Wearable Sensors-based Human Locomotion And Indoor Localization with Smartphone
Mehrab Rafiq, Ahmad Jalal
Abstract
Indoor localization is a hot topic due to the rising demand for location-based services and smartphone use. Actions in indoor spaces provide valuable information that systems can use to verify user's relative location within a building. Inertial sensors, like GPS, accelerometers, and gyroscopes, originally designed to enhance device features, now serve different purposes. Recognizing human locomotion is crucial in fields like robotics, security, and medicine. This paper presents an enhanced technique that leverages GPS, accelerometer, and gyroscope data from the Indoor Positioning and HAR Datasets to recognize and localize human movement. The system preprocesses input signals with a Chebyshev Type 1 filter and segments them using a Hamming window. Feature extraction occurs in two blocks: Human Locomotion and Localization. Techniques used include FFT, SSCE, DTW, and LPCC for locomotion, while localization relies on step count and step length. Features are optimized with Kernel Fisher Discriminant Analysis (KFDA), and classification is done using a Fuzzy Entropy Classifier (FEC). The method achieves an accuracy of 0.90% with the HAR dataset and 0.85% with the Indoor Positioning dataset.