Focus on the User: A User Relative Coordinate System for Activity Detection

The information about a person's activity, as a special type of contextual information, can be used for a multitude of applications. Thus, activity detection is a focus in research. For activity detection, usually measured sensor data are processed and matched to known activities. Unfortunately, measured sensor data can be inconsistent for the same activity. Depending on the device orientation and in which direction of compass the user is heading (while carrying out an activity) the sensor data changes. As a result, expected values on one sensor axis appear on another one and it is challenging for a pre-generated activity detection model to match the sensor data to the activity. However, the intuitive solution seems to be easy: focus on the user. Regardless of the device orientation or the cardinal direction, the activity never change for the user. In this paper, we present an approach to focus on the user in activity detection. We convert sensor data into a representation relative to the user. Thus, the sensor data stay consistent regardless of the device orientation or the cardinal direction. We show that by using the focus on the user approach the detection rates increase up to 21.7%. Also, the detection is more reliable (lower standard deviation). We prove the reliable activity detection for different device orientations, cardinal directions, types of algorithms, sensor sets, activities, and sensor positions.