This paper analyzes the influence of “the concept of integration of body and health” on the physical development of young children and proposes that scientific sports are an important way to intervene in the poor physical appearance of young children. In order to facilitate and accurately identify young children with poor posture, the boundary tracking algorithm is used to detect the human body contour, and based on the physical characteristics, a human posture recognition algorithm based on multi-feature fusion and image similarity is proposed. To model common toddler body postures and obtain toddler posture features, a star model is employed. Combined with the advantage of the SVM classifier, the principal component analysis algorithm is used to design a classification system for abnormal behaviors of toddlers, which combines the static and dynamic posture data images of toddlers to detect abnormalities of toddlers’ body postures. When the difference between the angle of the left elbow to the left shoulder and the angle of the right elbow to the right shoulder to the left shoulder is about 16.58°, the toddler has a high-low shoulder posture. The detection accuracy of the method in this paper meets the needs of young children’s posture detection, and the misjudgment rate reaches 10.25%, which can assist in detecting the abnormalities of young children’s posture and facilitate the teachers and staff to carry out the sports related to young children’s posture.