This paper introduces JCDR-TCN, a lightweight WiFi-based gesture recognition model that captures dynamic CSI variations from a first-person perspective by integrating channel dimensionality reduction and temporal convolutional networks. A Raspberry Pi 4B equipped with Nexmon firmware is employed to collect raw CSI data, which is then processed and classified into specific gestures by the proposed JCDR-TCN. Experimental results indicate that JCDR-TCN achieves a recognition accuracy of 95.63\%, surpassing conventional CNN, LSTM, and CRNN models. Furthermore, the model maintains approximately 50K parameters and requires 8.38M floating-point operations, demonstrating a favorable balance between accuracy and computational efficiency, thereby making it well-suited for embedded and mobile deployment.