This paper introduces an analytical study on generating and classifying millimeter-wave frequency modulated continuous wave (mm-Wave FMCW) radar systems operating in the 77-81 GHz frequency range, utilizing Synthetic Aperture Radar (SAR) imaging techniques for human gesture such as sitting, standing, sleeping, falling, and walking for detection. In order to recreate and enhance realistic radar images, stickman images collected from online sources are used as input for creating a dataset of 5,000 SAR reconstructed images with the help of data agumentation technique and Conditional Generative Adversarial Network (CGAN). The generated dataset is classified using the EfficientNet architecture, achieving overall validation accuracy of 92.34% in recognizing human gestures. This method effectively addresses data scarcity, enhances SAR image quality, and delivers reliable classification accuracy, offering significant potential for applications in gesture detection, surveillance, and related fields.