2019 · 14 citations · 18 references
ComputingCybersecurityMobile ComputingMovement AnalysisTime Series AnalysisAction RecognitionPattern RecognitionSignal ProcessingSupport Vector MachinesTime Series Feature ExtractionWearable Health TechnologiesActivity RecognitionChannel State InformationClassification AccuracyHuman Activity RecognitionLow-pass FilteringLow-pass FiltersSupport Vector Machine ClassificationWi-fi Channel State Information
Real-time recognition of human activities is an important functionality of smart spaces. It allows a wide range of security and healthcare applications. In this work, we use the Channel State Information (CSI) of WiFi signals to assess the patterns associated with dynamic human activities, including sitting-down and standing-up actions. We preprocess raw signals with both a Hampel filter and low-pass filter. The signals are then segmented into 20-packet labelled sequences. Features including kurtosis, maximum, mean, minimum, maximum peak, skew, standard deviation, and variance are extracted for each sequence, providing feature vectors of 168 variables to enable activity recognition. Features are normalized and a series of classifiers were trained and compared to predict three activity classes: stationary (seated or standing still), sitting-down, and standing-up. Preliminary results on data collected for a single subject achieve a classification accuracy of 98.4% with a medium Gaussian Support Vector Machine (SVM) to distinguish between these three classes.
18
ADASYN: Adaptive synthetic sampling approach for imbalanced learning
Haibo He, Yang Bai, Edwardo A. Garcia et al. · 2008 · 4.4K citations
Understanding and Modeling of WiFi Signal Based Human Activity Recognition
Wei Wang, Alex X. Liu, Muhammad Shahzad et al. · 2015 · 1.1K citations
Yan Wang, Jian Liu, Yingying Chen et al. · 2014 · 805 citations
Tracking Vital Signs During Sleep Leveraging Off-the-shelf WiFi
Jian Liu, Yan Wang, Yingying Chen et al. · 2015 · 479 citations