Change Log

All notable changes to the Kairos Human Analytics SDK will be documented in this file.

The format is based on Keep a Changelog and this project adheres to Semantic Versioning.

[1.1.0] - 11/15/2016

Added

  • Improved image calibration for feature point alignment in images
  • Calibration type parameter (that's available) when passing a custom input frame to the processor
  • Improved EmotionResponse for Fear and Disgust emotion
  • Default parameter to the HumanAnalysisService that can configure the processor to process up to MAX faces concurrently
  • CaptureError is now thrown when the camera goes offline or is otherwise interrupted while in the middle of processing frames
  • Redirecting OpenCV range assertion output that comes from the HOG extractor during featurepoint detection. Serves no purpose for a thirdparty developer - used internally for making improvements and thus a dev never needs to do anything with this assertion
  • Updated the existing demo with a simple UI to provide real-time visual of SDK output
  • Additional DEMO to showcase how to process a cameral offline scenario
  • Minor speed improvements
  • General enhancements

Fixed

  • The ghosting issue that caused more faces, than are actually in frame, to be returned from the face detector

[1.0.2] - 10/04/2016

Added

  • Up to 20 faces can be tracked simultaneously
  • Can now a pass a frame directly into the processor, that was acquired from an individual video capture, image or pre-recorded video acquisition
  • Other performance enhancements

Fixed

  • When face left the scene, invalid featurepoints still being tracked
  • OpenCV debug dlls (Windows version only) missing from the archive

[1.0.1] - 09/27/2016

Added

  • Support for multi-face tracking
  • Methods for width and height of the media source (live video, video file or image dimensions for pictures)
  • Now adding new people to a frame in realtime, as they pass in front of the camera
  • Created a simple tutorial app to show basic usage of the SDK, with source for a developer to compile (compiles and runs in Linux and Windows)

Fixed

  • Emotion response scores were too low because it wasn't normalizing the response to produce accurate result
  • Lookup of .dat file in multiple locations, causing issues during installation and setting up of development environment, now only search for in the models folder
  • Calculation of dwell now based on frame rate for video instead of start/end time based on system clock
  • Minor tweaks to make glances more accurate
  • Several performance improvements and overall enhancements
  • Improved exception handling for corrupt licenses