Thesis (PhD) - Cyprus International University. Institute of Graduate Studies and Research Computer Engineering

This research aims to address challenges in age prediction using gait analysis by
analyzing human gait derived from preprocessed video data. The study identifies
problems associated with the quality and preprocessing of video data, which can affect
the accuracy of age estimation algorithms. The lack of comprehensive research on
preprocessing methods and suitable datasets further complicates the development of
reliable age estimation systems. The research aims to extract gait features, estimate
human age based on gait, implement biometric gait recognition using machine
learning, and evaluate the performance of the models.
The results demonstrate the effectiveness of different deep learning-based models in
estimating age based on gait data. The proposed Hybrid Model consistently achieves
the highest accuracy, followed by DenseNet and ResNet models. The standard CNN
model is the least accurate but still provides reasonable estimations among all other
tasted CNN models. The Hybrid Model emerges as the most promising choice for age
estimation due to its high accuracy and consistent performance.
The study also addresses the issue of imbalanced class samples through the use of
sparse categorical cross-entropy and class weight balance techniques. Both approaches
improve the model's performance on imbalanced training and testing data, with the
hybrid model maintaining high accuracy and stability. Comparing these models with
state-of-the-art models, the proposed hybrid models with sparse categorical cross
entropy or class weight balance outperform existing approaches in terms of accuracy.
This research contributes novel approaches for preprocessing human gait images,
developing hybrid models for age estimation, and utilizing sparse categorical cross
entropy and class weight balance techniques to handle imbalanced class samples. The
results demonstrate the effectiveness and potential of these approaches in improving
the accuracy and reliability of age estimation algorithms based on human gait analysis.