节点文献
基于FSR传感器的假手运动模式识别及控制系统研究
Study on the Prosthetic Hand Motion Pattern Recognition Based on FSR Sensors and Control System
【作者】 孙超;
【导师】 姜力;
【作者基本信息】 哈尔滨工业大学 , 机械电子工程, 2009, 硕士
【摘要】 本文提出一种新型的手部姿态识别系统,用于控制多自由度假手。本系统利用FSR(Force Sensitive Resistor)传感器检测前臂肌肉的收缩和膨胀情况来实现不同动作的识别。通过前臂处的FSR传感器获取不同手部姿态对应的信号,经过支持向量机(Support Vector Machine)分类器分类后给出相应手部运动模式,这样使用者可以实现多自由度假手的控制。本文建立假手运动模式识别及控制平台,此系统由三部分组成。首先,多通道FSR传感器信号采集系统实现前臂与系统之间的连接,为后续处理提供可靠的、有效的信号。上位机内信号分析处理试验系统,针对不同个体特点选择不同的传感器位置和数量、模式分类方法及最优参数模型,最大限度满足操作者需求。最后引入基于双DSP的假手控制系统,包括基于DSP2810的假手驱动与传感器系统和负责信号处理与决策的基于DSP2812的上层控制系统,彻底实现嵌入式控制,为假手的商业化做好准备。对于多自由度假手的多模式控制,模式分类方法是核心内容。从实用的角度出发,尽量提高分类的快速性和保证高成功率。本文选择支持向量机这一模式分类方法,并针对假手控制特点讨论多类分类方法,引入基于量子粒子群的全局最优参数搜索方法,并通过实验证明其有效性。为方便假手随身佩戴及嵌入式控制要求,介绍基于DSP2812的假手动作模式识别程序的设计思想及流程,系统的存储器配置和SPI接口的寄存器配置。算法程序设计分为三个部分,数据采集部分并加入采集过程中动作切换提示;SVM模型建立部分通过简化SVM的约束条件,实现样本训练过程在DSP中完成;分类预测部分,结合一对一方法,分别采用投票法和模糊法实现模式分类。针对建立的多自由度仿人型假手系统,本文进行了大量的实验,实验结果为:在PC机中,根据训练难度将33种手部动作分为3类,按由易到难的顺序进行训练和分类,验证控制方式的有效性。在DSP中,利用10枚传感器,可以识别10种常用运动模式,并且成功率在95%以上。
【Abstract】 In this thesis a new recognition system for hand gestures developed for the purpose of controlling active prosthesis hand is presented. The recognition system allows for the measurement and classification of muscle contraction around the lower arm. The singles obtained by the FSR sensors would be analyzed by the SVM divider, which is developed based on the theory of setting maximal distance between different categories, and then assigned to certain category. So the user can control the prosthesis through muscle contraction of the forearm according to different hand gestures.The Multi-DOF prosthetic hand platform is constructed, which includes three parts. Firstly, the multi-channel acquisition system of FSR signals is established, which offer an interface between the human body and the system. Then, the analysis and processing system for experiment of FSR signals is founded. In this system, we can optimize the location and number of sensors; choose the most comfortable classification method and its parameters based on different operators to meet their demands. Lastly, a control system based on two DSP is also imported. The system contains the drive and sense system based on DSP2810 and the system based on DSP2812, which is used for management and decision-making. Preparing to be commercially produced, all of them can be achieved in DSP totally.As for the control of Multi-DOF prosthetic, the method of classification is the core content. From the point of view of practical utility, the improvement of rapidity and success rate of classification is the main aim. The widely used method of SVM is also applied in this paper, in addition, the way of searching optimal parameters through QPSO are provided. In the end, experiments validate the effectiveness of the proposed method.To meet needs of taking prosthetic hand every day, ideas and process of design for the prosthetic hand motion pattern recognition are introduced, which is used in embedded system .In the part of data acquisition, The configuration of SPI registers provided and the main program for sending and receiving are also introduced. As for the part of SVM model establishing, by simplifying the restrained conditions, the training can be accomplished in DSP. Lastly, by using one against one multi classification method, the class of test data is given through vote or fuzzy.Based on the established prosthesis hand system, a lot of experiments are carried. The result indicates that, in the PC, the thirty three hand gestures are divided into three categories form the easiness to the difficulty, this kind of control method is validated a good one through training and classification; in the DSP, ten often used hand gestures can be recognized by ten sensors and the average success rate is above 95%.