Human Moving Behavior Recognition Based on the D-S Evidence Theory and SVM utilizing Smartphone

With the popularization of smartphones, the technology of moving human behavior recognition utilizing smartphones has drawn wide attention of scholars. However, due to the limited computing power of smartphones, it is difficult for them to stably run recognition systems with highly complex. In addition, when a behavior switches, the transitional action is also difficult to recognize. To address these defects, this paper first proposes a feature extraction method using gait periodicity to extract characteristics of moving human behavior quickly and effectively. Secondly, a recognition framework combining SVM and the D-S evidence theory is proposed to recognize transitional action. A series of experiments and their results show that the both proposed methods can improve accuracy rate of moving human behavior recognition to 95.5%.