Children’s dance is an important part of quality education for children, and it is also an important channel for children’s physical and mental development. Traditional children’s dance education adheres to the teaching method taught by teachers, and teachers insist on teaching progress. In order to improve the level and quality of children’s dance creation, this paper analyzes the effect of children’s dance creation under video guidance and discusses the integration strategies in children’s dance creation. Children’s dance education in the video teaching environment advocates the combination of online and offline teaching characteristics, and it is necessary to pay attention to the subjective initiative of students. At the same time, how to effectively use video teaching to improve aesthetic ability, effectively summarize online and offline resources to enhance educational efficiency, and make arrangements to strengthen the network. The quality of video-guided dance teaching is an era subject for the development of video-guided learning. Children’s dance education in the video teaching environment needs to focus on the development of teaching forms and the transformation of teaching thinking ability and guide students to strengthen their aesthetic ability in practical activities and teaching methods.

Aiming at the problem that traditional human motion pose estimation methods cannot accurately capture and estimate the movement changes of children dancers, a hierarchical dance pose estimation method for children based on sequence multiscale feature fusion representation is proposed. By comparing the pose feature extraction algorithm with the actual recognition effect, the recognition rates of the dancer’s upper body, infiltration, and whole body have increased by 14.2, 10.6, and 12.6, respectively. The experimental results show that the proposed pose estimation algorithm achieves good pose estimation results on both the standard human pose estimation dataset and the self-built dance dataset.