Face tracking requirements

From Hanson Robotics Wiki
Jump to: navigation, search

This page comprises a list of requirements for tracking human faces in the robot's visual field. See also: perception tasks and attention allocation.

The system should be able to pick out faces (and hands) from video feeds or other sensor data (e.g. kinnect). It should be able to:

  • Determine the distance to the face, and its orientation.
  • Track multiple faces, including thier movement.
  • Work in poor lighting conditions, and/or chaotic environments (crowd scenes).
  • Locate and track faces that are not posed full-frontal facing the camera, but are turned partly away, or looking down. This includes finding/tracking a face of someone lying on the ground, or if the robot itself is lying on the ground (or a gurney) and is looking up.
  • Identify faces from a list of known people.
  • Extract facial features (lip, brow positions).
  • Extract emotional content from facial features (happy, sad,...).

Face tacking architecture diagram

Good tracking requires the following: 3D perception, spatial fusion, and facial recognition.