The recognition of human gait states holds paramount importance for disease diagnosis and the formulation of rehabilitation protocols. Compared to traditional methods, Wi-Fi-based gait recognition techniques offer the benefits of non-invasiveness and preservation of privacy. Nevertheless, owing to the rapid transitions in gait states during human walking, prevailing gait recognition methodologies relying on Wi-Fi Channel State Information (CSI) fall short in achieving fine-grained gait state recognition, thereby impeding their applicability in domains with stringent granularity requirements, such as medical care. In this paper, we introduce ID-Gait, a precise gait state recognition system aimed at detecting and characterizing human gait with fine-grained sensing capabilities. ID-Gait first undertakes a broad recognition of the user’s gait state (e.g., fast walking, slow walking, running), and subsequently perform fine-grained state recognition within the gait cycle to monitor the user’s health. Specifically, we initially enhance the Wi-Fi CSI data through quantization of noisy signals, followed by the estimation of gait speed and fine-grained gait state recognition facilitated by the extraction of Doppler Frequency Shift (DFS). Extensive experimentation conducted across diverse scenarios of 5 users demonstrates that the proposed ID-Gait method achieves an average recognition accuracy exceeding \(85\%\) .