What is gait recognition?
Biometric recognition
Walking patterns
Identify individuals
Gait recognition is a biometric application to identify individuals by their walking pattern. This walking pattern is unique to everyone. By capturing the gait sequence, our project aims to identify individuals in the real world from their gait.
Why gait recognition?
Gait is your walking style and is unique to you.
Compared to other popular forms of recognition using fingerprints and facial features, gait recognition is non-intrusive, it works from a distance, and is unaffected by changes in appearance (bags, clothes, masks).
Comparing with other biometrics
Helping society
GaitSearch aids in providing evidence in investigations to solve crimes, and may enhance public safety and security.
With a higher chances of getting identified, GaitSearch aids in deterring individuals from committing crime, thus reducing crime rates.
By identifying individuals, GaitSearch can help find missing individuals, bringing relief to their families and loved ones.
How we achieved this
Silhouette extraction
Subject silhouettes are segmented using a Deep Learning model called SCNet.
Individual gait signatures
Our ensemble model recognises an individual's unique gait signature. It is based on GaitSet (2018) and GaitPart (2020)
Local and global feature extraction
The model consists of several CNN blocks that allow local & global feature extraction.
Ease-of-use and metrics
50%
50% rank-1 accuracy for real world recognition, jumps up to over 80% when considering rank-3 to rank-5 accuracies.
↑3%
Improvement over the state-of-the-art models.
80%
80% of users found GaitSearch to be fun and easy to use.
