Logo for AiToolGo

Hailo-Apps v25.10.0 and HailoRT 4.23: Raspberry Pi Trixie Support, Multisource Video, and Tiling Detection

In-depth discussion
Technical
 0
 0
 5
The content comprises two posts from the Hailo Developer Zone. The first announces HailoRT 4.23 and Hailo-Apps-Infra v25.10.0, detailing installation steps for Raspberry Pi (Trixie) and x86 Ubuntu, new apps (Tiling, Multisource, REID Multisource), infrastructure improvements, bug fixes, and repository changes. The second is a technical guide to face recognition covering the pipeline stages: detection (SCRFD, RetinaFace, YOLOv8), alignment via landmarks, ArcFace embedding extraction, and LanceDB-based database matching, with code samples using the DeGirum PySDK.
  • main points
  • unique insights
  • practical applications
  • key topics
  • key insights
  • learning outcomes

  • • main points

    • 1
      The face recognition guide walks through each pipeline stage with runnable code, including detection, keypoint-based alignment, embedding extraction, and vector database matching
    • 2
      The release post provides concrete, copy-pastable installation commands for both Raspberry Pi and x86 Ubuntu, along with explicit compatibility notes (Trixie vs. Bookworm)
    • 3
      Clear explanation of why alignment matters for embedding consistency, with a reference-keypoint transformation implementation

  • • unique insights

    • 1
      Explains that alignment normalizes pose, orientation, and scale so the embedding model focuses on consistent facial features, improving reliability in surveillance and multi-angle scenarios
    • 2
      Describes database design for face recognition where multiple embeddings per identity improve robustness, enforced through a shared entity_name field

  • • practical applications

    • Provides developers on Raspberry Pi and Hailo-8/8L hardware with an updated runtime stack, ready-to-use multi-source and re-identification apps, and a reproducible end-to-end face recognition implementation that can be adapted to access control, surveillance, or analytics projects.

  • • key topics

    • 1
      HailoRT 4.23 and Hailo-Apps-Infra v25.10.0 installation and release features
    • 2
      Face recognition pipeline: detection, alignment, embedding extraction, and database matching
    • 3
      Multi-source video processing and cross-camera person re-identification

  • • key insights

    • 1
      Combines a fresh hardware runtime release with a full, code-backed face recognition pipeline tailored for Hailo edge accelerators
    • 2
      Covers both a practical deployment changelog and an educational end-to-end guide in one resource
    • 3
      Offers a choice of detection models (SCRFD variants, YOLOv8-face, RetinaFace) with explicit trade-off guidance between accuracy and speed

  • • learning outcomes

    • 1
      Set up HailoRT 4.23 and Hailo-Apps-Infra on Raspberry Pi (Trixie) or x86 Ubuntu and run the new tiling, multisource, and REID applications
    • 2
      Implement a face recognition pipeline including detection with keypoints, landmark-based alignment, ArcFace embedding extraction, and similarity-based matching
    • 3
      Design a vector database schema for storing face embeddings with multiple entries per identity using LanceDB
examples
tutorials
code samples
visuals
fundamentals
advanced content
practical tips
best practices

“ Introduction: Hailo-Apps v25.10.0 and HailoRT 4.23

Hailo has released HailoRT 4.23 for Raspberry Pi and Ubuntu, alongside a major new version of the Hailo Applications Infrastructure. This update, long anticipated by the developer community, delivers a refreshed runtime, new AI applications, and a range of stability improvements that make building edge vision solutions with Hailo hardware more practical than ever.

“ What's New in HailoRT 4.23

The entire Hailo software stack has been updated to version 4.23, covering drivers, runtime, tools, and supporting components. Keeping every layer on the same release reduces version mismatches and simplifies troubleshooting, giving developers a consistent foundation for their inference workloads on Hailo-8 and Hailo-8L accelerators.

“ Raspberry Pi Trixie Support and Compatibility

This release targets Debian Trixie and does not support Bookworm. Users who need to continue working on Bookworm should remain on the previous release. Checking your operating system version before upgrading is an important first step to avoid installation issues.

“ How to Install or Update on Raspberry Pi

Raspberry Pi users can install the complete stack through the unified hailo-all installer. First, update the system and install the required packages: run sudo apt update, sudo apt full-upgrade, and sudo apt install dkms (dkms is essential, otherwise the driver will not install correctly), then run sudo apt install hailo-all. Next, clone the hailo-apps-infra repository, navigate into the directory, run sudo ./scripts/cleanup_installation.sh to remove previous installations, and finish with sudo ./install.sh.

“ How to Install on x86 Ubuntu

On x86 Ubuntu systems, begin by installing the Hailo packages from the Hailo Developer Zone. Then clone the hailo-apps-infra repository, run the cleanup script to remove any prior setup, and execute the installer script with sudo ./install.sh to complete the configuration.

“ New Applications in Apps Infrastructure v25.10.0

The Tiling app (beta) detects small objects in high-resolution images by intelligently splitting frames, making it well suited to aerial footage; it includes a VisDrone MobileNetSSD model and runs with hailo-tiling. The Multisource app processes multiple video inputs simultaneously, including cameras, files, RTSP streams, and USB devices, with up to three sources on Raspberry Pi and per-source callbacks; launch it with hailo-multisource. The REID Multisource app (beta) tracks people across several cameras using face embeddings and re-identification, stores identities in LanceDB, and is optimized for around 15 FPS; try it with hailo-reid.

“ Infrastructure and Stability Improvements

The installation flow has been completely revamped, offering clearer error messages, better hardware detection, and easier upgrades. New and improved C++ post-processing functions enhance performance, while the face-recognition backend has been simplified and its user interface removed. GStreamer pipeline stability and video looping issues have been fixed, memory and buffer management has been strengthened, logging has been enhanced, and resource downloading and version handling are now smarter.

“ Bug Fixes

This release resolves compatibility issues on Hailo-8L devices, fixes a pipeline looping bug, corrects version and device detection, and addresses installation permission problems. Several additional stability fixes round out the update.

“ Repository Changes and Community Direction

hailo-apps-infra is now the primary development repository and will soon be renamed Hailo-apps. The hailo-rpi5-examples repository will shift its focus toward community projects and will open to community contributions, inviting developers to share and build upon each other's work.

“ Conclusion and Next Steps

HailoRT 4.23 and Hailo-Apps Infrastructure v25.10.0 represent one of the most significant updates to date, combining a unified software stack, new multi-source and detection applications, and substantial stability gains. Developers are encouraged to review the full release notes, report any issues, and start building with these tools on Raspberry Pi and Ubuntu.

 Original link: https://community.hailo.ai/t/a-comprehensive-guide-to-building-a-face-recognition-system/8803

Comment(0)

user's avatar

      Similar Learning

      Related Tools