File support for offline processing, Simplified BYO model experience, Pre-buffer for Clips, Homekit Streaming node, Docker containers for Jetson devices and Unattended install mode
Hot off the press, here are the highlights from the latest release:
Platform
🍀 File support : Upload files to Lumeo or link a self-hosted file to process it using any gateway. See Files section in left nav to start. See Stream page for details.
🍀 Simplified BYO Model experience: A more structured way to import your models into Lumeo. See Analytics Library -> My AI Models to start.
🍀 Export pipelines to JSON and import them into a new workspace.
🐛 Fix Bulk updating deployments causes ROI coordinates to be lost
Nodes
🍀 Save Clip Node Pre-buffer support. Save clips starting from a time before a trigger occurs.
🍀 Homekit Streaming node : Publish Lumeo output streams as Apple Homekit Cameras for quick viewing and trigger alerts based on any analytic condition.
🐛 Fix flickering when viewing square aspect ratio video snapshots in Line and ROI nodes.
Gateway (v 0.4.11)
(All connected Gateways will auto update and restart any running deployments)
🍀 Docker container support for Jetson devices, defaults to docker for new installs. See Setting Up a Gateway section for details.
🍀 Unattended install support. Specify LUMEO_APP_ID and LUMEO_API_KEY environment variables to the installer to automatically provision a gateway. See Unattended Install section for details.
🐛 Fix so that linked ONVIF cameras' streams update automatically. Fixes an issue with new streams not showing up automatically in the Milestone ONVIF Bridge.
🐛 Fix inferencing slowdowns for the full YOLO models. Typically yields a 2-3x speedup for YOLO models.
For more details, check out the Release notes.
Reorganized node navigation, Bug fixes and reliability updates
Hot off the press, here are the goodies from the latest release:
Nodes
🍀 Reorganized Node categories in the Node library to make it easier to find the one you need.
🍀 Proximity Detection node lets you choose the tracking point other than the centroid
🐛 Fix duplicate triggers and counts that did not respect the Occupancy Monitor node's object presence buffer
Lumeo Gateway (v 0.3.6)
(All connected Gateways will auto update and restart any running deployments)
🍀 Restart deployments automatically in more situations where the source is unresponsive of stops streaming
🐛 Fix inability to capture camera snapshots for some camera models
Multi-stream and Local License Plate Recognition pipelines, Send snapshots to Elasticsearch, Autoconfigure Milestone Camera IDs, and new nodes for Profiling Performance and Filter Objects
Core Platform
🔥 Pipelines
New template for Multi-stream pipelines that let you process more streams with a given GPU
New template for Local License Plate Recognition
Performance, Stability & Security
30% runtime performance improvement for pipelines with a large number of objects
Fixed processing slow down when ingesting variable frame rate streams
Console
Support double-click in multiline and polygon to finish or close a shape
Auto generate default URI stream name from URI/file + allow editing in creation form
Models
🔥 Models : Local License Plate Recognition so you can read license plates on the edge without making a trip to the cloud!
Pipeline Nodes
Rules
Line Counter NodeOccupancy Monitor - select alternate tracking points to improve performance when Centroid tracking doesn't work well enough.
Trigger Milestone Event Node - Automatically configures camera ID when you use Milestone ONVIF / Open Network Bridge as the source, making it even easier to associate alerts and events with the correct camera in Milestone.
Function Node - add, modify and delete object metadata using the Function node, for advanced use cases such as filtering out objects or creating new ones.
Docker container support for DGPU installs, New Models: Forklift, Face gear, Crowd People, New Integrations: Elasticsearch, SMS, Pushover, Azure Face Recognition and LPR enhancements.
Lumeo Core Platform
Pipelines
New, ready-to-use templates for more use cases
Performance, Stability & Security
Docker container support for DGPU installations, for ease of setup and deployment management
Installer enhancements to reduce memory utilization, free up disk space and support alternate disk installation for Jetson devices
Significant runtime performance and stability improvements in light of network issues.
Billing
View your invoices and update payment methods from the Lumeo Console.
Models
New Models : Forklift detection, Face gear, Crowd People Body detection (high density)
Jetpack 4.6 updates, New VMS Integration Nodes, New Rules: Proximity, Speed, Multi-line, and New Video transformations - Tile video, Multiplex/Demultiplex Nodes. Platform updates to enable easier ROI drawing, and Trigger Conditions for conditional actions.
Lumeo Core Platform
Pipelines
ROI Drawing Tools : Draw Lines and ROIs on video stream previews.
Trigger Metadata support: Most Lumeo Nodes now generate metadata that can be used in "Trigger Conditions" within other nodes to generate conditional actions (like take a snapshot when an object crosses a line).
Trigger Conditions : New Trigger Condition builder in Node properties makes it easier to build trigger conditions with drop down and autocompletion.
Performance, Stability & Security
Upgrade to Nvidia Jetpack 4.6 / Deepstream 6.0 (more models support with high accuracy & real-time performance).
This requires you to update any Jetson devices to Jetpack 4.6 to receive Lumeo software updates moving forward. Instructions here.
Auth updates : We've migrated to a new authentication system for console.lumeo.com. Use the "Forgot Password" link on https://console.lumeo.com to reset your password if you have trouble logging in.
Lumeo Pipeline Nodes
Integrations (read more here)
Send events and alarms to existing VMS installations.
