Human Activity Classification in Smartphones using Shape Descriptors

This paper presents a shape descriptor-based approach to human activity classification in devices such as iPod Touch, smartphones, and other similar devices. In this work, signals acquired from the built-in accelerometer and gyroscope sensors of iPod Touch are analyzed to recognize different activities performed by a user. In order to extract the discriminative information, shape descriptor-based features are computed from the captured signals. These features are then normalized and concatenated to form a consolidated feature vector. To recognize an activity performed by the user, k-nearest neighbor classifier is employed. The proposed approach is evaluated on the publicly available dataset namely, physical activity sensor data. Our experimental results demonstrate the effectiveness of the proposed shape descriptors for activity classification. Additionally, the experimental results on the aforementioned dataset show significant improvement in classification accuracy as compared to the existing work.