【作者】 苗苗；

【导师】 赵姝颖；

【作者基本信息】 东北大学 ， 模式识别与智能系统， 2015， 硕士

【摘要】 随着机器人相关技术的日趋完善,仿人机器人的拟人化程度也日益提高,实现了对人类在视觉、听觉、嗅觉、触觉上的高度模拟。听觉作为人类感官中最重要的组成部分,具有信息处理量小、实际应用性高、硬件成本低、隐蔽性强等独特优点,可以弥补视觉在视线和能见度等方面的限制,为机器人感知技术的研究提供了新的途径。自上世纪80年代开始,声源定位技术逐步受到国内外研究者的重视。作为声源分离和语音识别的基础能准确、快速的实现对目标的定位,本文基于仿人机器人NAO的四个麦克风阵列实现对声源目标的跟踪定位。主要的研究内容分为两个部分:声源目标定位和运动声源目标跟踪。相较于其他的声源定位算法,到达时间差(TDOA)方法的计算量较小,适用于实时在线处理,在声源定位算法中,基于机器人四个麦克风在三维空间进行,并分析了影响定位精度的因素,提出了几种去除模糊点的时差定位方法。并针对此定位算法目前的局限性进行了分析。目前,另一种广泛使用的声源定位方法是可控波束形成技术,相较于到达时间差法,推迟了做决策的时间,抗噪减弱混响的能力更强,融合了更多的麦克风信息,本文在基本的可控波束形成技术的基础上提出了改进,利用回响的指数衰减性与脉冲性,提出了有效的去除回响的方法,总结汉语拼音的发音频率特性,改变可控波束形成技术的加权系数,定位精确度高。另外,在可控波束形成技术的基础上应用卡尔曼滤波,使系统的跟踪性能更加的稳定。最后,我们在室内环境中,基于NAO机器人的软硬件条件在实际中采集数据实现对声源目标的定位以及声源目标的跟踪,并对基于到达时间差法的声源定位与改进的基于可控波束形成技术的定位算法进行分析比较。更多还原

【Abstract】 With the development of robot-related technologies,the anthropomorphic degree of humanoid robots is increasing day by day,also achieved the high level simulation of the human on vision,auditory,smell and touch.Hearing as the most important part of the human senses,with some unique advantages,such as a small amount of information processing,high practical applications,low hardware cost,strong concealment and so on.It can compensate for visual limitations in sight and visibility,provides a new way for robot senses technology research.Since the 1980s,the technology of sound source localization has been paid more attention by domestic and foreign researchers.As the basis of sound source separation and speech recognition can realize the target position accurately and quickly.This thesis based on the humanoid robot NAO’s four microphone array realized the sound source localization and tracking.The main research content of this paper is divided into two parts:one is sound source target locating,the other is moving sound source target tracking.Compared with other sound source localization method,TDOA is suitable for on-line processing due to its small amount of calculation,low requirements in hardware configuration and simple arithmetic.The sound source locating algorithm is in 3D space based on robot’s four microphones,this thesis mainly analyzes the factors that affect the positioning accuracy,proposes some new methods for removing the fuzzy points,and analyzes the limitation of this localization method.At present,another widely used sound source localization method is beamforming,compared to the Time Of Arrival Difference method,it delays the decision making time,has a stronger ability to resist noise weaken echo,fuses more information of the microphones.Based on primary beamforming technology,this thesis presents some improvements,proposes an effective method to weaken echo using the exponential decay and the impulse of echo,summarizes the pronunciation frequency of Chinese Pinyin,changes the weighting efficient of beamformimg.Based on the above improvements,the location performance get good result.In addition,A combination of kalman and beamforming technology,which makes tracking more robust and stable.In the end,we make experiments in the indoor environment,based on the NAO robot’s hardware and software collect data in a real time to achieve the target of sound source location and sound source target tracking,and compare the TDOA and the improved beamforming technology更多还原