Study On Gesture Recognition Algorithm And Robustness Of B-mode Ultrasound-based Human-machine Interface

The clinical applications of prosthesis have not been effectively promoted because of some limitations for the existing human-machine interfaces(HMIs).However,ultrasound images have extremely high temporal and spatial resolution,which can provide multiple information including muscular morphological features.This paper introduces a new kind of HMIs based on B-mode ultrasound,verifies its off-line and online gesture recognition performance and further studies its robustness following the research routes of traditional HMIs.Firstly,this paper designs the contrast experiment between ultrasound and surface electromyography(sEMG).The experimental results demonstrate that the ultrasound has better average accuracy(95.88%)than the sEMG(90.14%)in the recognition of discrete finger motions.And ultrasound outperforms sEMG in the prediction of finger angle as well.Both of these confirm that the off-line performance of the ultrasound-based HMI has a strong superiority compared to the existing sEMG-based HMIs.Secondly,this paper builds an online gesture recognition platform of ultrasound-based HMI.Through gray-level co–occurrence matrix,the online indexes in 10 hand gestures in average: motion selection time(0.24s),completion time(1.27s),completion rate(97.2%),and real-time accuracy(94.6%),reveal the excellent online performance for this HIM.Finally,this paper further studies the robustness of B-mode ultrasound-based HMI in gesture recognition.In order to solve the problems like probe shift in long-term usage,the paper proposes a method,which uses image correlation analysis for probe positioning,extracts robust features by BOF model and self-updates the classifier.The experimental result shows that the average accuracy of 6 hand gestures in 6 days is 97.6%,which proves that the robustness of the proposed scheme is excellent.In addition,the robustness of the proposed features in different force levels and different degrees of fatigue is also discussed experimentally.The experimental result shows that muscle fatigue has little effect on gesture recognition while different force levels can make the recognition accuracy slightly decreased but the average accuracy of the proposed scheme is still stabilized at more than 90%.All of the above demonstrate the feasibility of B-mode ultrasound-based HMI and the effectiveness of the proposed scheme.