Skip to main content
Ctrl+K

Open Edge Platform

    • Open Edge Platform Overview
    • Metro
    • Manufacturing
    • Retail
    • Robotics
    • Education
    • Health and Life Sciences
    • Federal and Aerospace
    • Libraries, Tools, Services
    • DL Streamer
    • Scenescape
    • ViPPET
    • Geti
    • OpenVINO
    • OpenVINO Physical AI
    • Platform Software
  • Open Edge Platform Overview
  • Metro
  • Manufacturing
  • Retail
  • Robotics
  • Education
  • Health and Life Sciences
  • Federal and Aerospace
  • Libraries, Tools, Services
  • DL Streamer
  • Scenescape
  • ViPPET
  • Geti
  • OpenVINO
  • OpenVINO Physical AI
  • Platform Software

Section Navigation

  • OEP SDK Manager
  • OEP CLI Installer
  • Metro Analytics Catalog
  • Metro Application Builder

Samples

  • Agentic Predictive Maintenance for Critical Infrastructure
    • Get Started
      • System Requirements
    • How It Works
    • Build from Source
    • Training a Defect Detection Model with Intel Geti
    • API Reference
    • Troubleshooting
    • Release Notes
  • Agentic Smart Community
    • Get Started
      • System Requirements
      • Ready-to-Run Demo
    • How It Works
    • How-To Guides
      • Docker Setup Script Reference
      • Add Swap Space
      • Register a New Use Case
      • Configure the Prefilter Model
      • MCP Tools Guide
    • API Reference
      • MCP Tools and Resources API
      • MCP Subscription
      • MCP Webhook Event API
      • Dashboard API
      • Videostream Analytics HTTP API
    • Release Notes
  • Enterprise Data Intelligence - Multi-Agent Enterprise Knowledge Assistant
    • Get Started
      • System Requirements
    • Release Notes
  • Image-based Video Search
    • Get Started
      • System Requirements
    • How It Works
    • How to use GPU for inference
    • How to use NPU for inference
    • Troubleshooting
    • Release Notes
  • Live Video Alert Agent
    • Get Started
      • System Requirements
      • Build from Source
      • Deploy with Helm
    • How It Works
    • API Reference
    • Known Issues
    • Release Notes
  • Live Video Captioning
    • Get Started
      • System Requirements
      • Model Preparation
      • Build from Source
      • Deploy with Helm Chart
      • Setup Simulated RTSP Stream
      • Testing and Coverage
    • Quick Start: Live Video Captioning
    • How-to Guides
      • Enable Alert Mode
      • Configure Object Detection Pipeline
    • How it Works
    • API Reference
    • Known Issues
    • Release Notes
  • Live Video Captioning RAG
    • Quick Start Guide
    • Get Started
      • System Requirements
      • Build from Source
      • Deploy with Helm Chart
      • Testing and Coverage
    • How It Works
    • API Reference
    • Known Issues
    • Release Notes
  • Live Video Search
    • Get Started
      • System Requirements
      • Build from Source
      • Deploy with Helm* Chart
    • How It Works
    • API Reference
    • Release Notes
  • Real-Time Scene Understanding
    • Get Started
      • System Requirements
      • Model Preparation
      • Build from Source
      • Testing and Coverage
    • Quick Start
    • How It Works
    • API Reference
    • Troubleshooting
    • Release Notes
  • Smart Building Digital Twin
    • Get Started
      • System Requirements
    • How It Works
    • How to Use the Application
    • Troubleshooting
    • Release Notes
  • Smart Corridor
    • Get Started
      • System Requirements
    • Troubleshooting
    • Release Notes
  • Smart Route Planning Agent
    • Get Started
      • System Requirements
      • Build from Source
      • Environment Variables
      • Deploy with Helm
    • Release Notes
  • Smart Traffic Intersection Agent
    • Get Started
      • System Requirements
      • Build from Source
      • Deploy with Helm
      • Deploy with Trusted Compute using Helm
      • Deploy with Trusted Compute
    • API Reference
    • Release Notes
  • Loitering Detection
    • Get Started
      • System Requirements
    • How-to Guides
      • Use the Console UI
      • Customize the Application
      • Updating the Pipeline
      • Generate and Deploy Offline Package
      • Use GPU for Inference
      • Use NPU for Inference
