Personal navigation using novel methods of human motion modeling

The widespread use of smartphones and other personal devices provides a low cost sensing platform giving easy access to a variety of data. The availability of this rich data provides opportunities to develop new applications for personal use in areas such as health monitoring, situational awareness, and location-awareness. Of these, personal navigation and localization is of rapidly-growing commercial interest. There have been considerable research efforts to improve navigation capabilities using the embedded inertial, optical, and magnetic sensors in personal devices. This spans GPS augmentation, inertial and vision based solutions, map matching, and other sensor fusion approaches. One emerging method is to improve contextual awareness by detecting and classifying relevant human motions. This may be done by building human locomotion models primarily based on inertial and magnetic data. Once reliable models are constructed, they can be calibrated to a motion's magnitude and frequency. The derived information can then be integrated into the navigation solution; improving performance in indoor and other GPS challenged navigation environments. A case application is a human motion aware advanced pedometer. Several methods of dynamically modeling human motions have been proposed in literature. Each method has constraints and often non-obvious drawbacks. This paper first provides a survey of existing methods along with important but often overlooked details. Although the processing power of small personal devices is quickly growing, the computational load for real-time applications is still a constraint. Therefore, an evaluation of these methods based on their computational cost of reliable performance is provided. Finally, a case study with field test results will be presented. Three motions states were chosen for field tests; walking forward, walking backward, and running. Conclusions regarding suitability of personal navigation will be presented.