Multi Sensor System for Pedestrian Tracking and Activity Recognition in Indoor Environments

• Main Author: Marron Monteserin, Juan Jose
• Format: Others
• Published: Scholar Commons 2014
• Subjects: Inertial Navigation; Pervasive Computing; Sensor Fusion; Smartphones; Ubiquitous Localization; Computer Sciences; Engineering
• Online Access: https://scholarcommons.usf.edu/etd/5068; https://scholarcommons.usf.edu/cgi/viewcontent.cgi?article=6264&context=etd

The widespread use of mobile devices and the rise of Global Navigation Satellite Systems (GNSS) have allowed mobile tracking applications to become very popular and valuable in outdoor environments. However, tracking pedestrians in indoor environments with Global Positioning System (GPS)-based schemes is still very challenging given the lack of enough signals to locate the user. Along with indoor tracking, the ability to recognize pedestrian behavior and activities can lead to considerable growth in location-based applications including pervasive healthcare, leisure and guide services (such as, museum, airports, stores, etc.), and emergency services, among the most important ones. This thesis presents a system for pedestrian tracking and activity recognition in indoor environments using exclusively common off-the-shelf sensors embedded in smartphones (accelerometer, gyroscope, magnetometer and barometer). The proposed system combines the knowledge found in biomechanical patterns of the human body while accomplishing basic activities, such as walking or climbing stairs up and down, along with identifiable signatures that certain indoor locations (such as turns or elevators) introduce on sensing data. The system was implemented and tested on Android-based mobile phones with a fixed phone position. The system provides accurate step detection and count with an error of 3% in flat floor motion traces and 3.33% in stairs. The detection of user changes of direction and altitude are performed with 98.88% and 96.66% accuracy, respectively. In addition, the activity recognition module has an accuracy of 95%. The combination of modules leads to a total tracking error of 90.81% in common human motion indoor displacements.