Count: the smart tool on wrist-worn device

The use of wrist-worn devices such as smartwatches is constantly increasing, smartwatches represent the most popular type of wearable device. Previous works used smartwatches for muscle and hand gesture activity. This study focuses on recognizing finger counting activity by using an electrocardiography sensor. This method is based on using Apple’s off-the-shelf smartwatch sensor to collect raw ECG data. These digital signals will be converted to analog signals and saved as images. Then a deep learning algorithm will be used to understand these signal images to build the Conventional neural network (CNN) model. Finally, this work designs a model that recognizes how many fingers were being raised. This study aimed to use deep learning algorithms, specifically CNN. The COUNT model was based on CNN and has seven layers; each layer represents different operations and tasks. Starting with the dataset as an input and resulting with a COUNT-trained model- being built. To implement the COUNT model, ECG data was collected from the user by using Apple’s watch ECG application, then the ECG reading was retrieved from the apple watch by using the health application in the IOS device. ECG data were converted into images by using the stackedplot function in MATLAB then these images were divided into 70% and 30% as training and testing sets respectively, to build the COUNT model. After building COUNT, using MATLAB a new dataset was used after conversion to allow COUNT to classify the input images as 1,2...,5. In order to evaluate the model, there was more than one experiment and different results for each. The best accuracy of COUNT reached about 98% after using the data augmentation method.