Modeling wearable bend sensor behavior for human motion capture

The possibilities offered by variable resistance bend sensors, applied as wearable devices on body garments, to recover human joint bend angles for body segment movement tracking, have been investigated, underlying their advantages and drawbacks in real-time applications. Due to their pliability, sensitivity, and cheapness, they could be a valid alternative to movement analysis systems, based on optoelectronic devices or inertial electronic sensors. This paper suggests a new method for sensor characterization under fast bend and extension movements, to extract few parameters of a synthetic model, which provide to the users the chance to foresee their electrical performance in different applications. The sensor and their extracted models were applied to register the human knee rotation during a gait cycle, either at slow speed (83 deg/s) for a walking pattern at 5 km/h, and at high speed (650 deg/s) for a running pattern of a sprinter at 10 m/s, and finally the finger joint rotations at their maximum angular velocity (900 deg/s). This was done for a twofold purpose: from one hand, to assess the model capability to predict the sensor performance, tracking human body segment rotations at different speed, without the need of measurement; from the other hand, to recover in real time the actual sensor rotation from its resistance measurement, especially in high speed applications, where its response is distorted. With this technique, the mean error decreases from 22.5° to 3.7° in the worst cas