Research on Rehabilitation Exercise Guidance System based on Action Quality Assessment

As a result of the growing participation and awareness in sports, there has been an increasing prevalence of musculoskeletal injuries. Rehabilitation exercises get more and more attention in the treatment because of its affordability and long-term effectiveness. However, many patients opt for home-based rehabilitation due to the prolonged convalescence of rehabilitation exercises for various musculoskeletal diseases, which lacks professional oversight and threaten the safety and effectiveness as a consequence. To address this issue, we developed an intelligent Rehabilitation Exercise Guidance System (REGS), including an assessment module and a consultation module. Built on exercise standards provided by physiotherapists, we compute spatial-temporal indices using Blazepose for skeletal tracking and DTW for comparative analysis, to assess the accuracy, stability, and completeness of patients' exercise quality. In conjunction with the assessment results as the individualized prompt, we fine-tuned the Qwen-14B-Chat model using a dialogue dataset in the consultation module, configuring which with an agent that efficiently filters out irrelevant or invalid queries. The system's spatial-temporal indices calculation exhibits high accuracy when tested on pre-established video and image datasets, verifying its reliability in assessing rehabilitation exercise performance. Moreover, our new chat model surpasses baseline models such as Qwen-14B-Chat and ChatGPT4 in consistency and expertise, reflecting its enhanced capability to deliver specialized guidance on rehabilitation exercises. In summary, the Rehabilitation Exercise Guidance System offers a robust solution to the challenges faced by patients when doing rehabilitation exercise at home, integrating advanced AI techniques with medical expertise to provide accurate, reliable exercise assessment and professional, personal support throughout the recovery process.