Metadata format for the LPR, Line Counter, Presence Counter and Queue nodes has been updated to reduce duplicate information, and introduce a uniform structure with new properties that can be used to trigger snapshots, video clip capture, webhooks, etc.
Summary
Metadata format for the following nodes has been updated to reduce duplicate information, and introduce a uniform structure with new properties that can be used to trigger snapshots, video clip capture, webhooks, etc. :
These updates take effect when you create new deployments or update existing ones. Starting and stopping existing deployments will not be impacted.
If you are expecting the old(er) format metadata, you can turn on legacy metadata option in the node property to keep the older format while you transition to the new format.
What's changed
Metadata format updates
With this update, these nodes insert their metadata inside the root field nodes, while previously those nodes used top-level fields like lpr, lines, presence, queue. The contents of those fields have now been moved into nodes.<node_id> as per below:
lpr -> nodes.<node_id>.license_plates
lines -> nodes.<node_id>.lines
presence -> nodes.<node_id>.rois
queue -> nodes.<node_id>.rois
To provide for backward compatibility during this transition, the Line Counter Node and License Plate Detection Node have the option to turn ON the legacy metadata format in node properties. Enabling that option will revert the metadata for those deployments to the older format.
New metadata properties
These nodes now add new metadata to make it easier to trigger webhooks, snapshots, etc. based on the difference in counts between previous and current frame. This makes it easier to trigger when a new car is detected, or when a new object crosses a line, etc.
For specifics of the new format, see node documentation.
What you need to do
If you have code that parses Lumeo metadata, either in a Custom Function, or sent to you via webhooks, or from a saved clip, you will need to update it to account for the format changes.
If you are using the Publish to Google Sheets node, you will need to update the Metadata to Publish property to look for metadata in the new format. Here's an example of the new property value:
Spring is (almost) here! And with it, we bring you new nodes, bug fixes and (pretty awesome, if we may say so ourselves) performance improvements. Our focus for the upcoming release is to address issues we are seeing with clip storage, adding trigger conditions for clips & snapshots to make working with streaming metadata easier, and improve Console's responsiveness. Ah, and a new feature that we are sure you'll love.
What's new
New nodes :
-- Presence Detector Node : Count and track how long objects have been present in a specific region of interest.
-- Line Counter Node : Count the number of objects that have crossed the line in one direction or the other.
-- Webhook Node : Send metadata to a webhook.
New models : Face detection (for infrared cameras), Vehicle Detection (for Dash cam perspective)
Deployment management UX has a new look, with a split pane view that leads to less clicking and easier access to the deployments you are looking for.
x86 + DGPU support : You can now run the lumeo gateway installer on x86 linux (Ubuntu 18.04) with discrete Nvidia GPUs. For those times when Arm just won't do :). This means you can install Lumeo in AWS, Azure, or Google Cloud GPU-based VMs, or on local linux boxes.
Fixes
Major under the hood revisions to our MQTT infrastructure, making connections from lumeod to our cloud much more reliable in the face of local network issues.
Starting, stopping and updating Deployments is much more reliable now. Deployments auto start when created. When gateway auto updates or restarts, previously running pipelines will now auto start. When you update a running deployment, it will automatically restart to pick up new parameters - all the stuff you'd not want to think twice about.
Lumeo can discover and link ONVIF enabled NVRs. We also fixed a pesky issue that resulted in some NVRs locking out the Lumeo host IP completely.
Camera status is now accurately reflected in the Console.
Pipelines with WebRTC and RTSP output streams now work as expected.
Happy 2021! We've been heads down over the last few weeks, working on a ton of goodies for you. Now that we have this core release out, expect to see faster paced updates, as we light up a lot more nodes out of the box for analytics and integrations with external systems in the coming weeks.
Revamped UX for Streams making it easier to find and manage streams: Streams are now split into input & output streams. Input streams are the ones you import (RTSP streams, Media files hosted on a web server, etc.) and Output streams are ones generated by Lumeo pipelines. Note that streams from a Camera now only appear under the Cameras section. However the Video source selector allows you to select any and all input streams easily.
Revamped metadata format, making it easier to access metadata in the function node. We are deprecating the old method used to access object detection metadata, so you will need to update your function nodes. See Function Node for details on how to access, and each node reference for details about the metadata it adds to the pipeline.
Fixes
Deployments now transition from Stopped to Deploying to Running when you Start them. This resolves a common issue where you would try to play a stream before the deployment was actually running, and hence it would just not play.
RTSP URLs & Cameras with special characters in the credentials now work as expected.
Improved thresholds on Person detection models to reduce false positives.
Merry X'mas and happy holidays! This is a lightweight update as the team takes a bit of a break at the end of the year - we will be back in January with more updates. We are working on a number of improvements in how the pipelines work, the user experience around managing streams & deploying pipelines and a bunch of new nodes. Meanwhile, please feel free to let us know if there is anything we can help with or issues that are blocking you. Cheers!
What's new
Track Objects Node in the pipeline editor. Use it to track and assign unique id's to any detected objects as they move about in the view. This is particularly useful for determining the velocity of the object, or figuring out if the object has been in the view for more than X amount of time (loitering). See Track Objects Node for reference, and Pipeline Examples for an example pipeline that uses this to detect loitering.
Fixes
The People model in the marketplace is now functional.
