Orientation invariant gait matching algorithm based on the Kabsch alignment

Accelerometer and gyroscope sensors in smart phones capture the dynamics of human gait that can be matched to arrive at identity authentication measures of the person carrying the phone. Any such matching method has to take into account the reality that the phone may be placed at uncontrolled orientations with respect to the human body. In this paper, we present a novel orientation invariant gaitmatching algorithm based on the Kabsch alignment. The algorithm consists of simple, intuitive, yet robust methods for cycle splitting, aligning orientation, and comparing gait signals. We demonstrate the effectiveness of the method using a dataset from 101 subjects, with the phone placed in uncontrolled orientations in the holster and in the pocket, and collected on different days. We find that the orientation invariant gait algorithm results in a significant reduction in error: up to a 9% reduction in equal error rate, from 30.4% to 21.5% when comparing data captured on different days. On the McGill dataset from 20 subjects, which is the other dataset with orientation variation, we find a more pronounced effect; the identification rate increased from 67.5% to 96.5%. On the OU-ISIR data, which has data from 745 subjects, the equal error rates are as low as 6.3%, which is among the best reported in the literature.