      • View Open Telemetry Data
      • View System Performance Dashboard
      • Benchmark Performance
      • GPU and NPU Stream Density Benchmark
    • Troubleshooting
    • Release Notes
  • Smart Intersection
    • Get Started
      • System Requirements
      • Deploy with Trusted Compute
    • How It Works
    • How to use GPU for inference
    • How to use NPU for inference
    • RTSP Stream Setup
    • Export and Optimize Geti Model
    • Troubleshooting
    • Release Notes
  • Smart Parking
    • Get Started
      • System Requirements
      • Get Started with Smart Parking using SceneScape
    • How-to Guides
      • Customize the Application
      • Generate and Deploy Offline Package
      • Use GPU for Inference
      • Use NPU for Inference
      • View Open Telemetry Data
      • View System Performance Dashboard
      • Benchmark Performance
      • Demonstrating NPU Value: GPU and NPU Stream Density Benchmark
    • Troubleshooting
    • Release Notes
  • Smart Tolling
    • Get Started
      • System Requirements
      • Bring Your Own Data
    • How It Works
      • The Perception Layer
      • Optimization
      • Analytics Pipeline
    • Troubleshooting
    • Release Notes
  • Sensor Fusion for Traffic Management
    • Post-Fusion
      • Get Started
        • Prerequisites
        • System Requirements
      • How it Works
      • Advanced user guide
      • APIs
    • Intermediate Fusion
      • Get Started
      • Prerequisites
      • Testing
      • Training
        • Installation
        • Convert NVIDIA Checkpoint
    • Troubleshooting
    • Release Notes
      • Release Notes 2025
  • Smart NVR
    • Get Started
      • System Requirements
      • Build from Source
      • Deploy with Helm
    • How to Use Smart NVR
    • Development Guide
    • Integrate Scenescape with Smart NVR
    • Multiple Scenescape Deployment
    • API Reference
    • Troubleshooting
    • Release Notes
      • Release Notes 2025
  • Video Processing for NVR
    • Get Started
    • How It Works
    • Release Notes
  • Virtual Security Guard Agent
    • Get Started
      • System Requirements
      • Set Up the Prerequisite Stack
      • Build from Source
      • Run Unit Tests
    • How-to Guides
      • Investigate a Person
      • Configure the Floorplan
      • Manage Live Cameras
    • How It Works
    • Known Issues
    • Release Notes
  • Visual Search and Q&A
    • Get Started Guide
      • System Requirements
      • Deploy with Helm
    • Tutorials
    • Troubleshooting
    • Release Notes
  • VMS Adapter Plugin
    • Get Started
      • System Requirements
      • Build from Source
      • Deploy with Helm
    • How It Works
    • How-To Guides
      • Integrations
        • Nx Witness
          • Set Up Nx Witness
          • Run Loitering Detection
          • Run Live Video Captioning
          • Run Both Applications
          • Reference and Troubleshooting
        • Milestone XProtect
          • Set Up XProtect and AI Bridge
          • Run Loitering Detection
          • Events and Alarms
          • Reference and Troubleshooting
      • TLS and Certificates
    • API Reference
    • Troubleshooting
    • Release Notes
  • Video Search and Summarization

Software Development

  • OpenVINO™ Toolkit
  • OpenVINO™ Model Server
  • DL Streamer
  • DL Streamer Pipeline Server

Tools

  • Geti™ Software
  • Visual Pipeline and Platform Evaluation Tool
  • Scenescape

Guides

  • Hardware Partner Workflow
  • Developer Workflow
  • Migrate from NVIDIA
  • Achieve High Performance <
  • Improve Security
  • Customize Visual AI Kit
  • Choose the Right Components

Blueprints

  • Blueprint: Critical Infrastructure Predictive Maintenance
    • Vision AI Pipeline
    • Multimodal Pipeline
  • Blueprint: Smart Buildings
  • Blueprint: Edge Manageability and Resiliency

---------------

  • Intel® Edge System Qualification
  • Get Help or Contribute
  • Metro AI Suite
  • Live Video Search
  • Release Notes: Live Video Search

Release Notes: Live Video Search#

Version 2026.3.0-rc1#

Release Date: November 6, 2026

New:

  • Added the --pull-always option to setup.sh (PULL_ALWAYS=true for make deploy) to refresh cached images from the registry before starting the stack.

