Skip to content

Release notes - CVEDIA-RT - 2023.4.0

Solutions

SecuRT

  • Emitting starting/stopping events for intrusion and loitering
  • Implemented independent trackers per class (unknown/person/vehicle/animal)
  • Improved tripwires by adding bandwidth settings
  • Using feature tracking to improve movement detection of target classes
  • It now allows switching between movement-based detection and full-frame detection
  • Backtrace of an event in the case of a delayed locking of a track
  • Implemented LOW/NORMAL/HIGH profiles for detection and movement sensitivity
  • Implemented object size filtering on zones and tripwires
  • Improved edge detection filtering to prevent movement false positives
  • Removed ignore stationary object option when implicit

Crowd Estimation

  • Added instance_id in the count export data

AI Models

PVA Detector

  • Improved night-time people detection
  • Improved vehicles and headlights at night

ModelForge

  • Implemented quantization of our models for OpenVino, resulting in a significant performance improvement of 2x to 10x

CVEDIA RT

  • Added support for Hailo on Windows
  • SNPE: Addeded batch size support on classifiers
  • Added information about the AI module and the used backend into Tracy
  • runinstance CLI now uses -b to specify the solution path instead of the parent folder, and it's now mandatory
  • Added Output settings to specify the output preset to use when rendering the output stream
  • Added JSON Exporter in Python as an example of how to consume multiple types of events and data

Bug fixes

  • Switching between instances could crash RT
  • Rendering to the output stream fails when fonts are missing
  • Rendering to the output stream fails when the preset is missing
  • buffer:copyToRegion destination offset doesn't work
  • RTSP path with GET parameters is not accepted
  • The get_instances REST API uses the solution name but should use the solution ID
  • The get_instances REST API doesn’t report the state of a solution correctly
  • The get_state REST API sometimes takes a while to work after starting the instance
  • Saving exporting settings is failing at the first attempt
  • Deleting areas or tripwires while the instance is running causes an error that stops the instance