Pedestrian Dead Reckoning Based on Complex Motion Mode Recognition Using Hierarchical Classification

The pedestrian positioning algorithm based on inertial measurement unit (IMU) has been widely applied in scenarios such as firefighting and counterterrorism where global navigation satellite system (GNSS) performance is limited. However, most existing systems do not take into account crawling-type motions. Additionally, the zero velocity update error constraint method often used in pedestrian navigation is not applicable to crawling-type motions. As a result, errors introduced by crawling-type motions become the primary source of error in complex motion scenarios. In order to enhance pedestrian positioning accuracy during complex motion modes, this study analyzed data across various motion modes and developed specific positioning algorithms based on a hierarchical classification. First, we observe that data patterns for standing mode and crawling-type mode are distinct based on the analysis of motion mode. Therefore, to improve step detection accuracy, support vector machine (SVM) is employed to classify motion modes into three coarse categories: standing mode (forward/lateral/backward walking, and running), crawling mode (baby/military crawling), and one side crawling mode (one side military crawling). Specific step detection algorithms are designed for each category. Furthermore, the step lengths of standing motions and crawling-type motions exhibit relatively strong correlations with data characteristics and height, respectively. Hence, different step length models are designed. Fine classification of motion modes is accomplished using the minimally random convolutional kernel transform (MINIROCKET), optimizing step length model parameters and heading correction for various fine motion modes. Finally, precise position estimation is achieved. Experimental results demonstrate that this algorithm exhibits stronger adaptability and superior performance in complex motion scenarios.