  • Multimodal DataPrep object detection now uses Hugging Face OpenVINO/* detection models instead of the bundled YOLOX download. Select the model with DATAPREP_DETECTION_MODEL (Compose, setup.sh, Makefile) or global.env.detectionModelName (Helm). The model is downloaded on first start and cached in the dataprep-detection-models volume (Helm: detectionmodels sub-path).

  • Added DATAPREP_DETECTION_CLASSES (Helm: global.env.detectionClasses) to keep only selected classes from the detection model’s label set.

  • Exposed the DataPrep pipeline stall watchdog through MM_DATAPREP_PIPELINE_STALL_TIMEOUT_S (Helm: global.env.pipelineStallTimeoutS, default 300, 0 disables).

Improved:

  • Aligned LVS with the Video Search and Summarization 2026.3.0-rc1 stack (Multimodal DataPrep, Vector Retriever, and Multimodal Embedding Serving), including the updated GPU drivers in the DataPrep and Multimodal Embedding Serving images.

  • setup.sh, make validate-env, and the Helm chart now fail early with a clear message when object detection is enabled and no detection model is set.

  • Updated the documentation links to the new Metro AI Suite repository.

Upgrade Notes:

  • Breaking: DATAPREP_DETECTION_MODEL (Helm: global.env.detectionModelName) is required while object detection is enabled. There is no default model. For example, set OpenVINO/yolox_s-int8-ov, or set ENABLE_OBJECT_DETECTION=false (Helm: global.env.enableObjectDetection: 'false') to embed whole frames.

  • Each detection model carries its own license (for example, OpenVINO/yolox_* is Apache-2.0 and OpenVINO/YOLO11* is AGPL-3.0). Review the model card before use.

  • The legacy docker_dataprep-yolox-models volume is no longer used. source setup.sh --clean-data removes it.

Version 2026.2.0#

Release Date: September 10, 2026

New:

  • Replaced the legacy vdms-dataprep orchestration with backend-neutral multimodal-dataprep in Docker Compose and Helm.

  • Added an always-on Vector Retriever layer so Video Search no longer accesses a vector database directly.

  • Added selectable VDMS (default) and Milvus backends through VECTORDB_BACKEND for Compose and global.vectordbBackend plus milvus_override.yaml for Helm.

  • Added pinned standalone Milvus/etcd orchestration and updated build, architecture, device, and deployment guidance for both retriever flavors.

  • Replaced the collector/Pipeline Manager WebSocket telemetry path with Metrics Manager for both Docker Compose and Helm, gated by ENABLE_METRICS_MANAGER (default true) in setup.sh and the Makefile. Multimodal DataPrep publishes throughput metrics directly and the UI consumes the same-origin SSE stream through NGINX.

  • Added NPU-capable device orchestration for the VSS search stack used by LVS in Docker Compose setup.

  • Added global.accelGroupIds so the host gids owning /dev/dri (GPU) and /dev/accel (NPU) are injected into the pod supplementalGroups, letting the non-root container open the accelerator device.

  • Added a persistent OpenVINO cache (ovCacheDir, default /app/ov_models/ov_cache) for MME and DataPrep so GPU/NPU model compilation is reused across pod restarts.

  • Exposed asynchronous watcher-batch size, polling interval, and timeout settings for Search MS and Smart NVR continuous ingestion through Compose, Helm, and setup.sh.

  • Added a single, case-insensitive Helm global.pullPolicy override for all application images selected through the LVS, VSS, and Smart NVR stack tags.

Improved:

  • Updated LVS compose deployment to a pure per-component device model (DATAPREP_EMBEDDING_DEVICE, DATAPREP_DETECTION_DEVICE, MME_EMBEDDING_DEVICE; each defaults to CPU) and mounted /dev/accel for NPU execution. Retired the redundant VDMS_DATAPREP_DEVICE baseline.

  • Updated LVS Helm deployment templates and values to a pure per-component device model via global.devices.multimodalEmbedding.* and global.devices.multimodalDataprep.{embedding,detection}.* (each defaults to CPU), retiring the legacy global.gpu.* block to remove device-configuration ambiguity.

