This paper proposes CARM, a CSI-based human activity recognition and monitoring system. CARM uses two theoretical models: a CSI-speed model that quantifies the correlation between CSI value dynamics and human movement speeds, and a CSI-activity model that quantifies the correlation between movement speeds of different body parts and specific human activities. These models allow CARM to quantitatively build the correlation between CSI value dynamics and specific human activities. CARM uses this correlation as a profiling mechanism to recognize activities by matching them to the best-fit profile. CARM was implemented using commercial WiFi devices and evaluated in multiple environments, achieving an average accuracy of over 96%. The system addresses several technical challenges, including estimating human movement speeds from CSI values, building robust CSI-activity models for different humans, denoising CSI values, handling carrier frequency offset, and detecting the start and end of human activities. CARM also introduces a PCA-based CSI denoising scheme that leverages the correlation between CSI streams to improve signal quality. The system uses HMMs to model human activities and extract features from CSI values using DWT. CARM achieves high accuracy in recognizing human activities, even in new environments and for new users. The paper also discusses the theoretical foundations of CSI-based human activity recognition, including the relationship between CSI value dynamics and human movement speeds, and the use of time-frequency analysis tools to separate movement components. The results show that CARM outperforms traditional methods in terms of accuracy and robustness.This paper proposes CARM, a CSI-based human activity recognition and monitoring system. CARM uses two theoretical models: a CSI-speed model that quantifies the correlation between CSI value dynamics and human movement speeds, and a CSI-activity model that quantifies the correlation between movement speeds of different body parts and specific human activities. These models allow CARM to quantitatively build the correlation between CSI value dynamics and specific human activities. CARM uses this correlation as a profiling mechanism to recognize activities by matching them to the best-fit profile. CARM was implemented using commercial WiFi devices and evaluated in multiple environments, achieving an average accuracy of over 96%. The system addresses several technical challenges, including estimating human movement speeds from CSI values, building robust CSI-activity models for different humans, denoising CSI values, handling carrier frequency offset, and detecting the start and end of human activities. CARM also introduces a PCA-based CSI denoising scheme that leverages the correlation between CSI streams to improve signal quality. The system uses HMMs to model human activities and extract features from CSI values using DWT. CARM achieves high accuracy in recognizing human activities, even in new environments and for new users. The paper also discusses the theoretical foundations of CSI-based human activity recognition, including the relationship between CSI value dynamics and human movement speeds, and the use of time-frequency analysis tools to separate movement components. The results show that CARM outperforms traditional methods in terms of accuracy and robustness.