Data collection for HAR and/FER is a challenging task for many reasons, including privacy. As a result, open source quality datasets are limited in number. For the HAR implementation in use case one, we are using a very popular and open source Wireless Sensor Data Mining (WISDM) lab dataset . The dataset consists of 54,901 samples collected from 36 different subjects. For privacy reasons, usernames are masked with ID numbers from 1-36. The data was collected for six different activities undertaken by the subjects: standing, sitting, jogging, walking, going downstairs, and climbing upstairs. The dataset contains three-axis accelerometer data with more than 200 time steps for each sample. The following screenshot is a sample of the dataset:
For the FER-based emotion detection in use case two, we used two different datasets. The first one is the popular and open...
