الفهرس | Only 14 pages are availabe for public view |
Abstract In this thesis, we propose an inertial HAR system to recognize complex human activities using motion primitives. Inertial sensor readings are segmented into set of finite motion primitives, then recognized motion primitives are used to classify the complex activity. Complex activities are classified using 2-level hierarchical HMM classifier. We introduce a motion primitive generation algorithm that extracts most distinct time-series segments from a set of complex activities. We also apply three different features selection approaches to reduce the processing time. SBHAR and PAMAP2 public datasets are used to evaluate the system’s performance, where we show that our approach achieves 93.77% and 86.84% accuracies respectively. A comparison with related researches which used the same datasets is conducted to compare our results regarding methodology, features, accuracy, time complexity and classification rate |