Distributed Learning for Gait-based Human Recognition

Gait, a biometric characteristic observed during walking, is unique, similar to fingerprints and facial features. Analyzing an individual’s gait pattern enables accurate recognition. Gait recognition technology uses wearable sensors, such as accelerometers and gyroscopes, to collect data from individuals for identification purposes. This study proposes a FedGait method for distributed learning to address privacy concerns in centralized gait recognition. Unlike centralized methods, the proposed approach trains client and global models by sharing model parameters instead of raw data with the central serverensuring privacy during distributed training. This study analyzes both collected gait data and public datasets, allocating the data to local clients based on the number of participants. Experimental results demonstrate that the distributed learning method not only reduces privacy risks but also effectively recognizes gait identity.