8.2.1. Data Collection

Several experiments were carried out for collecting an adequate amount of data with ground truth labels to train and test the designed motion mode classifier. Data from four adults (two males and two females) of different heights and weights were collected to evaluate the performance of the proposed classifier. The data collections were conducted in an open field, and each participant was equipped with a chest-mounted IMU device. The participants were requested to walk approximately 500 m of distance while performing several activities. The experiments were performed using a predefined protocol that consisted of six activities including static, walking, turning right, turning left, opening a door, bending, and random movement. As the data were collected from an open field, the participants were asked to simulate the door-opening activity. To complete one run of the defined protocol, a participant required four minutes, and each participant repeated the protocol three times. Thus, enough data was collected for training and testing of the classifier.

To facilitate the labelling of ground truth, another IMU device, named NGIMU (https://x-io.co.uk/ngimu/ (accessed on 10 February 2021)), was utilised. Pressing the power button of the device while powered on will send a message to the receiver with timestamped information for the button-pressing event. During the experiment, the participants were asked to press the button of the NGIMU each time they began an activity. Thus, by annotating the exact number and sequence of performed activities and mapping this information with the timestamped button event's data, the ground truth labelling was

performed. For precise evaluation of the classifier, the data recorded during the transition of two activities were removed manually.
