A Multi-Gait Dataset for Human Recognition under Occlusion Scenario

Abstract

Biometric gait has found importance in recognition, security, behavior learning and also in clinical analysis. From a few decades research on gait recognition and dataset available publically focused on a single moving person. But in real time applications (such as shopping malls, railway stations, airport parking, etc.) where people walk in a group and occlusion issue affects the gait recognition performance. Considering this issue, we constructed a new database which focused on Multi-Gait (dynamic) occlusion situation. The dataset is classified into two categories i.e. first, Multi-Gait (MG), subjects walk in a group, and second, Single-Gait (SG), same subjects walk alone. Therefore, the dataset included both occluded and non-occluded gait patterns. The objective of this dataset is to analyze gait pattern variations when a person walks in a group or the same person walk separately. This dataset is also useful for researchers for identification of SG to MG.