Traditional dance drama creation mainly relies on choreographers’ experience, which has problems such as limited innovativeness and low efficiency of movement design. This study explores the use of artificial intelligence image processing methods to enhance the creation of dance drama. A binocular stereo vision depth motion camera is used to capture the dance performer’s movements, combined with wavelet transform for multi-scale filtering of dance video images, background subtraction is applied to extract dance movement features, and a dance automatic generation system based on Seq2Seq model is constructed. The system takes audio features as input, and realizes the matching generation of dance and music through the generator, discriminator and self-encoder working together. The experimental results show that the joint angle changes of the dance movements generated by this method are smoother, and the average value of knee pitch angle is 1.315 rad, which is only 1.15% different from the initial set value of 1.30 rad; in the objective evaluation index, the method achieves a beat coverage rate of 26.4%, and a diversity index of 42.1, which is significantly improved compared with the traditional method; in the subjective evaluation, 15 subjects showed a positive response to this method’s realism and musical style consistency of the generated dances were highly evaluated by 15 subjects. The study proves that the artificial intelligence image processing method can effectively improve the stability, authenticity and innovation of dance movements in the creation of dance dramas, providing a new technical path for the creation of modern dance dramas.
