Novel Step Detection Algorithm for Smartphone Indoor Localization Based on CEEMDAN-HT

Current smartphone-based step detection algorithms do not comprehensively consider the randomness of user movements in daily life, such as carrying a smartphone in various postures, false walking, and walking in various gaits. To overcome the problem, this article proposes a novel step detection algorithm based on complete ensemble empirical mode decomposition with adaptive noise Hilbert transform (CEEMDAN-HT). First, the CEEMDAN partial reconstruction (CEEMDAN-PR) algorithm is presented to reduce noise for inertial data. In order to compensate for the step count misdetection due to the mix-postures. Second, we combine support vector machine (SVM) with CEEMDAN energy entropy for accurate identification of user movement state and smartphone-carrying posture. This approach addresses the issue of false walking and selects appropriate step detection strategies for different scenarios. Finally, HT is used to obtain instantaneous phase information from inertial data for step detection. This approach is capable of adapting to mixed gaits and mixed postures without complex threshold settings. Experiment results show its superior performance, achieving 98.10% average accuracy in free walking and 93.85% in false walking scenarios.