  • Removed the ambiguous ENABLE_EMBEDDING_GPU shortcut; indexing and query embedding devices are configured independently with DATAPREP_EMBEDDING_DEVICE and MME_EMBEDDING_DEVICE.

  • Renamed the Compose/setup model input from EMBEDDING_MODEL_NAME to MULTIMODAL_EMBEDDING_MODEL, the DataPrep variables from VDMS_DATAPREP_* to MM_DATAPREP_*, and VDMS_PIPELINE_MANAGER_UPLOAD to VIDEO_UPLOAD_ENDPOINT.

  • Vector Retriever now waits for a healthy Multimodal Embedding service before starting, using a configurable init check (embeddingService.waitForHealthy, healthUrl, healthTimeoutSeconds, retryIntervalSeconds).

  • Multimodal DataPrep waits for Metrics Manager readiness when metrics are enabled, and carries component labels for scheduling.

  • Made Search MS batch-status polling resilient through SEARCH_DATAPREP_POLL_MAX_RETRIES, SEARCH_DATAPREP_POLL_TIMEOUT_MS, and SEARCH_DATAPREP_POLL_RETRY_DELAY_MS.

  • Improved Milvus readiness handling in the Helm deployment.

  • Removed the obsolete vss-collector subchart, telemetry compose overlay, Telegraf configuration, shared signal PVC, and the pod co-location constraint they required.

  • Updated LVS documentation (get-started, deploy-with-helm, how-it-works) with NPU usage guidance, accelerator configuration examples, and Metrics Manager deployment notes.

Fixed:

  • Fixed Docker Compose backend selection so Milvus deployments do not start or depend on the VDMS service, and stale backend containers are removed when switching backends.

  • Corrected the Helm multimodal DataPrep completion-queue default to satisfy the service’s minimum queue size and prevent pod startup validation failures.

  • Aligned the Helm Multimodal Embedding Serving probe timeout with its Compose healthcheck to avoid one-second startup probe timeouts during model loading.

  • Fixed configuration rollouts so a Helm upgrade that changes only ConfigMap values restarts the affected pods; environment variables injected through envFrom previously stayed stale until a manual kubectl rollout restart.

  • Fixed Milvus data persistence so global.keepPvc=true also retains the etcd metadata volume; etcd previously used an ephemeral emptyDir, which orphaned the retained segment data on pod rescheduling or reinstall.

  • Fixed the nvr-event-router container healthcheck, which now bypasses proxy settings instead of relying on curl.

  • Miscellaneous documentation corrections.

Version 2026.1.0#

Release Date: June 17, 2026

New:

  • Deployment with Helm chart.

Known Issues:

  • First‑time model downloads may take several minutes.

  • Time‑range queries require the clock and timezone on the host to be accurate.

Version 1.0.0#

Release Date: April 1, 2026

Live Video Search is a new sample application which implements embedding and visual data ingestion microservices (available in Edge AI Libraries) for processing RTSP camera streams and user query-based search. The application converts the input camera data to embeddings continuously, using models like Clip. The embeddings are stored in a Vector Database (VectorDB ) and enable search on live camera feed and historical video data. A rich UI is provided to configure the camera used for data ingestion, enter the search query, and view telemetry data, currently, for CPU, GPU, and memory utilization. The sample application introduces camera streaming with Frigate.

New:

  • Live Video Search stack integrating Smart NVR with VSS Search.

  • Time‑range filtering in search via UI or natural‑language query parsing.

  • Telemetry visualization in VSS UI for live system performance.

Known Issues/Limitations:

  • Deploy with Helm is not yet supported for Live Video Search.

  • First‑time model downloads may take several minutes.

  • Time‑range queries require the clock and timezone on the host to be accurate.

The application has been validated on Intel® Xeon® 5 + Intel® Arc™ B580 GPU.

On this page
  • Version 2026.3.0-rc1
  • Version 2026.2.0
  • Version 2026.1.0
  • Version 1.0.0
Show Source
Enable cookies to use AI chat
Chat is locked. Click the blue chat button, then enable Functional cookies to unlock it.