Sensor orientation invariant mobile gait biometrics

Abstract

Accelerometers and gyroscopes embedded in mobile devices have shown great potential for non-obtrusive gait biometrics by directly capturing a user's characteristic locomotion. Despite the success in gait analysis under controlled experimental settings using these sensors, their performance in realistic scenarios is unsatisfactory due to data dependency on sensor placement. In practice, the placement of mobile devices is unconstrained. In this paper, we propose a novel gait representation for accelerometer and gyroscope data which is both sensor orientation-invariant and highly discriminative to enable high-performance gait biometrics for real-world applications. We also adopt the i-vector paradigm, a state of the-art machine learning technique widely used for speaker recognition, to extract gait identities using the proposed gait representation. Performance studies using both the naturalistic McGill University gait dataset and the Osaka University gait dataset containing 744 subjects have shown dominant superiority of this novel gait biometrics approach compared to existing methods.

Pace independent mobile gait biometrics

“…The goal of this analysis is to investigate the effectiveness of the proposed algorithm for practical gait biometrics, and in particular, for data with variations in walking speed. We compare the performance of the proposed algorithm to [41] in two scenarios: with normal walking speed, and with multiple speeds.…”

“…We use the i-vector approach, a state of the art speaker biometrics method [8], for gait authentication in a similar manner as in [41] where promising gait authentication performance has been achieved. Although speech and gait authentication represent different application domains, these two problems are similar in nature as both need to extract subject specific signatures from sensory data confounded with variations from a range of irrelevant sources (such as speech content and microphone channel effect for speaker authentication; footwear and pocket size for mobile gait authentication).…”

“…As a result, it is too computationally intensive to be applicable to large gait databases. Zhong and Deng [41] recently proposed a sensor orientation invariant gait representation called gait dynamic images where they leveraged pairwise interactions to factor out the sensor orientation. They demonstrated that by using this invariant representation with advanced machine learning method, they can drastically reduce the recognition error on the largest gait dataset [28] containing more than 700 subjects.…”

“…Gait dynamics images (GDIs) are a recently proposed sensor orientation invariant gait representation [41]. Given a 3-D time series captured by a 3-axis accelerometer, its GDI is a 2D array where the horizontal axis is for time t and the vertical axis is for time lag l. Each entry encodes the interaction in terms of time lagged self-similarity of the 3D acceleration vectors.…”

“…In particular, we address the challenge of varying walking speed, a constant presence in real world gait. We also build our gait representation based on Gait Dynamics Images (GDIs) [41], a sensor orientation invariant gait representation, and further advance it to compute a highly discriminative mobile gait representation that is both compensated for varying walking speed and invariant to sensor orientation. We move forward the state of the art for accelerometersbased gait biometrics in the following manner: 1.…”