Phase registration in a gallery improving gait authentication

Orientation-Compensative Signal Registration for Owner Authentication Using an Accelerometer

“…The reason is that the gait model-based methods require an expert knowledge about the human gait to train the finite state machine, and hence they may fail when the input data is considerably deviated from the expert knowledge due to the variation. Meanwhile, since the frequency analysis-based methods require properly long and stable signal sequence to compute frequency coefficients or gait statistical characteristics, they may not work well with signals containing the temporal distortion induced by the covariates [19], [21], [25].…”

“…While an initial signal correspondence is given by using a sensor-orientationinvariant 1D resultant signal, the proposed method uses full three dimensions of signals for matching without any information loss, unlike the existing orientation-invariant methods [29], [30]. For period detection, we apply the above mentioned period detection-based method [21] since it has a good performance [34] and can work with any initial sensororientation.…”

“…The problem is because human gait signals always vary owing to many natural covariates such as mood, physical condition, walking speed, ground condition, weight being carried, shoes, and so on. To solve this problem, some existing period detection-based methods [21], [24], [26] apply dynamic time warping (DTW) for signal matching to overcome the temporal distortion. Particularly, there exists a period detection-based method [21] that can overcome the distortion not only in signal matching but also in gait period detection.…”

“…To solve this problem, some existing period detection-based methods [21], [24], [26] apply dynamic time warping (DTW) for signal matching to overcome the temporal distortion. Particularly, there exists a period detection-based method [21] that can overcome the distortion not only in signal matching but also in gait period detection. This method modifies the Self DTW [33] to accurately detect a period and remove the temporal distortion in constructing a gallery without any heuristic information, and thus achieved the best performance in large-scale evaluation in [34].…”

“…There are many existing methods for recognizing a person carrying a sensor based on the gait signals of an inertial sensor, such as identification [7], [18]- [20] or authentication [21]- [26], which showed a promising application of inertial sensors for biometric recognition. Because the human gait is a periodic motion, there are several approaches for gait-based owner recognition: gait model-based, frequency analysis-based, and period detection-based approaches.…”