Review of gait recognition approaches and their challenges on view changes

Abstract

Gait or walking pattern has been known as one of the alternative biometric solutions used in surveillance monitoring and control. The methods of gait recognition have been developed for decades using various techniques in different concepts. This study is a review paper collecting gait recognition approaches in both perspectives of model-based approaches relying on key joints/parts of the human body and appearance-based approaches relying on gait silhouettes. The existing methods addressing one of the most important real-world challenges, i.e. view changes, are emphasised and summarised in this study. Also, recent methods based on convolutional neural network solving the gait recognition and their challenges of view changes are illustrated. In addition, the publicly-available gait datasets and corresponding recognition performance and comparison are concluded in each section. The state-of-the-art gait recognition methods can achieve up to a perfect score of 100% accuracy for the normal walking, and above 80% in average for the view changes ranging from 0° to 180°.