In order to generate 3D virtual human animation with smoother actions and richer and more vivid expressions, this article proposes a system based on dual cameras to synchronously capture human facial expressions and human posture, so as to better generate 3D virtual human animation. Firstly, the 2D features of the image are extracted through facial expression, and the benchmark is provided for the standard 3D coordinate values. At the same time, the 3D pose fusion is realized by using the pose measurement algorithm for human motion capture. Through the test, it is found that the two fusion algorithms can meet the requirements of the static accuracy of 0.5° and the dynamic accuracy of 0°, and the three-dimensional model of a virtual human body has a good effect in data-driven animation. The experiments show that the system can better capture facial expressions and human posture synchronously, and the frame rate in the experimental test can reach 20 fps, which can generate more natural and realistic 3D virtual human animation in real time.