Human detection and face recognition in indoor environment to improve human-robot interaction in assistive and collaborative robots

Abstract

Human detection in indoor environment is essential for Robots working together with humans in collaborative manufacturing environment. Similarly, Human detection is essential for service robots providing service with household chores or helping elderly population with different daily activities. Human detection can be achieved by Human Head detection, as head is the most discriminative part of human. Head detection method can be divided into three types: i) Method based on color mode; ii) Method based on template matching; and iii) Method based on contour detection. Method based on color mode is simple but is error prone. Method based on head template detects head in the image by searching for a template which is similar to head template. On the other hand, Method based on contour detection uses some information to describe head or head and shoulder information. The use of only one criteria may not be sufficient and accuracy of human head detection can be increased by combining the shape and color information. In this thesis, a method of human detection is proposed by combining the head shape and skin color (i.e., Combination of method based on Color mode and method based on Contour detection). Mainly, curvature criteria is used to segment out curves having similar curvature to find human head. Further, skin color is detected to localize face in image plane. A curve represents human head curve if only it has sufficient skin colored pixel in its closed proximity. Thus, by using color and human head curvature it was found that promising results could be obtained in human detection in indoor environment. iv After detecting humans in the surrounding, the next step for the robot could be to identify and recognize them. In this thesis, the use of Gabor filter response on nine points was investigated to identify eight different individuals. This suggests that the Gabor filter on nine points could be applied to identify people in small areas, for example home or small office with less individuals.