Inertial Measurement Units (IMUs)-based gait analysis is a promising and attractive approach for user recognition. Recently, the adoption of deep learning techniques has gained significant performance improvement. However, most existing studies focused on exploiting the spatial information of gait data (using Convolutional Neural Network (CNN)) while the temporal part received little attention. In this study, we propose a new multi-model Long Short-term Memory (LSTM) network for learning the gait temporal features. First, we observe that LSTM is able to capture the pattern hidden inside the gait data sequences that are out-of-synchronization. Thus, instead of using the gait cycle-based segment, our model accepts the gait cycle-free segment (i.e., fixed-length window) as the input. By this, the classification task does not depend on the gait cycle detection task, which usually suffers from noise and bias. Second, we propose a new LSTM network architecture, in which, one LSTM is used for each gait data channel and a group of consecutive signals is processed in each step. This strategy allows the network to effectively handle the long input data sequence and achieve improved performance compared to existing LSTM-based gait models. In addition, besides using the LSTM alone, we extend it by combining with a CNN model to construct a hybrid network, which further improves the recognition performance. We evaluated our LSTM and hybrid networks under different settings using the whuGAIT and OU-ISIR datasets. The experiments showed that our LSTM network outperformed the existing LSTM networks, and its combination with CNN established new state-of-the-art performance on both the verification and identification tasks.

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 sensorbased 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-of-the-art researches.

Similarity preserving is a key ingredient of cancellable biometric scheme design. The notion ensures the accuracy performance of the biometric systems can be preserved after the cancellable biometric technique is applied. Random Projection is among the most commonly adopted method in cancellable biometric schemes. However, it is reversible subject to certain conditions, which disrupts the template irreversibility criterion. This invites vulnerabilities for random projection-based schemes. In this paper, we propose a novel transform function, namely Absolute Value Equations Transform (AVET), which non-linearly projects feature vectors to another domain. The transformed templates hold two main merits ensuring the user’s privacy, and maintaining the system’s performance simultaneously. First, by relying on the hardness of the Absolute Value Equations problem, we guarantee that AVET satisfies irreversibility. Second, by using Johnson–Lindenstrauss lemma and the inverse triangle inequality, we prove that the proposed approach has the similarity preserving property. Notably, rigorous theoretical proofs and empirical experiments are provided. The efficacy of AVET is comprehensively evaluated on both physiological and behavioral biometrics including face, ear, fingerprint, and gait. With unimodal approach, we achieve competitive performances compared to related algorithms on eight public datasets. Regarding bimodal mode, the AVET surpasses the state-of-the-art technique on all three observed datasets. To the best of our knowledge, this is the first study that attempts to develop a secure transformation to augment the role of Random Projection in the existing cancellable biometric schemes.

Smart environments established by the development of mobile technology have brought vast benefits to human being. However, authentication mechanisms on portable smart devices, particularly conventional biometric based approaches, still remain security and privacy concerns. These traditional systems are mostly based on pattern recognition and machine learning algorithms, wherein original biometric templates or extracted features are stored under unconcealed form for performing matching with a new biometric sample in the authentication phase. In this paper, we propose a novel gait based authentication using biometric cryptosystem to enhance the system security and user privacy on the smart phone. Extracted gait features are merely used to biometrically encrypt a cryptographic key which is acted as the authentication factor. Gait signals are acquired by using an inertial sensor named accelerometer in the mobile device and error correcting codes are adopted to deal with the natural variation of gait measurements. We evaluate our proposed system on a dataset consisting of gait samples of 34 volunteers. We achieved the lowest false acceptance rate (FAR) and false rejection rate (FRR) of 3.92% and 11.76%, respectively, in terms of key length of 50 bits.