This paper proposes a method for data acquisition of yoga movements and analyzes it using the Fourier transform. It aims to help residents choose a more scientific and effective exercise method to relieve physical fatigue, reduce body stiffness, and improve body flexibility. Initially, we transform the yoga movements into RGB and optical flow, feeding them into the STF-ResNet network to extract the temporal and spatial features of the movement data. Next, we combine the spatial and temporal flow features with residuals to compensate for the loss of high-level feature information. Finally, we add the convolutional block attention module to further filter the yoga movement features from both channel and spatial dimensions. Next, we introduce the Fourier transform to parse the gathered yoga movement features. We conducted experiments both with and without yoga movements interfering with body soft rhythms to investigate their effects. The experimental group improved the test scores of the forward and backward splitting test and the backward bending body bridge building test, with the mean value of standing forward bending being 5.11 higher after the experiment than before. It shows that yoga intervenes in the flexibility of the human body through auxiliary training, and yoga postures can have a significant beneficial effect on body flexibility.