Face-recognition based security robot incorporating omnidirectional vision

Abstract

Security robots are gathering an increasing interest as a supplement to current security systems due to their advantages of continuous effective surveillance and low cost of operation. Current security robots, however, have their own limitations which is preventing them from being used on a large scale in real world environment. The main objective of this research is to develop a security robot that can efficiently be used in real-world environments by overcoming the limitations of most security robots, to guard properties and humans from intruders that may compromise the safety of the premises or its inhabitants. The robot developed in this research uses motion detection by image subtraction. When motion is perceived a face detection programme using the Viola Jones method coupled with skin-color-content detection is used to find out whether the motion was caused by a person or otherwise. If a face of a person or more is found, then the face recognition stage is utilized to verify the familiarity of the person(s) detected by using three different face recognition algorithms (Fisher faces, Eigen faces and LBPH (Local Binary Pattern Histogram)) to ensure maximum identification under different pose, facial expression and lighting conditions. If the person(s) is determined to be a stranger, the robot would raise an alarm and navigate towards the subject to authenticate him/her through a password entry check.