Pulse Detection from Head Motions in Video

Technology #16417

Questions about this technology? Ask a Technology Manager

Download Printable PDF

Image Gallery
Overview of the pulse estimation approach: (a) A region is selected within the head and feature points are tracked for all frames of the video. (b) The vertical component is extracted from each feature point trajectory. (c) Each trajectory is then temporally filtered to remove extraneous frequencies. (d) PCA decomposes the trajectories into a set of source signals. (e) The component which has clearest main frequency is selected. (f). Peak detection identifies the beats of the signal.
Professor Fredo Durand
Department of EECS & CSAIL, MIT
External Link (graphics.csail.mit.edu)
Professor John Guttag
Department of EECS & CSAIL, MIT
External Link (people.csail.mit.edu)
Guha Balakrishnan
Managed By
Daniel Dardani
MIT Technology Licensing Officer
Patent Protection

Pulse detection from head motions in video

US Patent Pending 2015-0005646
Detecting pulse from head motions in video
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2013, pp. 3430-3437


The Inventors exploit subtle head oscillations that accompany the cardiac cycle to extract information about cardiac activity from videos. This method provides an unobtrusive way of measuring heart rate and can be used to extract other clinically useful information about cardiac activity, such as the subtle changes in the length of heart-beats that are associated with the health of the autonomic nervous system.

Problem Addressed  

Heart rate is a critical vital sign for medical diagnosis. There is a growing interest in extracting it without contact, particularly for populations such as premature neonates and the elderly who have fragile skin that may be damaged by traditional sensors. Furthermore, as the population ages, frequent monitoring outside of clinical environments can provide doctors with not just timely samples but also long-term trends and statistical analyses. This continuous monitoring requires a non-obtrusive technology. The Inventors have developed one such method of non-invasive heart monitoring by detecting pulse from head motions in video.


The cyclical movement of blood from the heart to the head via the abdominal aorta and the carotid arteries causes the head to move in a periodic motion. The Inventors’ algorithm detects pulse from this movement. This method takes an input video of a person’s head and returns a pulse rate as well as a series of beat locations which can be used for the analysis of beat-to-beat variability. The motion of the head is extracted using feature tracking, upon which the motion corresponding to the pulse is isolated and projected onto a 1D signal to extract individual beat boundaries from the peaks of the trajectory. For this, principal component analysis (PCA) is used and the component whose temporal power spectrum best matches a pulse is selected.

The algorithm projects the trajectories of feature points onto this component and extracts the beat locations as local extrema. The correct source for analyzing and computing the duration of individual beats is chosen by examining the frequency spectra and selecting the source with the clearest main frequency. Average pulse rate is identified using this frequency. For more fine-grained analysis and calculation of beat durations, peak detections are performed in the time domain.


  • Non-invasive, non-obtrusive method for continuous heart monitoring
  • Can be used as a diagnosis tool for patients for whom traditional heart monitoring methods might be risky, including premature neonates and the elderly