Human activity recognition using tag-based radio frequency localization

This article provides a comparative study on the different techniques of classifying human activities using tag-based radio-frequency (RF) localization. A publicly available dataset is used where the position data of multiple RF tags worn on different parts of the human body are acquired asynchronously and nonuniformly. In this study, curves fitted to the data are resampled uniformly and then segmented. We investigate the effect on system accuracy of varying the relevant system parameters. We compare various curve-fitting, segmentation, and classification techniques and present the combination resulting in the best performance. The classifiers are validated using 5-fold and subject-based leave-one-out cross validation, and for the complete classification problem with 11 classes, the proposed system demonstrates an average classification error of 8.67% and 21.30%, respectively. When the number of classes is reduced to five by omitting the transition classes, these errors become 1.12% and 6.52%, respectively. The results indicate that the system demonstrates acceptable classification performance despite that tag-based RF localization does not provide very accurate position measurements.