Multi-Channel Deep Learning Filtering Method for Accurate Heart Rate Detection and Performance Comparison with Existing Technique

Abstract

A multi-channel deep learning filtering method is suggested in this study to make heart rate tracking more accurate. This method uses strong deep learning frameworks. Advanced signal processing methods, such as noise removal, butter bandpass filtering, and peak recognition algorithms, were used on the UBFC-RPPG dataset to get ground truth heart rate signals. After that, frames from the movies in the collection were taken out, resized to standard sizes, and face regions of interest (ROIs) like the left cheek, right cheek, and temples were found. These ROIs were used to figure out heartbeat signals over time. These signals were then sent to mixed deep learning models. Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) networks are combined in the deep learning design to make a mixed system for efficiently extracting time features. As shown by the results, this combination model works better than either LSTM or GRU models used alone. The suggested method is very good at finding heart rate signs, as shown by loss metrics and accuracy graphs. A comparison with other methods shows that this one works better on standard datasets in terms of accuracy and dependability. This multi-channel deep learning filtering method is a big step forward in heart rate recognition, especially when it comes to using deep learning and face video processing to get better results. The method could be used in healthcare tracking systems because it gives a safe and effective way to estimate heart rate without touching the person.