Smartphone Inertial Sensors for Human Locomotion Activity Recognition based on Template Matching and Codebook Generation

In the recent past, recognition of human locomotion activities has become a growing research area. Health monitoring, detection of a crowd’s behavior and indoor-localization are some examples of the fields that benefit from the diversity of this research area. Locomotion activity recognition while using the data from smartphone sensors has become a popular approach because it provides the user-independence. While working with smartphone sensors, where it helps to rectify the annoyance of the subject, it creates difficulty in the recognition of the activity due to smartphone’s possible random orientations. In this paper, a combination of techniques like template matching and codebook generation is proposed that not only eliminates the random orientation factor but also lessens the computational complexity of the activity recognition. Experimental results are evaluated over publicly available benchmark MobiAct dataset which include a combination of smartphone embedded accelerometer, gyroscope sensors and orientation data. Our model has secured an accuracy of 87.50% and 81.73% over static and dynamic activities respectively.