Embedding an Accelerometer-Based Heart Rate Evaluation Algorithm in a Smart Eyewear

We present a study dedicated to refining and optimizing algorithms for efficient estimation of heart rate. To this end, we apply seismocardiography techniques on inertial measurement unit signals collected from a proprietary smart eyewear device. Measurements were performed on 50 subjects, for different static activities. The wearers were equipped with electrocardiogram, photoplethysmogram and smart eyewear inertial measurement unit data sensors, connected to a computer for data acquisition. The sampling frequency and sensitivity of the accelerometer has been configured to fit the cardiac frequency characteristics. As expected, the signals were characterized by significant levels of noise and difficult to be exploited directly by peak detection. Therefore, the accelerometer signals should be processed to reduce the body artifacts effects and to focus on the cardiac activity. For this purpose, a Radix-2 type algorithm has been preferred to mitigate excessive degradation of the accelerometer signals, improve processing speed, and minimize memory usage. With a reporting coverage of 74.13%, the optimized embedded algorithm achieved 96.56% of accuracy (at ±5 bpm). The algorithm was embedded into the proprietary low power smart eyewear after significant speed and memory footprint improvements without compromising accuracy. The total processing duration is completed in just 1 second. Deployment on different embedded systems and inertial motion units showed consistent performance. We have developed a very promising measurement system for a permanent wearer, complying with the constraints of an eyewear. This technology opens new opportunities for cardiac tracking and new fields of investigations, especially for cardiovascular diseases detections.