts/metro-vision-ai-sdk.sh | bash ```  The installation process configures the following components: - Docker containerization platform - DL Streamer video analytics framework - OpenVINO inference optimization toolkit - Pre-trained model repositories and sample implementations ## Object Detection Implementation This section demonstrates a complete object detection workflow using the installed components. ### Step 1: Create Working Directory Create a dedicated working directory for the tutorial: ```bash mkdir -p ~/oep/metro-vision-get-started-tutorial cd ~/oep/metro-vision-get-started-tutorial ``` ### Step 2: Download Sample Video and Model Run the Docker commands from your regular non-root host shell, without `sudo`. The `--user "$(id -u):$(id -g)"` option gives generated files your host UID and primary GID, so you can copy, move, and delete them without root privileges. Create the working directory before mounting it; these commands do not repair files created by earlier root-run containers. The download container uses the mounted workspace as its writable home for model preparation environments and caches. Download the Intel-hosted sample video, which shows pedestrians and a vehicle from a street-side perspective: ```bash wget -O sample.mp4 https://raw.githubusercontent.com/open-edge-platform/edge-ai-resources/main/videos/VIRAT_S_000101.mp4 ``` Download and convert the COCO-trained YOLO11s detector using the DL Streamer container: ```bash # Download YOLO11s model using DL Streamer docker run --rm --user "$(id -u):$(id -g)" \ -e HOME=/home/dlstreamer \ -e http_proxy -e https_proxy -e no_proxy \ -v "${PWD}:/home/dlstreamer/" \ intel/dlstreamer:2026.2.0-ubuntu24 \ bash -c "export MODELS_PATH=/home/dlstreamer && /opt/intel/dlstreamer/samples/download_public_models.sh yolo11s" ``` The downloader creates `public/yolo11s/FP16/yolo11s.xml` and `yolo11s.bin` in the working directory. YOLO11s detects COCO classes including `person`, `car`, `bus`, and `truck`; DL Streamer reads its class labels and preprocessing metadata directly from the exported model, so no model-proc file is needed. ### Step 3: Pipeline Execution Execute the object detection pipeline using the configured assets. ```bash # Allow the invoking user to access the local X11 display xhost +si:localuser:"$(id -un)" # Start the container docker run --rm -it --name dlstreamer \ --user "$(id -u):$(id -g)" \ -e HOME=/data \ -w /data \ -v "${PWD}:/data" \ --ipc=host \ --net=host \ -e DISPLAY=$DISPLAY \ -v /tmp/.X11-unix:/tmp/.X11-unix \ intel/dlstreamer:2026.2.0-ubuntu24 ``` ```bash # Run the GStreamer pipeline inside the running container gst-launch-1.0 filesrc location=/data/sample.mp4 ! \ decodebin3 caps="video/x-raw(ANY)" ! videoconvert ! \ gvadetect model=/data/public/yolo11s/FP16/yolo11s.xml device=CPU ! queue ! \ gvawatermark ! videoconvert ! autovideosink ``` The video begins with pedestrians; a vehicle enters the scene later (around the one-minute mark). Let the pipeline play until then to see both detection classes.  ```bash # To exit the container exit ``` Revoke the display permission from the host shell when finished: ```bash xhost -si:localuser:"$(id -un)" ``` ### Pipeline Architecture Analysis The executed command implements a processing pipeline with the following stages: 1. **File Input**: Reads the specified video file from the filesystem 2. **Video Decoding**: Converts encoded video stream to raw frame data 3. **Object Detection**: Applies the pre-trained model to identify pedestrians and vehicles 4. **Visualization**: Renders bounding boxes and classification labels on detected objects 5. **Display Output**: Presents the annotated video stream through the system display The resulting output displays the original video content with overlaid detection annotations indicating identified objects and their classification confidence levels. ## Technology Framework Overview ### DL Streamer Framework DL Streamer provides a comprehensive video analytics framework built on GStreamer technology. Key capabilities include: - Multi-format video input support (files, network streams, camera devices) - Real-time inference execution on video frame sequences - Flexible output rendering and storage options - Modular pipeline architecture for custom workflow development ### OpenVINO Optimization Toolkit OpenVINO delivers cross-platform inference optimization for Intel hardware architectures. Core features include: - Model format standardization through Intermediate Representation (IR) - Hardware-specific performance optimization - Extensive pre-trained model repository - Development tools for model conversion and validation ## Next Steps Continue your learning journey with these hands-on tutorials: ### [Tutorial 1: OpenVINO Model Benchmark](./tutorial-1.md) Learn to benchmark AI model performance across different Intel hardware (CPU, GPU, NPU) and understand optimization techniques for production deployments. ### [Tutorial 2: Multi-Stream Video Processing](./tutorial-2.md) Build scalable video analytics solutions by processing multiple video streams simultaneously with hardware acceleration. ### [Tutorial 3: Real-Time Object Detection](./tutorial-3.md) Develop a complete object detection application with interactive controls, performance monitoring, and production-ready features. ### [Tutorial 4: Advanced Video Analytics Pipelines](./tutorial-4.md) Create sophisticated video analytics using Intel® DL Streamer framework, including human pose estimation and multi-model integration. ### [Tutorial 5: Profiling](./tutorial-5.md) Profiling and monitoring performance of Metro Vision AI workloads using command-line tools. ## Additional Resources ### Technical Documentation - [DL Streamer](https://docs.openedgeplatform.intel.com/dev/edge-ai-libraries/dlstreamer/index.html) \- Comprehensive documentation for Intel's GStreamer-based video analytics framework - [DL Streamer Pipeline Server](https://docs.openedgeplatform.intel.com/dev/edge-ai-libraries/dlstreamer-pipeline-server/index.html) \- RESTful microservice architecture documentation for scalable video analytics deployment - [OpenVINO](https://docs.openvino.ai/2026/get-started.html) \- Complete reference for Intel's cross-platform inference optimization toolkit - [OpenVINO Model Server](https://docs.openvino.ai/2026/model-server/ovms_what_is_openvino_model_server.html) \- Model serving infrastructure documentation for production deployments - [Edge AI Libraries](https://docs.openedgeplatform.intel.com/dev/ai-libraries.html) \- Comprehensive development toolkit documentation and API references - [Metro AI Suite](https://docs.openedgeplatform.intel.com/dev/ai-suite-metro.html) \- Complete application suite documentation with implementation examples ### Support Channels - [GitHub Issues](https://github.com/open-edge-platform/edge-ai-libraries/issues) \- Technical issue tracking and community support :::{toctree} :hidden: tutorial-1 tutorial-2 tutorial-3 tutorial-4 tutorial-5 tutorial-6 :::