ALFaceCharacteristics

NAOqi PeoplePerception - Overview | API


What it does

ALFaceCharacteristics updates every person with some additional information such as an estimation of age and gender. It also tries and detects whether the person is smiling or what facial expression the person is having.

How it works

Like most perception modules this one works on top of ALPeoplePerception which will be started automatically by ALFaceCharacteristics.

This module analyzes the face of a person and try to find some specific characteristics to a gender or an age category. As for the smile, it gives a measurement of the “smile degree” on a face, which means it can make the difference between a faint smile and a frank smile. An event called FaceCharacteristics/PersonSmiling() is raised every time the smile degree is higher than a given threshold. The value of this threshold can be changed by calling ALFaceCharacteristicsProxy::setSmilingThreshold().

ALFaceCharacteristics also analyses facial expressions and tries to estimate whether a face is neutral, happy, surprised, angry or sad.

In addition to the age, gender, smile and facial expression estimations, the module also provides a confidence score between 0 and 1 associated to each value to indicate how likely an estimation is. 0 means that the estimation is “very unlikely” and 1 means “very likely”). It makes it possible to filter the results in order to keep the most reliable information only.

Performances and limitations

  • Lighting conditions and accessories such as glasses and hat may alter the results.
  • Face deformations (such as a pronounced grin or grimace) can also have an impact on age estimation.
  • Gender estimation on young children may also be less accurate than on adults since they have less distinctive facial features.
  • The estimation results may fluctuate over time because they are computed on different facial images each time. This is to be expected and the age estimation for example can be used to estimate an “age category” (child/adult/elderly for example).
  • Facial expression estimation is not able to recognize too subtle facial expressions.