Edge AI Suites v1.2.0

Edge AI Suites

Release Overview

Edge AI Suites v1.2.0 delivers enhancements across intelligent video analytics, multimodal sensor fusion, and edge AI deployment. This release supports scalable, real-time AI solutions for smart cities, transportation, safety and security, industrial, robotics, and retail environments. Key improvements include unified sample applications, GenAI integrations, expanded hardware support, and streamlined developer experiences.

Suite-Specific Updates

Metro AI Suite

The Metro AI Suite in version 1.2.0 introduces a comprehensive set of enhancements aimed at accelerating the development and deployment of intelligent edge AI solutions for urban environments, particularly in smart city, transportation, and public safety domains. This release focuses on improving modularity, scalability, and real-time analytics capabilities through updated sample applications, SDKs, and developer tools.

Smart NVR Sample Application

• A newly released sample application that functions as a next-generation Network Video Recorder (NVR)
• Leverages Generative AI-powered vision analytics to enable advanced event detection, summarization, and automation
• Reduces bandwidth and storage requirements by performing video processing and analysis directly at the edge
• Integrates with existing video analytics pipelines such as:
  • Video Search and Summarization
  • Image-based Video Search
• Uses Frigate, a third-party open-source NVR, as a reference implementation
• Deployment is supported via Docker Compose on Ubuntu

Metro Vision AI App Recipe

• • Provides a unified architectural framework for building Smart Transportation applications
    • Includes curated deployment packages and improved tutorial guidance
• Introducing the Smart Intersection sample application:
  • Combines analytics from multiple traffic cameras to create a unified intersection view
  • Enables object tracking across multiple viewpoints, motion vector analysis (e.g., speed and heading), and 3D object interaction understanding
  • Demonstrates how existing camera infrastructure can be repurposed for real-time, multi-camera scene analytics
  • Highlights a shift from traditional frame-based analysis to integrated, edge AI-driven traffic management
  • Ingests output from DL Streamer Smart Parking pipeline, which includes:
    • License Plate Recognition
    • Optical Character Recognition
    • 3D object detection using third party/commercial models from DeepScenario
• Tutorials cover:
  • Building an AI-powered Tolling System
  • Customizing applications for specific use cases
• Enhancements to previously released sample applications:
  • Loitering Detection
  • Smart Parking

Metro AI Suite SDK

• A modular and comprehensive toolkit designed to accelerate development of visual AI solutions.
• Includes updated versions of:
  • Intel® Distribution of OpenVINO Toolkit
  • DL Streamer
  • FFmpeg
  • OpenCV (with OneVPL support)
• Refreshed documentation and tutorials.
• New tutorial added to demonstrate running inference pipelines using DL Streamer.
• SDK image now includes the latest version of the DL Streamer Pipeline Server.

Visual Pipeline and Platform Evaluation Tool

• Introduces the Simple Video Structurization pipeline, a flexible and use-case-agnostic solution.
• Supports:
  • License Plate Recognition
  • Vehicle Detection with Attribute Classification
  • Other object detection and classification tasks
• Adaptable based on selected models
• Adds live output support, allowing users to view real-time results directly in the UI
• Includes new pre-trained models:
  • YOLO v8 License Plate Detector
  • PaddleOCR
  • Vehicle Attributes Recognition Barrier 0039

Edge System Qualification Tool

• Designed for device and equipment manufacturers
• Enables verification and benchmarking of hardware performance
• Generates system qualification reports
• This release includes high-priority bug fixes for improved user experience, including:
  • iGPU ID resolution
  • Media runner stability

Video Search & Summarization

• Provides intelligent video search and summarization capabilities.
• Supports:
  • Video Summary for longform videos
  • Agentic Video Search using a Visual Data Management System
• Validated on Intel® AI Edge systems with:
  • Intel® Arc™ Graphics B580
  • Intel® Arc™ Graphics A770
• Deployment supported via:
  • Helm
  • Docker Compose on Ubuntu
  • Edge Microvisor Toolkit (limited support)

Video Processing Platform (VPP)

• Enables evaluation and optimization of video processing workflows for NVR use cases.
• Supports a wide range of hardware platforms including:
  • Intel® Core™ 13th and 14th Gen
  • Intel® Core™ Ultra Series 1 processors
• Addresses performance and compatibility improvements based on user feedback.

Sensor Fusion for Traffic Management

• Adds support for multiple sensor fusion configurations:
  • 2 Camera + 1 Radar on Intel® Celeron™ Processor 7305E
  • 4 Camera + 4 Radar on Intel® Core™ Ultra 7 Processor 165H
  • 16 Camera + 4 Radar on Intel® Core™ i7-13700 with Intel® Arc™ Graphics A770/A750E

Interactive Digital Avatar Platform Blueprint

• Now supports 2D Digital Avatars with:
  • English and Chinese language capabilities
  • Integration of QWen 2 and LLM RAG models
  • Enhanced multimodal intelligence

Visual Search Question and Answering (Q&A) Sample Application

• Combines a multimodal search engine and a visual Q&A assistant.
• Integrates models such as:
  • CLIP
  • CN-CLIP
  • QWen2.5-VL
• Includes a persistent vector database for improved multimodal search and question answering.

Image-Based Video Search Sample Application

• Includes bug fixes and component upgrades to improve stability and performance.

