Project Description Head injuries are one of the most severe incidents with a high death rate. Finite element (FE) models are powerful tools to study head injury mechanisms, and accurate loading inputs such as velocity, acceleration, and angular velocity are critical for reliable injury predictions. Markerless motion capture system based on pose estimation algorithms offers solutions for obtaining these biomechanical parameters; especially the contactless tracking in the natural environment minimizes the interference to the subjects being tracked. Current pose estimation methods, for instance, OpenPose, DensePose, and Alphapose, take images as input and predict 2D joints locations. The predicted 2D joints will guide the 3D reconstruction by triangulating from multiple corresponding cameras. The accuracies of the predicted 2D joints have extensively been evaluated by human-labeled data. However, the 3D prediction results have rarely been evaluated. Furthermore, it is unclear whether such methods are feasible to be used to estimate biomechanical parameters during fast sports. The goal of this project is to develop a multi-camera markerless capture system that is applicable for reconstructing 3D human motion and then validate its accuracy using a marker-based motion capture system as ground truth. Method Two subjects will perform tackle with soft contact at the end to avoid injuries and to punch the HIII dummy head with gloves on. During the data collection session, the subjects will wear necessary protective equipment. The experiments will be conducted in KTH MoveAbility Lab (Dept. of Engineering Mechanics, SCI) and data will be collected simutaneuously with markerless multi-view cameras and marker-based 3D motion capture systems (VICON). Ethical permit ethical application for data collection is under preparation. Expected Outcome The expected outcome of this project is a validated markerless motion capture system that can be used for studying the biomechanical mechanisms of head injuries. This can be achieved through the following steps. First, the markerless motion capture system will be developed. The system will take images as inputs and predict 2D joints locations. Multiple cameras will be used to triangulate the 3D positions of the predicted 2D joints. To validate the accuracy of the proposed markerless motion capture system, the predicted 3D positions will be compared with the ground truth 3D positions from the marker-based motion capture system. The comparison will be evaluated in terms of both accuracy and robustness. Second, the proposed markerless motion capture system will be used to estimate the biomechanical parameters during the experiment. The estimated parameters will be compared with the ground truth parameters from the marker-based motion capture system. The comparison will be evaluated in terms of accuracy and robustness. Finally, the feasibility of using the proposed markerless motion capture system to study the biomechanical mechanisms of head injuries will be discussed.