查看文章

Biometric methods for verifying, ie authenticating, someone’s identity are increasingly being used. Today’s commercially available biometric systems show good reliability. However, they generally lack user acceptance. Users show an antipathy touching a fingerprint scanner and they dislike looking into an iris scanner that might eventually malfunction and impair their vision. In general, they favour systems with the least amount of interaction. Using gait as a biometric feature would lessen such problems since it requires no subject interaction other than walking by. Consequently, this would increase user acceptance. And since highly motivated users achieve higher recognition scores, it increases the overall recognition rate as well.

This monograph describes a biometric system that uses individual characteristics of human gait for authentication. Two sensors measuring different physical properties of the walking person were used. First, a force sensor measures the Ground Reaction Force (GRF) perpendicular to the floor and second, a video sensor captures a side view of the passing person. Computationally efficient algorithms were developed to extract five different feature types, ie modalities, from the acquired gait data. A novel variant of the Generalised Principal Component Analysis (GPCA) was devised to reduce data dimensionality without losing, or even better, with improving person separability. Last but not least, a Bayes Risk Criterion approach is used to fuse the five modalities.