Influence of walking in groups in gait recognition

Abstract

This contribution presents an evaluation on the performance of wearable gait recognition of individuals walking in groups. Three case scenarios are presented: Individual, when the user walks alone; Group, in which a group of users walk at the same time at a normal pace; and Formation, in which users walk in military formation. The evaluation was performed with the collaboration of the University Center of the Civil Guard (CUGC), whose students were used to form our database. Since all of their students are instructed to walk in formation, it was possible to perform a study on this particular pattern. This scenario is of interest as it simulates users trying to mimic each other's gait, for instance in a spoofing attack. The database consists of 10 students (7 males, 3 females), and every participant recreated each scenario 8 times The data was collected by a smartphone attached to the waist by means of a holster. With this database, a study on the influence of walking in groups was possible. Results obtained show similar accuracy in the cases of individual and group walking. However, when walking in formation, there is a significant drop in accuracy. These results point out that there is no significant influence of walking alone or with someone else, but, that there is a plausible vulnerability if users decide to walk imitating someone else.