Upper Limb Position Sensing: A Machine Vision Approach

Abstract

Numerous approaches to sensing limb position for controlling neural prostheses have been proposed, evaluated and even incorporated into commercial products. In general, these technologies have focused on the goals of accuracy, convenience and cost. Here we propose an approach to sensing upper limb posture for a stroke rehabilitation system that does not require any devices attached to the subject This is achieved through the use of a machine vision approach, which involves focusing a digital video camera on the subject and processing the video stream using a specialized algorithm running on a PC. This algorithm will produce a trigger signal whenever the arm posture conforms to a predefined profile. While the approach itself can be applied to a variety of sensing and control applications, we have demonstrated it by developing and characterizing an algorithm that can accurately sense elbow flexion and extension. The machine vision algorithm performs 3-D recovery of the arm position and calculates the elbow angle accordingly, which we have compared to a commercially available goniometer. It also involves a model based prediction and correction technique that improves the accuracy where the model is trained at the outset of a sensing session. The system uses a commercial off-the-shelf webcam, which is widely available and cost effective. The experiments were done in vivo, and the results have shown that the accuracy of the system is about 90% accurate on average compared to our benchmarking device, and that it has strong potential to facilitate control of neural prostheses