Smart Contactless Vital Monitoring System Using Mobile Camera and Smart AI Coach

Abstract

As an AI intern, I was associated with Teksun Microsys and was part of two projects: 1. Smart Contactless Vital Monitoring System Using Mobile Camera and 2. Smart AI Coach. Vital signs monitoring is important for clinical diagnostics and in-home health monitoring. It is measurements of the basic functions of the human body. Medical professionals monitored four main vital signs routinely as Temperature, Pulse Rate, Respiration Rate, Blood Pressure. Vital signs such as heart rate (HR), HR variability (HRV), and respiratory rate (RR), are usually measured with non-invasive electrocardiography (ECG) or photoplethysmography (PPG) sensors in clinical examination or with commercial wearable devices in health monitoring. The measurements in both scenarios often employ contact sensors, which may be inconvenient or cause discomfort in long-term monitoring sessions. For example, it is hard to put sensors on young children and ask them to keep them still during the monitoring session. Some pioneering works reveal a possible approach of remote vital signs measurement with contactless sensors. Extracted HR from face videos, based on the small colour change on the face that is consistent with the pulse signal. This technology is called remote PPG (rPPG). We are using this technique in our project. We are using the rPPG signal and extract the values of the RGB signal after applying some signal processing and based on that information we predict the values of HR, RR and SpO2. Smart AI Coach: A fitness trainer leads or assisting a client to reach their physical fitness goal. Trainer determines the clients’ fitness level, individual goal, skills, develop training program according to clients’ need and also, monitor the progress. To accomplish all tasks there need to trainer and client present physically, which is not possible for both client and trainer. Smart AI Coach overcomes this issue and helps a client to get online training. AI coach useful to a user so that s/he do not need to go the gym and guided at home only to perform an exercise, also, no need to keep count of sets and measure correct posture thus avoid injuries. We trained a CNN model for 21 different exercises, for that we have created our dataset and used it in our model. We faced so many challenges for exercise which has the same posture while doing exercise, its result comes in misclassification. To resolve this misclassification we used the LSTM algorithm and we got good accuracy with accurate classification.