基于视觉和听觉融合的移动机器人目标识别与定位方法研究

发布时间：2018-04-01 12:23

本文选题：移动机器人　切入点：声源定位　出处：《南京理工大学》2017年硕士论文

【摘要】：随着信息技术水平的提高,机器人在家庭服务中的作用也变得尤为突出,研究具有感知和决策能力的服务机器人具有重要的意义。基于人类在其生活环境中扮演着至关重要的角色,本文以人为目标,分析其听觉和视觉特征,主要研究了基于视听觉融合的移动机器人目标识别与定位技术,着重做了以下的研究工作:首先,总结了移动机器人的目标识别与定位方法的研究现状,对移动机器人的视听觉融合进行了系统平台设计,分析了移动机器人交互控制以及场景的应用。其次,研究了移动机器人的目标定位。对声源定位的方法进行了研究,在多信号分类算法中,采用广义特征值分解抑制噪声的影响。移动机器人根据声源定位得到的方位角,结合运动距离,利用三角测量得到声源距离。通过实验分析验证了该方法精度和鲁棒性。再次,研究了说话人识别和人脸识别技术。在人脸识别之前,首先进行人脸检测定位人脸区域,将人脸图像进行分块处理,利用小波分解和奇异值分解相结合提取特征后,采用稀疏表示的人脸识别;在说话人识别中,语音预处理后,经过同态处理和倒谱分析后,提取语音的特征Mel频率倒谱系数,进行矢量量化,通过LBG聚类处理,为说话人建立码本模型。最后通过实验表明,说话人识别和人脸识别的有效性。进一步地,研究了基于视听信息融合的识别技术。分别在匹配层和决策层上对语音和人脸信息进行了融合识别。在匹配层上提出了基于语音优先的匹配的加权融合和基于人脸优先匹配的加权融合,并通过实验与非加权融合进行了对比,验证了加权融合的识别率更高。在决策层上利用模糊积分将说话人识别和人脸识别的输出结果进行非线性的加权,最后通过实验表明了模糊积分对于视听融合的有效性。最后,在目标定位与识别的基础上,搭建了移动机器人的目标定位与识别系统平台。
[Abstract]:With the improvement of the level of information technology, the role of robots in home service has become more and more prominent. It is of great significance to study service robots with the ability of perception and decision making.Based on the fact that human beings play an important role in their living environment, this paper analyzes the auditory and visual characteristics of human beings, and mainly studies the target recognition and localization technology of mobile robots based on audiovisual fusion.The following research work is emphasized: firstly, the research status of target recognition and localization method of mobile robot is summarized, and the system platform of audio-visual fusion of mobile robot is designed.The interactive control of mobile robot and the application of scene are analyzed.Secondly, the target location of mobile robot is studied.In this paper, the method of acoustic source location is studied. In the multi-signal classification algorithm, generalized eigenvalue decomposition is used to suppress the influence of noise.According to the azimuth of sound source and the distance of motion, the distance of sound source is obtained by triangulation.The accuracy and robustness of the method are verified by experimental analysis.Thirdly, the techniques of speaker recognition and face recognition are studied.Before face recognition, first, face detection is carried out to locate the face region, then the face image is divided into blocks. After the feature is extracted by wavelet decomposition and singular value decomposition, the sparse representation of face recognition is adopted.After speech preprocessing and homomorphism processing and cepstrum analysis, the speech feature Mel frequency cepstrum number is extracted, vector quantization is carried out, and the codebook model is established for the speaker by LBG clustering processing.Finally, experiments show that speaker recognition and face recognition are effective.Furthermore, the recognition technology based on audiovisual information fusion is studied.The speech and face information are fused and recognized at the matching layer and the decision level respectively.The weighted fusion based on speech first matching and the weighted fusion based on face priority matching are proposed on the matching layer, and compared with non-weighted fusion through experiments, it is proved that the recognition rate of weighted fusion is higher than that of non-weighted fusion.The output results of speaker recognition and face recognition are weighted by fuzzy integral in decision level. Finally, the effectiveness of fuzzy integral for audio-visual fusion is demonstrated by experiments.Finally, on the basis of target location and recognition, a mobile robot target location and recognition system platform is built.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP242

【参考文献】