Software: Feature Recognition for Stereotype Learning in Human-Robot Interaction

Researchers: Azfar Aziz and Alan Wagner

Stereotype learning and use requires that a robot be capable of recognizing a person's higher level features. This project, funded by the Naval Surface Warfare Center, located in Crane, IN, develops software for the recognition of high level attributes such as gender, the presence of glasses, and the presence of facial hair. Our recent research has developed methods that allow a robot to learn and reason about categories of people. In order for these methods to work, however, the robot must be able to use its perceptual modalities (camera for vision, microphone for auditory, etc.) to create a set of features describing a person. We believe that the more features available to the robot the better the methods will work. Unfortunately, very little software is freely available that allows a robot to recognize high-level features such as gender, age, hair style, body type, etc. This project therefore represents an initial step towards the creation of a suite of software classification methods that will generate these high level features.

The link below provide the software to create and test the classifiers. The XML and .dat files are for simply loading and using the classifier. Right now these algorithms only seem to work on Linux systems. Please read the ReadMe file for more information.

Results

Software

Software used to create and test classifiers. [download]

Xml and .dat files. [download]

ROC Curves for the different classifiers. [ROC]