Analyzing Pulse From Head Motions in Video

We extract heart rate and beat lengths from videos by measuring subtle head oscillations that accompany the cardiac cycle. Our method tracks features on the head, temporally filters their trajectories and performs principal component analysis (PCA) to decompose the trajectories into a set of ID component motions. It then chooses the component that is most periodic as the pulse signal. Finally, we identify peaks of the chosen signal, which correspond to heartbeats. When evaluated on 18 subjects our approach reported heart rates nearly identical to an electrocardiogram (ECG) device for all subjects and obtained similar beat length distributions for 17. In addition we obtained pulse rate from videos of the back of the head and of sleeping newborns. Initial findings also show that our method can measure heart rate from body parts other than the head and can measure respiration rate by selecting a different frequency band. Finally, we present visualization techniques such as motion magnification for subjective analysis of body motions. Thesis Supervisor: John Guttag Title: Professor, Electrical Engineering and Computer Science Thesis Supervisor: Fredo Durand Title: Professor, Electrical Engineering and Computer Science