Today Dwight Goins shares a great example on using one of the coolest features of the Kinect v2, heart rate detection...
When the latest Kinect sensor was unveiled more than a year ago at Build 2014, demos showed how it could determine a user’s heart rate without attaching sensors or wires to his or her body. But that was old news to regular followers of D Goins Insperience, the personal blog of Dwight Goins, a Microsoft Kinect for Windows MVP and founder of Dwight Goins Inc. As Goins revealed in February 2014, he had already devised his own application for detecting a person’s heart rate with the preview version of the latest Kinect sensor.
Goins’ app, which he has subsequently refined, takes advantage of three of the latest sensor’s key features: its time-of-flight infrared data stream, its high-definition-camera color data stream, and face tracking. The infrared stream returns an array of infrared (IR) intensities from zero to 65,536, the color stream returns RGB data pixels, and the face tracking provides real-time location and positioning of a person’s face. He thus knew how to capture a facial image, measure its infrared intensity, and gage the RGB color brightness level in its every pixel. The following video shows Goins' Kinect v2 heart rate detector in action.
Project Information URL: https://blogs.msdn.com/b/kinectforwindows/archive/2015/06/12/detecting-heart-rate-with-kinect.aspx
Kinect for Windows v2 Heart Rate Library
his application is a .Net WPF application which uses the R Statistical programming language engine version > 3.12. This application requires the R engine to be installed on the system running the application. R can be installed from here: http://cran.r-project.org/ The WPF application utilizes the Kinect RGB, IR, and Face streams of data to determine a region around the face and calculate a spatially averaged brightness over time. The averaged values are then divided by their respective standard deviations to provide a unit variance value. These values are required for feeding into ICA algorithms. The values are saved into a csv file for processing with other Machine Learning techniques and algorithms.
The basic approach is simple. When a person's heart pumps blood, the volume of blood is pushed through various veins and muscles. As the blood pumps through the muscles, particularly the face, the more light is absorbed, and the less brightness the a web camera sensor picks up. This change in brightness value is very minute and can be extracted using mathematical tricks. The change in brightness is periodic. In other words, a signal or wave. If we can match the signal/wave to that of a blood pulse, we can calculate the heart rate.
In order to match the change in brightness to a blood pulse we use the Independent Component Analysis (ICA) concept. This concept is the cocktail party concept and is the basis for finding hidden signals within a set of mixed signals. If you have two people talking in a crowded room, and you have microphones placed at various locations around the room, ICA algorithms let you take a mixed sample of signals, such as sound waves, and calculates an estimated separation mixture of components. If you match the separate components to the original signal of a person speaking you have found that person in the crowded room.
This ICA concept is also known as blind source separation, and this project uses the JADE algorithm for R, to provide the separation matrix of components for the R,G, B, IR mixture of data. The separate components then have their signals extracted using a fast Fourier transform to find a matching frequency range of a heart rate.
Project Source URL: https://github.com/dngoins/Kinectv2HeartRate
Couple other times we've highlighted Dwight's work;