Identifying New Persons in the Context of the Robocup@Home Competition Using KNN

• David Segalle UTFPR
• Gustavo Fardo Armênio UTFPR
• Gustavo Fontes Lewin UTFPR
• Everson De Souza Silva UTFPR
• João Alberto Fabro UTFPR

Resumo

Robocup@Home proposes a challenge related to Person Recognition: after presented, a new ‘operator’ should become ‘immediately’ recognizable by the robot. The presentation procedure may require the operator to correctly interact with the robot, following a certain procedure, as instructed by the robot itself (for example, staying in front of the robot, so that the robot can take pictures of this person). In this paper, we propose the use of the KNN (K-Nearest Neighbor) supervised machine learning algorithm to include a new ‘operator’ in a database of persons recognizable by the robot. This algorithm uses information taken from an image segmentation of the face of the operator. The experiment evaluates how long it takes to include a new operator if the robot has from 1 to 12 current operators, evaluating also how long it takes to include this operator based on 1, 2 or more images of the new operator, taken from slightly different points of view. The results confirm that KNN can be used to ‘present to the robot’ up to 13 new operators, with up to 15 images for each operator, in less than 60 seconds.

Palavras-chave: KNN algorithm, person recognition, face recognition, online training