DEVELOPMENT OF MULTI-CAMERA SYSTEM FOR EVALUATING REPROJECTION ERROR IN 3D MARKER ESTIMATION ON SIMPLE MOTION CAPTURE SYSTEM

• Main Author: Ananta, Moses
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/75264
• Institution: Institut Teknologi Bandung

With the increasing advancement of artificial intelligence (AI)-based technology, the production cost and complexity of creating a motion capture system have become cheaper, simpler, and more accessible for small content creation industries. However, because the motion estimation process of such AI-based motion capture systems is not marker-based, unlike conventional motion capture systems, and only estimates the key points of an actor's body based on visible body parts, the accuracy of AI-based motion capture systems can become an issue for industries that require precise results without incurring significant costs like conventional motion capture systems. Based on this issue, a cost-effective and simple version of a conventional motion capture system will be developed, which utilizes color-based markers and multiple cameras to capture the motion of actors. The focus of this final project is to build a multi-camera system that will be used to estimate the 3-dimensional coordinates of detected markers using triangulation techniques. The system's performance will be evaluated using the reprojection error metric. The constructed multi-camera system is tested with 4 different inter-camera rotation ranges, and from the experiment's results and analysis of the system, it is found that the reprojection error can be used as an indication of inaccurate 3D estimation results. Rotation does not affect the reprojection error, and triangulation with 3 cameras simultaneously produces the lowest reprojection error and the best 3D results when all cameras are synchronized in capturing images.