Data Augmentation for Inertial Sensor-Based Gait Deep Neural Network

Abstract

Inertial sensor-based gait has been discovered as an attractive method for user recognition. Recently, with the approaching of deep learning techniques, new state-of-the-art researches have been established. However, the scarcity of training data still endures as an obstacle that impedes to build a robust deep gait model. In this study, we address that problem by proposing a novel approach for inertial sensor-based gait data augmentation. First, two label-preserving transformation algorithms, namely Arbitrary Time Deformation (ATD) and Stochastic Magnitude Perturbation (SMP), are proposed to generate more training data from the real gait data. The ATD algorithm adjusts the timing information of gait data with random values, on the other hand, SMP alters the magnitude arbitrarily, to create variations on the augmenting data. Then, we design a generic gait recognition model using convolutional neural network, in which, the ATD and SMP algorithms are coordinated appropriately to produce augmenting data varied naturally in both time and magnitude as real data. The proposed approach was evaluated on two public datasets, one was collected in unconstrained conditions, and the other had the largest number of participating users. The experiment showed that, under different amounts of training data, using ATD or SMP alone could increase the recognition performance effectively, and their combination even attained higher accuracy. With ATD and SMP, our model achieved competitive performance on both two datasets comparing to state-ofthe-art researches.

Multi-Model Long Short-Term Memory Network for Gait Recognition Using Window-Based Data Segment

“…The model achieved the FAR of 6.4% and FRR of 5.4% for a dataset of 36 users. After that, this field received more research attention, and a large number of gait recognition models have been proposed in the literature [9], [13], [16]- [20], [31].…”

“…Most prior IMUs-based gait recognition models used handcrafted methods for feature extraction, which could be summarized in [12]. The later researches adopted deep learning techniques to automatically extract more discriminative and stable features, which showed the improved performance [17]- [20], [22]. In addition, some researches improved the recognition performance by combining multiples sensors [32], [33].…”

“…In that study, the authors used it to confirm the performance of existing methods published so far, and the best accuracy as 70.2% was reported when measured in each segment. After that, it has been used to evaluate the recognition performance in many studies [18]- [20], [44]. On this dataset, our LSTM network achieved higher identification performance comparing to the methods used in studies [19], [20], [27].…”

“…After that, it has been used to evaluate the recognition performance in many studies [18]- [20], [44]. On this dataset, our LSTM network achieved higher identification performance comparing to the methods used in studies [19], [20], [27]. In addition, our proposed hybrid network over-performed the method used in [44].…”

“…Gait is a special biometric modality that contains both spatial and temporal information [14]. However, major efforts in recent studies focused on using Convolutional Neural Network (CNN) [15] to extract the spatial information (e.g., [16]- [20]), while there was little attention for the temporal part. Some latest studies adopted Long Short-term Memory (LSTM) network [21] to capture the temporal information from the gait data sequence and showed promising results [19], [22].…”