The purpose of this project was to provide forensic investigators with a new method of evidence collection, specifically by use of gait analysis. Gait analysis is currently used in the medical and athletic fields, but has not been widely accepted as a forensic tool. This is mostly due to concerns regarding whether a person’s gait is stable and unique, and the extent to which gait is influenced by external factors. This project has found evidence to support that gait does meet these criteria.This was done by developing a machine learning model that analyze a person’s gait through Gait Energy Images (GEI).
Within the scope of our model, we were able to predict a person’s walking gait with around 80% accuracy. We were not able to eliminate false positive identifications. Gait was found to be fairly repeatable between instances of a person walking. Gait was found to be unique between different people. We had the most success with GEI's over other image analysis methods. A CNN trained with a validation set, 50 epochs, Adam optimizer, learning rate scheduler, and categorical cross entropy was found to yield the best results over other models tested.
Semester of Project:
Team Photo:
Team Poster:
Problem Statement/Summary:
MS Teams Meeting Link:
Project Department:
Project Presentation Video Embed Code:
no client approval