We address the problem of estimating the physical burden of a human body. This translates to monitor and estimate muscle tension and joint reaction forces of a musculoskeletal model in real-time. The system should minimize the discomfort generating by any sensors that needs to be fixed on the user. Our system combines a 3D pose estimation from vision and IMU sensors. We aim to minimize the number of IMU fixed to the subject while compensating the remaining lack of information with vision.