Human Gait Analysis by Mining and Classification of Skeletal Data
Farnoush Banaei-Kashani, CMTC Fellow


In this project we develop automated data-driven methodologies to identify abnormalities and signatures in human gait solely based on the human skeletal gait data.

Problem that Inspired Research:
Our main research question is how one can accurately and efficiently identify frequent movement patterns from multivariate spatiotemporal skeletal gait data. Such patterns often correspond to abnormalities and/or signatures in human gait.

Objective / Proposed Solution:
Our main challenge in this project is accurate and efficient “gait classification”, i.e., the ability to label an instance of human gait based on pre-existing classes of gait, e.g., normal versus abnormal. We strive to develop accurate and efficient gait classification methodologies. Once such methodologies are developed, they will be applicable in a variety of application domains such as medical diagnosis for mobility disorders (e.g., Parkinson’s disease), gender and age classification, psychological evaluations, action/gesture/movement recognition, and most recently gait identification for individualization.

Greatest Challenge to Overcome:
Mining patterns from single variate data has been well studied and understood. Our main challenge in this project is to extend such solutions for multivariate pattern mining to be applicable for gait classification.

Benefits of Research:
If successful, the methodologies we introduce can, for example, help doctors to diagnose mobility disorders, law enforcement officers to identify criminals, sports trainers to study mobility of athletes, and motion detectors to label human actions.

Real-World Application(s):
Medical diagnosis, health monitoring, law enforcement, authentication (e.g., in airports), physical training, home automation, etc.

Innovations to Media and Technology:
As an automation methodology applicable in numerous application domains (see above for examples), the contribution of our proposed solution to the world of technology is obvious. Moreover, given that these methodologies derive knowledge from multimedia data, they help exploiting such data and enriching the applications for such data.

Cutting-edge Technology Being Used:
In this project we introduce and study various pattern mining and deep learning methodologies for accurate and efficient gait classification. In particular, deep learning is a cutting edge technology that is shown to significantly outperform existing solutions in other application domains. In this project, we are extending this technology to be applicable to our research problem.

Transdisciplinary Collaboration:
Our solutions are developed at the intersection of the fields of Machine Learning, Data Mining, and Computer Vision.