Skeletons on the Stairs: Are They Deceptive?

Abstract

Physiological signals have been widely applied for deception detection. However, these signals are usually collected by devices attached to subjects. Such attachments can cause discomfort and unexpected anxiety, and thus will be noticed. Alternatively, skeleton-based gait data collected in a non-contact setting can be a solution to detect deception. Therefore, in this paper, we aim to investigate whether liars can be recognized using their skeletal motion trajectories. We extract skeletal gait data from videos of participants going up and downstairs after they conduct a mock crime. With the extracted skeletal gait data, a simplified version of Multi-Scale Graph 3D (MS-G3D) network is able to recognise participants’ deceptive behaviour with an average accuracy of 70.9%. This result is higher than those obtained from traditional classifiers such as neural networks, support vector machines and decision trees, which are trained on hand-crafted features calculated from the gait data. Show more