IDNet: Smartphone-based Gait Recognition with Convolutional Neural Networks (1606.03238v3)

Abstract: Here, we present IDNet, a user authentication framework from smartphone-acquired motion signals. Its goal is to recognize a target user from their way of walking, using the accelerometer and gyroscope (inertial) signals provided by a commercial smartphone worn in the front pocket of the user's trousers. IDNet features several innovations including: i) a robust and smartphone-orientation-independent walking cycle extraction block, ii) a novel feature extractor based on convolutional neural networks, iii) a one-class support vector machine to classify walking cycles, and the coherent integration of these into iv) a multi-stage authentication technique. IDNet is the first system that exploits a deep learning approach as universal feature extractors for gait recognition, and that combines classification results from subsequent walking cycles into a multi-stage decision making framework. Experimental results show the superiority of our approach against state-of-the-art techniques, leading to misclassification rates (either false negatives or positives) smaller than 0.15% with fewer than five walking cycles. Design choices are discussed and motivated throughout, assessing their impact on the user authentication performance.

• The paper introduces a robust smartphone-based gait recognition framework that leverages CNNs to autonomously extract orientation-invariant features.
• It achieves a misclassification rate below 0.15% within five walking cycles by integrating a one-class SVM with a multistage authentication process.
• The study demonstrates the potential for non-obtrusive mobile security and encourages further research into deep learning for biometric authentication.

Overview of IDNet: Smartphone-based Gait Recognition with Convolutional Neural Networks

The paper presents IDNet, an authentication framework leveraging smartphone-acquired inertial signals for recognizing an individual's unique gait. The research aims to affirm the feasibility and efficacy of using commercial smartphones equipped with accelerometers and gyroscopes as gait recognition tools, providing an unobtrusive and orientation-independent solution for user authentication.

Key Innovations and Methodology

IDNet introduces several noteworthy components to tackle the challenges of smartphone-based gait recognition:

1. Walking Cycle Extraction: It features a robust cycle extraction technique that remains invariant to the smartphone's orientation, addressing a common obstacle in wearable sensor-based recognition.
2. Feature Extraction Using CNNs: The paper applies Convolutional Neural Networks (CNNs) as universal feature extractors. Unlike traditional methods relying on manually engineered features, this approach autonomously extracts features, thus facilitating more accurate and generalizable recognition.
3. One-Class SVM and Multistage Authentication: IDNet integrates a one-class Support Vector Machine (SVM) trained on data from the target user alone, alongside a multistage decision framework that aggregates classification results over successive walking cycles to enhance accuracy.

Experimental Results and Comparisons

The paper provides comprehensive experimental results demonstrating significant improvements over state-of-the-art methods. IDNet achieves a misclassification rate of below 0.15% with fewer than five walking cycles, a substantial advancement compared to existing techniques, which are typically characterized by error rates ranging from 5% to 15%.

Implications and Future Developments

Practically, the development and deployment of IDNet offer a promising pathway for secure, non-obtrusive user authentication on mobile devices, which could be particularly impactful for mobile security, health monitoring, and user identification applications. Theoretically, the fusion of CNNs and machine learning classifiers in wearable sensor-based recognition underscores a methodological shift towards more automated and scalable feature extraction techniques.

Looking forward, this research could spur further exploration into adaptive learning models that continuously refine the feature extraction and authentication processes as more data is acquired. Moreover, investigating the applicability of similar technologies to other biometric signals collected from wearable devices could broadens its scope beyond gait analysis. Future work could also explore the integration of additional sensor data, or the application of more advanced deep learning architectures to further enhance authentication accuracy and robustness.

Overall, the paper convincingly argues for the adoption of deep learning methodologies in biometric authentication, paving the way for more sophisticated and reliable security solutions across various mobile applications.