Performing repeated ultrasound scans on human limbs requires a high degree of precision and trajectory consistency between iterations. Because human operators often struggle to achieve the exact reproducibility necessary for these tasks, robotic systems can be utilised to improve consistency. A significant technical challenge involves determining how to define the robotic path and the appropriate contact pressure for specific points on the limb.
This project addresses the first phase of this problem by developing a pipeline to accurately define the trajectory. Methods such as YOLO, MediaPipe, and IR tracking were evaluated to determine their capability for tracking the index finger of a human hand with high precision. With these results, a pipeline has been designed and developed to efficiently run a tracking algorithm and add necessary information to the trajectory, such as reference points, and a force estimation of an existing estimation network called PressureVision from Grady et al. Extensive testing was conducted to ensure the system achieves the required tracking precision, maintaining a maximum error margin of 1 cm along the trajectory.
With the final system, a baseline has been set for the Master's Thesis project, which will investigate the accurate estimation of force and stiffness from a human finger pressing on the forearm using RGB-D images solely. This information can subsequently be utilized by a robotic system to execute high-precision, repeated ultrasound scanning tasks.
