Feature Learning for Accelerometer based Gait Recognition
This work addresses gait recognition for biometric applications, but it is incremental as it builds on existing deep learning approaches with minor variations.
The paper tackled gait recognition from accelerometer data by comparing supervised and unsupervised feature learning methods, finding that autoencoders performed nearly as well as discriminative models and that fully convolutional models learned effective representations regardless of training strategy.
Recent advances in pattern matching, such as speech or object recognition support the viability of feature learning with deep learning solutions for gait recognition. Past papers have evaluated deep neural networks trained in a supervised manner for this task. In this work, we investigated both supervised and unsupervised approaches. Feature extractors using similar architectures incorporated into end-to-end models and autoencoders were compared based on their ability of learning good representations for a gait verification system. Both feature extractors were trained on the IDNet dataset then used for feature extraction on the ZJU-GaitAccel dataset. Results show that autoencoders are very close to discriminative end-to-end models with regards to their feature learning ability and that fully convolutional models are able to learn good feature representations, regardless of the training strategy.