This paper proposes a scoring system to evaluate the correctness of movements performed by trainees during single and parallel bar exercises. First, the system employs a two-step multi-view 3D human pose estimation algorithm to capture the 3D key point coordinates of all individuals in the training scene. The process begins with 2D pose estimation for each individual view, followed by cross-view matching and subsequent 3D reconstruction. Next, a feature selection strategy is introduced to extract the 3D pose information of a specific individual (the trainee) based on his positional features and movement characteristics. The system then assesses and scores the trainee’s 3D poses according to standard action scoring criteria. In multi-person scenarios, the proposed scoring system demonstrates high accuracy in pose estimation and in selecting and scoring the trainees’ movements.