On-device gait-based identification: Real-time recognition using CWT and deep CNN on mobile accelerometer data

The proliferation of smartphones has driven efforts to harness these devices for secure digital identification. For smartphones to serve as digital keys, they must first reliably verify the identity of the device holder before granting access. This study introduces Smart Identification of Legitimate or Counterfeit (Smart ILC), a novel on-device user identification system that leverages behavioral biometrics—specifically gait analysis—using smartphone accelerometer data and deep learning. Raw X-axis accelerometer signals are transformed into two-dimensional scalogram images via Continuous Wavelet Transform (CWT) and analyzed using a Convolutional Neural Network (CNN), enabling real-time, contactless identification by capturing both temporal and frequency-domain features of individual walking patterns. The system was evaluated on a private dataset of 31 participants—20 for model development and 11 for real-world validation—achieving 98% accuracy and successfully rejecting all but two non-owner attempts out of 220, corresponding to a False Accept Rate of 0.91% and a True Negative Rate of 99.09%. To enable direct benchmarking, the statistical method from our prior work was re-applied to the same dataset, achieving 70.30% accuracy, demonstrating the discriminative advantage of CWT-based deep feature extraction over Gaussian-based peak scoring. Cross-dataset evaluation on the publicly available MotionSense benchmark further confirmed the system’s generalization capability, with both methods achieving a False Accept Rate of 0.00% against 24 independent public participants—providing direct evidence of robust impostor rejection beyond the training population. These findings demonstrate that Smart ILC achieves 98% accuracy on the private dataset—outperforming the statistical baseline (70.30%)—while matching its 0.00% False Accept Rate on the independent public benchmark, confirming robust cross-dataset generalization. Once verified, the smartphone functions as a secure digital key for applications such as smart door and vehicle entry, with all processing performed locally on the device—ensuring full user privacy and eliminating cloud dependency.