Motion perception, pivotal to myriad specialized tasks, necessitates the enhancement of proficiency through sustained repetition of perception-action cycles to meet standard benchmarks. Attaining advanced skill levels often demands additional practice. However, traditional pedagogical and evaluation systems predominantly hinge on the subjective experiences of instructors and evaluators. This dependence precipitates two principal challenges in the domain of motion perception-related professional work. Firstly, learners grapple with securing timely, adequate guidance during learning and practice, given the slow, trial-and-error nature of key point acquisition. Secondly, objective evaluation, fraught with instability, fails to consistently deliver accurate quantitative assessments, thereby adversely impacting the learning process. In response to these challenges, this study introduces a deep learning-based approach for standardized evaluation and human pose estimation. The methodology begins with the utilization of OpenPose for body joint detection. This is followed by a Deep Neural Network (DNN)-informed strategy for posture information extraction. Lastly, leveraging our team's extensive experience in dance instruction, a novel method for describing and discerning differences in dance movements is proposed. This approach enables a quantitative evaluation and provides intuitive feedback on the mechanics of dance movements, thereby enhancing the monitoring of participants' progress. Validated experimentally, the proposed methodology demonstrates precision in motion perception and quantitative evaluation. It not only offers practical guidance for enhancing the quality of dance instruction but also provides a valuable reference for other applications.