Multi-modal learning for gait-based human analysis and authentication
Abstract
Gait analysis is a key element in many tasks and application domains,including sports bio-mechanics,peoplere-identification and authentication, robotics, clinical evaluations and rehabilitation.In particular, human authentication from gait has attracted increasing interest in recent years, for different reasons- gait is difficult to precisely replicate, and it allows for authentication in unobtrusive and ecological conditions. On the negative side, current approaches show limitations in the presence of distracting factors like fatigue, impairment, or carrying conditions, that negatively affect the authentication performances. Typical sensors to address gait-based authentication are wearable devices (IMUs,smartwatches,...),videocam-eras and motion capture systems, and ground contact forces. Wearable devices are largely available, portable, small, and easy-to-use.As a drawback, they are noisy and poor. Among the other choices, video cameras are the ones coupling the richest information with the availability of very wide background knowledge in terms of methods and data.The general goal of this proposal is to conceive novel machine learning methodologies for gait-based humananalysis and authentication from inertial sensors in challenging conditions“ in the wild�. Our main scientific objective will be to investigate strategies to improve gait representation and learning by injecting knowledge from an alternative sensor modality, specifically videos. Their richness and descriptive power can provide a valuable source of knowledge to be transferred to models employing inertial data, sparser, and more sensitive to noise and instabilities. Moreover, visual data have been extensively studied, and there exists a large literature (including datasets, effective pre-trained models, and the potential for data synthesis) that can be used as starting point for our study.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 22, 2024
- Source ID
- FA86552317074
Entities
People
- Nicoletta Noceti
Organizations
- Air Force Office of Scientific Research
- United States Air Force
- University of Genoa