Enhancing Gait Recognition through Edge Detection and Hybrid Feature Extraction Techniques
Abstract
Gait recognition is the process of identifying individuals based on their unique
walking patterns. This method has gained importance in biometric systems for its
non-intrusive nature and applications in security, surveillance, and healthcare, where
it helps monitor and identify individuals from a distance. However, accurately distin-
guishing between gait patterns under varying conditions poses significant challenges.
These include variations in walking speed, changes in clothing and footwear, and
different environmental conditions, which all affect the accuracy of gait recognition
systems. Moreover, the lack of diverse and high-quality datasets complicates the
development of robust models that generalize well across different populations.
To overcome these challenges, we have developed a novel approach that incorpo-
rates various preprocessing techniques, such as edge detection, contrast enhancement,
and noise reduction, with feature selection methods to improve data quality and model
performance. Our hybrid feature extraction model combines Kolmogorov-Arnold Net-
works (KANs), ResNet, EfficientNet, and Principal Component Analysis (PCA) for
spatiotemporal features, with traditional methods like Histogram of Oriented Gradi-
ents (HOG) and Local Binary Patterns (LBP). This approach bridges the research
gap by demonstrating improved performance on the Chinese Academy of Sciences In-
stitute of Automation (CASIA) datasets A, B, and C, enhancing robustness. These
advancements are particularly promising for real-time applications in healthcare, such
as early detection of neurological conditions like Alzheimer’s disease (AD).
Evaluations using metrics such as Accuracy, F1 Score, and Area Under the Curve
(AUC) have been conducted. The results show that the KAN model achieved the
highest overall accuracy, surpassing 95%, while the ResNet model excelled across all
metrics, effectively handling complex data variations. Integrating traditional and deep
learning features boosts model accuracy and robustness. Comparative analysis with
existing techniques confirms that our proposed models outperform previous methods
in key metrics, validating the hypothesis that integrating diverse features and robust
training strategies can substantially enhance gait recognition systems.