Smartphone-based Walking Speed Estimation in the Context of Mindful Walking

With smartphones being more widespread and having more users than ever, the possibilities of using them for mHealth purposes are on the rise. Since users carry their smartphone on them constantly, they are the ideal tools for non-invasive fitness and health monitoring. The main focus in this thesis being the walking speed which is a
general indicator of health and can be seen as a vital sign. Walking speed can for example show mortality, high stress levels or diseases such as depression. One application aiming to counteract such conditions is mindful walking. The application monitors the walking speed of users and through haptic feedback reminds them to keep a steady pace which is below a predefined limit. This thesis serves as a framework for state-of-the-art walking speed estimation methods in the setting of free outdoor walking with a live speed monitoring, only using a smartphones’ internal sensors. Three general methods were used: GPS, step counting and raw sensor data in combination with machine learning. Due to the known inaccuracies of GPS in certain locations, the measurements were evaluated in the following four locations: city centre, forest, suburbs and open field. To be able to use the step counting method, first, the step length of each participant has to be determined which also tests two methods. The first one is an estimation by height and gender, the second one uses GPS and step counting over a predefined distance. To evaluate all these methods and find the most precise one, a series of experiments was done. Overall, GPS delivers the worst accurate walking speed estimation with an average root mean square error of at least 2.429 km/h doing the manual calculation and 1.126 km/h extracting the speed directly from the GPS signal. The method utilizing machine learning offers a root mean square error of 1.143 km/h, but shows possible ways of improvement. The most accurate walking speed estimation method for both Android and iOS uses step counting and has a root mean square error of 1.084 km/h and 1.014 km/h, respectively.