Release 2026/07/14 (Target)
v3.10.x
Observ July Updates
This release expands Observ's scenario coverage with traffic flow and directional event logic, strengthens VLM detection precision, and improves platform traceability and API flexibility, making Observ more powerful for real-world traffic and safety monitoring and more automation-ready for large-scale deployments.
New Features:
🚦 New Scenario Events: Traffic Flow & Directional Logic
We've added a new family of traffic-oriented events to support smart city and roadside monitoring scenarios.
Flow Threshold Alerts
Vehicle Flow Threshold Exceeded
Pedestrian Flow Threshold Exceeded
Sequence-Based Events
Vehicle Wrong-Way Driving
Vehicle Illegal U-Turn
All of the above can be flexibly combined within Custom Events. In particular, temporal events let you capture direction-aware behavior by chaining zone-entry sequences, ideal for distinguishing intent (e.g., wrong-way vs. normal flow).
📋 Activity Log
A new Activity Log provides baseline operational traceability by recording key user and task actions:
Task operations — create, edit (pause, resume, terminate), delete
User operations — create, edit, remove
🤖 VLM Enhancements
Two upgrades make VLM detection more precise and more efficient:
Image Resize Before Inference: Images are now resized before being sent to the VLM. Higher input resolution improves detail recognition but increases inference time and token usage.
BBox / Zone Drawing for VLM: In the BBox VLM Flow Entry Node, when selecting a bbox you can now choose to draw a warning bbox / zone for the VLM. This lets the VLM focus its reasoning on specific regions, improving targeted analysis.
🔗 Expanded API Support
We've broadened API coverage for richer integration and more scalable, automation-friendly workflows:
Event Type now includes
token+AliasEvent Record WebSocket now pushes event status-change updates
Batch control of event detection on/off and VLM frequency
Event Zone adjustment via API
⚡ Streaming Stability
We've upgraded the GPU streaming service to improve the stability of the GPU stream-decode service for more reliable, continuous video processing.
🧩 General Model Update
We've expanded the object detection model's training coverage and added privacy-focused capabilities:
New training data for weapon and road pothole detection
New vehicle marking class to support license-plate / identifier anonymization for stronger privacy protection
Coming Soon:
We're continuing to push toward a more modular, flexible, and agent-ready platform:
More batch-modification APIs to fully modularize the environment, enabling greater flexibility and agentic application workflows.
Higher VLM accuracy through multi-model selection and multi-image / video inference.
Transformer-based object detection model, targeting accuracy improvement over the current model.
New scenario: instantaneous flow-change detection, surfacing sudden traffic surge/drop signals in real time.
Support
For technical support, please contact our support team during business hours.
Last updated