In China, the training of adolescent track and field athletes primarily relies on traditional coach-apprentice teaching methods, which are characterized by a certain degree of subjectivity and ambiguity. To provide an objective and quantitative basis for athletic performance evaluation, this study developed a biomechanical analysis system based on machine vision technology and musculoskeletal inverse dynamics analysis. The system uses 100 Hz orthogonal-view calibrated images to synchronously capture athletes’ movements, employs the HRNet model for human pose estimation to extract keypoint positions, and utilizes Gaussian process regression to optimize outliers in pose data, thereby calculating joint velocities and angular velocities. By integrating the collected motion data, inverse dynamics analysis is performed using the lower limb model in the AnyBody modeling system to assess the forces on joints and muscles. This study used the standing triple jump as an example, selecting cases of full-score and non-full-score athletes to test the system’s effectiveness and explore differences in kinematics, dynamics, and muscle activation among athletes. The results showed that in the full-score athlete case, the maximum knee joint flexion angle during the flight phase was 120°, and 60° at landing; in the non-full-score athlete case, the anterior force of the knee joint and the upward force of the ankle joint were weaker during the takeoff phase, and the activation of the knee flexor muscles and soleus muscle was also lower. These findings indicate that the system can enable precise measurement and quantitative assessment of athletic movements, providing coaches and athletes with a scientific basis for training, thereby helping athletes to posture optimization and performance enhancement.