saba-gul / gait-analysis-using-3d-cnn

Gait Recognition with 3D CNN. This project proposes a novel approach using 3D convolutional neural networks (3D CNN) to capture spatio-temporal features of gait sequences for robust recognition in an un-intrusive manner.

Gait Recognition with 3D CNN

Introduction

Gait recognition is a promising avenue for identification and authentication due to the uniqueness of individual strides. This project proposes a novel approach using 3D convolutional neural networks (3D CNN) to capture spatio-temporal features of gait sequences for robust recognition.

Methodology

• 3D CNN Architecture: The proposed network architecture employs a holistic approach using gait energy images (GEI) to capture shape and motion (Spatio-Temporal features) characteristics of human gait.

• Dataset: Evaluation was conducted on two publicly available datasets, OULP and CASIA-B, which exhibit substantial gender and age diversity.

• Optimization Strategies: Bayesian algorithms were explored to tune hyperparameters and enhance network performance.

Key Features

• Robust Gait Recognition: The optimized 3D CNN and GEI effectively capture unique gait characteristics despite challenging covariates such as change in speed, viewpoint, clothing, and carrying accessories.
• State-of-the-Art Results: Achieved state-of-the-art results on multi-views and multiple carrying conditions of subjects in the CASIA-B dataset.

Future Directions

• Overcoming Overfitting: Address potential overfitting issues due to limited variance and frames per subject in the OULP dataset.
• Genetic Optimization Algorithms: Explore genetic optimization algorithms to further enhance performance.
• Real-life Scenarios: Extend the framework to practical environments by tackling more challenging real-life scenarios for person identification based on walking patterns.

Usage

• Clone the repository.
• Install necessary dependencies.
• Execute the main script for gait recognition.

Citation

Gul S., Malik M.I., Khan G.M., Shafait F. (2021) Multi-view Gait Recognition System using Spatio-temporal Features and Deep learning, Expert Systems with Applications,115057, ISSN 0957-4174, https://doi.org/10.1016/j.eswa.2021.115057.