Ensemble Model of VGG16, ResNet50, and DenseNet121 for Human Identification Through Gait Features

Human Gait Recognition is a type of behavioral biometric authentication based on walking pattern of an individual. Every individual has a unique way of walking making authentication based on gait features difficult to masquerade. It is non-intrusive in nature as it does not require any active participation from the subject making it an apt identification technique. In the past, techniques used in this field required significant amount of expert knowledge for feature identification and also lacked the ability to capture all the complex gait patterns resulting in low accuracy of the predictions made. The paper presents an ensemble model for gait recognition, combining knowledge of multiple Convoluted Neural Network (CNN) models namely VGG16, ResNet50 and DenseNet121. The individual models were fine-tuned on the Casia-B dataset and an ensemble model was created to exploit the individual strengths of the models making this proposed model, a powerful one in terms of accuracy.