Deep learning pipelines for recognition of gait biometrics with covariates: a comprehensive review

This paper presents a comprehensive exposition of deep learning architectures and pipelines for biometric applications using complex characteristics of human gait. The variety and complexity that we have come across encompass the majority of deep learning techniques. Recognizing humans by their walking patterns is a complicated biometric processing approach that identifies people without intervention and works well even with low-resolution images. Several covariates, such as footwear, heavy clothing and carry situations, view angle, occlusion, speed transition, and others, can affect the gait recognition rate. We provide an extensive literature review focusing on the performance of deep learning models in covariate conditions. It shall help researchers to understand specific aspects of deep learning pipelines in gait recognition. The discussion is done on data acquisition, input, dataset, preprocessing, feature extraction, transformation, activation function, classification and training parameters. We also demonstrate the most often used strategies for each parameter over the previous five years. The datasets used so far are compared with the accuracy achieved for each covariate condition. Finally, we listed the benefits and drawbacks of deep learning approaches in covariate conditions along with open problems in the identification based on behavioral traits and concluded the paper by highlighting important lessons.

Automated summary

This paper provides a comprehensive overview of deep learning architectures and pipelines for biometric applications using complex characteristics of human gait. The authors discuss the challenges in gait recognition due to various covariates and present a literature review on the performance of deep learning models in covariate conditions. They also cover various aspects of deep learning pipelines in gait recognition, such as data acquisition, preprocessing, feature extraction, and classification. The paper concludes by highlighting the benefits and drawbacks of deep learning approaches in covariate conditions and identifying open problems in identification based on behavioral traits.