【作者】 吴剑锋；

【导师】 孙守迁；

【作者基本信息】 浙江大学 ， 数字化艺术与设计， 2008， 博士

【摘要】 在现实社会中,受到各种主观和客观因素的限制,人的肢体可能会失去部分或全部的行动能力,这不仅影响病患的心理和生理,同时带来复杂的社会问题。如何帮助行动受限的人群恢复其独立生活的能力,已经成为当今医疗康复和养老助残等领域的重要研究课题。智能肢体动作辅助系统可以有效改善或解决这一问题,其基本目标是:在增强人类现有运动能力的同时,保留人的灵活性和直接操作的感觉。因此如何有效而精确地实时获取使用者的肢体运动信息成为目前急需解决的问题。本文重点关注下肢运动,以能够直接反映人体肌肉功能状态的肌电信号为手段,对人体动作实现的“内因”——肌肉力和关节力矩的计算以及“外现”——动作识别预测两个方面进行了研究。论文的主要研究内容包括:1.对国内外当前的运动信息获取技术方法及其发展趋势做了综述分析,指出各自的优点及不足之处,阐述本文的研究方法和内容。2.通过实验研究不同下肢动作模式与肌电信息的对应关系:利用动作捕获系统获取下肢运动过程中关节角度变化;利用肌电信号分析相关肌肉活动情况,对起立、行走和上下楼梯等最常见的日常下肢动作进行实验分析。实验结果表明,不同的下肢动作模式对应不同的肌肉兴奋时间和兴奋程度,为后文的运动信息获取提供了实践基础。3.研究基于支持向量多元分类的下肢运动模式识别技术:首先,阐明基于肌电信号的下肢动作模式识别的目标及难点;其次,在实验研究的基础上,对肌电信号特征提取做分析,并建立模式识别的特征向量空间;另外,在识别策略上采用“数据流分割”和“移动窗”的概念,降低计算复杂度,提高算法的鲁棒性;在识别算法的建立上,提出基于核聚类简化的支持向量多元分类改良算法;最后,以下肢动作模式识别实验来验证算法的有效性。4.研究基于肌电信号的肌肉力和关节力矩预测技术:首先,对当前基于肌电信息的肌肉力和关节力矩算法进行分析,指出基于单块肌肉受激行为生理特征的肌肉力和关节力矩计算方法的优点;其次,建立基于肌肉生理模型、肌肉骨骼几何模型和多刚体动力学模型三个层次的人体力学分析模型;然后,提出基于肌肉力-电关系的肌肉力和关节力矩预测模型;最后以实验验证该模型的有效性。5.实现一个基于肌电信号的运动信息检测与反馈原型系统,对自主开发的老年人起立辅助座椅进行功能分析和评价,并验证本文所提出的理论、方法和技术的正确性与可行性。更多还原

【Abstract】 The human body maybe lost movement ability because of various subjective and objective limits. It not only affects the patients’ psychological and physical state but also brings complex social problems. How to help them live independently has become very important nowadays.The concept of intelligent limb-assisted system aims at enhancing human movement capabilities and retaining the flexibility and feeling for direct operation at the same time. So how to obtain the limb’s information effectively and precisely in real-time becomes urgent at present. EMG signals can directly reflect human muscles’ function, based on that the dissertation mainly concentrates on two important contents. One is how to calculate muscle force and joint torques, and the other is how to recognize and predict the human motions.The main contents of this dissertation are shown as following:1. Summarized the tendency and methods for acquiring motion information at home and abroad. The advantages and disadvantages were pointed out. Narrated the main research contents and research methods for this thesis.2. The relationships between different lower-limb motion modes and EMG signals were analyzed through experiments. Standing up, going up and down stairs and walking were performed as the daily lower-limb motion in the experiment. The angle of each joint was measured by motion capture system and the activity levels of muscles were accessed by the muscle EMG signals. The results showed that different lower-limb motions had different muscle excitation time and excitation grade. It provided a practical basis for the following study.3. The recognition technology of lower-limb motion modes was studied based on Multi-Class SVM. First of all, the aims and difficulties of mode recognition on lower-limb motion based on EMG signals were expounded. Secondly, the analysis on EMG characteristics extraction has been done and an eigenvector space of mode recognition was built. On the other hand, developed an improved algorithm based on Multi-Class SVM. The results of mode recognition experiments showed this method could effectively resolve the lower-limb motion identifications in real-time.4. The muscle force and the joints torque prediction technique have been researched based on EMG Firstly, current application using EMG as a method for muscle force and joint torque calculation was analyzed. Then the benefits on muscles force and joint torque calculation based on single muscles stimulated were pointed out. Secondly, this research established a human body mechanics model based on the muscle physiological model, skeletal muscle model and multi-body dynamic model. Then the prediction model of the muscles force and joint torque was advanced. At last, the experiment results validated the model’s availability.5. A prototype system was established to obtain motion information. Then the function of standing-up-assisted seat for the aged was analyzed and evaluated. It verified the feasibility and validity of theory, methods and techniques in this dissertation.更多还原