A WiFi-Based Method for Recognizing Fine-Grained Multiple-Subject Human Activities

Device-free human activity recognition (HAR) has gained attention in recent years. While much has been done in coarse-grained HAR, the recognition of fine-grained human activities is still a research challenge. In this article, we present a novel method to combine channel state information (CSI) and received signal strength indicator (RSSI) signals at the feature level to improve the performance of device-free fine-grained HAR using WiFi data. We extract seven CSI and three RSSI non-segmented frequency-domain features, 12 segmented time-domain features, and five segmented frequency-domain features to select the feature set. We evaluate our method using a dataset containing 12 human-to-human fine-grained interactions. We utilized various classification methods like support vector machine (SVM), Gaussian-Naive-Bayes (GNB), decision tree (DT), logistic regression (LR), linear discriminant analysis (LDA), K-nearest neighbors (KNN), and random forest (RF) using the feature set as input. Our evaluation result yields 94.16% of accuracy, 94.3% of precision, 94.24% of recall, 94.13% F1-score, 93.18% of k-score, and 95.91% area under the curve (AUC) in recognition of seven human-to-human interactions using RF.

Automated summary

In this article, a novel method combining channel state information (CSI) and received signal strength indicator (RSSI) signals is proposed to improve the performance of device-free fine-grained human activity recognition (HAR) using WiFi data. The method is evaluated using a dataset of 12 human-to-human fine-grained interactions and various classification methods. The results show that the method achieves high accuracy, precision, recall, F1-score, k-score, and area under the curve (AUC) in the recognition of seven human-to-human interactions using random forest (RF).