Determining gender of walking people using multiple sensors

Describes a method for determining the gender of walking people using multiple sensors: a camera, an accelerometer, and a tactile sensor. In the field of marketing, a system is needed that automatically recognizes the gender of customers. However, difficulties in collecting information that can be used to effectively discern gender and in obtaining robust and realtime computation have prevented the development of such a system. To solve these problems, we propose a system using multiple sensors. The features for recognition are the entropy value of a morphological pattern spectrum of the silhouette of a person as captured by a camera, the FFT cepstrum of footsteps from an accelerometer, and the pressure distribution pattern of footprints from a tactile sensor. By fusing these features in a statistical method, we achieve efficient data acquisition, are able to establish a discriminant axis that clearly separates men from women, and offers adequate robustness. Experiments show that these advantages result in high recognition rates