Deep Learning-based Action Recognition for Pedestrian Indoor Localization using Smartphone Inertial Sensors
Sakorn Mekruksavanich, Ponnipa Jantawong, Anuchit Jitpattanakul · 2022 Joint International Conference on Digital Arts, Media and Technology with ECTI Northern Section Conference on Electrical, Electronics, Computer and Telecommunications Engineering (ECTI DAMT & NCON) · 2022
For decades, the identification of human action has been one of the most critical study areas in artificial intelligence research. The study in human activity recognition aims to classify people's behavior based on data that is already accessible, such as pictures, visual, sensing, or stream recorded data, among other things. A common distinction made by most human activity recognition methods is between daily tasks such as walking and specific behaviors like standing or sitting. More specialized functions targeted to the use-case of pedestrian interior navigation are investigated in this study. These activities include several types of stationary and movement activities and various other activities. A deep learning-based human activity recognition framework is presented that can be used to recognize people's actions inside an indoor localization situation by analyzing smartphone inertial sensor data. Furthermore, we introduced a 1D-ResNet-SE model in the framework and assessed it using a publicly available dataset known as the Activity Logs dataset. According to the experimental measurements, the 1D-ResNet-SE surpasses other standard deep learning techniques in terms of accuracy, with a maximum accuracy of 84.11% in the general scenario.