Manufacturing AI Suite

The Manufacturing AI Suite in version 1.2.0 delivers a robust set of tools and sample applications designed to streamline the development, deployment, and scaling of AI solutions in industrial environments. This release emphasizes configurability, real-time integration, and support for edge-native deployment scenarios, enabling developers and system integrators to build intelligent manufacturing applications with greater efficiency and precision.

HMI Augmented Worker Application

• Introduces a new Retrieval-Augmented Generation (RAG)-enabled Human-Machine Interface (HMI) application.
• Designed for deployment in virtualized environments, specifically for evaluation in conjunction with Edge Microvisor Toolkit.
• Architecture includes:
  • A Windows VM hosting the HMI application
  • A watcher service that monitors the knowledge base for updates
  • A RAG pipeline running in the Linux environment, enabling dynamic response generation based on updated knowledge.
• Optimized recipe for Intel® Core™ 13th generation CPU + virtualized setup.
• Showcase how AI can enhance operator interfaces by integrating real-time data, contextual awareness, and intelligent assistance

Predictive Maintenance for Wind Turbines Sample Application

• Introduces a new predictive analytics application for wind turbine anomaly detection
• Includes time series data ingest (Telegraf), data storage (InfluxDB), analytics (Kapacitor), visualization (Grafana), and publishing (Kapacitor)

Customizable Basic Application Architecture

• Migrate to a modular and configurable application framework tailored for common industrial use cases
• Supports easy customization through variable configuration, allowing developers to adapt the architecture to specific deployment needs
• Deployable through Docker and Helm charts for simplified evaluation and streamlined adoption.
  • A deployment package for Edge Management Framework (EMF) will be released alongside the future EMF 2025.2 release.
• Supported use cases include:
  • Pallet Defect Detection (PDD): Detects structural or surface-level defects in pallets using vision-based AI
  • Weld Porosity Detection: Identifies porosity and inconsistencies in welds, improving quality assurance in metal fabrication
  • Anomaly Detection on PCB: Uses AI to detect irregularities in printed circuit boards, aiding in electronics manufacturing
  • Worker Safety: Monitors worker behavior and environment to identify safety violations or hazardous conditions
• Added support for diverse accelerators including integrated-GPU, discrete-GPU, and NPU in addition to broad CPU support.

DL Streamer Pipeline Server Enhancements

• The DL Streamer Pipeline Server now supports enhancements to improve interoperability with existing industrial data platforms and robotic control systems, facilitating closed-loop automation and monitoring.
  • Metadata publishing to InfluxDB, a time-series database commonly used in industrial monitoring systems
  • ROS2 publisher, enabling real-time metadata transmission to robotic systems and middleware
• These enhancements improve interoperability with existing industrial data platforms and robotic control systems, facilitating closed-loop automation and monitoring.
• Support MQTT trigger for image processing and inference.
• Configurable OPCUA in REST API.

Retail AI Suite

The Retail AI Suite in version 1.2.0 introduces a set of targeted enhancements designed to support edge-native AI workloads in retail environments. This suite helps solution builders evaluate hardware configurations, optimize AI pipelines, and deploy intelligent retail applications that improve operational efficiency, customer experience, and loss prevention.

Loss Prevention

• The Loss Prevention module provides advanced capabilities to detect and analyze retail fraud and operational anomalies.
• Supported scenarios include:
  • Fake Scans: Identifies instances where items are scanned incorrectly or not at all.
  • Items in Baskets: Detects unscanned items placed directly into baskets.
  • Multi-Product Identification: Recognizes multiple products in a single frame or scan.
  • Product Switching: Detects when a shopper switches one product for another, typically to exploit pricing differences.
  • Shopper Behavior Analysis: Identifies behaviors such as obscuring or hiding items from view.
  • Event Video Summation: Generates summarized video clips of detected events for review and audit.
• This release introduces base workloads for key scenarios including:
  • Product Switching
  • Hidden Items
• Adds Stream Density Estimation, a feature that calculates the maximum number of supported shopping lanes per hardware configuration based on frames-per-second (FPS) thresholds. This helps retailers assess throughput and scalability.

Automated Self-Checkout

• Enhancements to the Automated Self-Checkout module focus on improving reliability and stability.
• Key updates include:
  • Bug Fixes: Addressed known issues affecting pipeline stability and accuracy.
  • Item Age Prediction Pipeline: Introduced a new pipeline that estimates the age of items, which can be used for freshness tracking or age-restricted product validation.
  • YOLO v11 Model Support: Added support for the latest YOLO v11 object detection model, improving detection accuracy and performance across diverse product categories.

Order Accuracy for Quick Service Restaurants (QSR)

• The Order Accuracy sample application is tailored for QSR environments where speed and precision are critical.
• Features include:
  • An object detection pipeline optimized for identifying food items and packaging.
  • Video LLM Integration: Initiated integration with video-based large language models to enhance contextual understanding and accuracy.
  • Demo Exploration: Provides a demonstration setup for evaluating advanced accuracy insights using AI-driven video analytics.

Known Issues and Limitations

• SceneScape: https://github.com/open-edge-platform/scenescape/blob/main/docs/user-guide/release-notes.md#known-issues

Breaking Changes

• None reported for this release.