【作者】 潘华东；

【导师】 刘济林；

【作者基本信息】 浙江大学 ， 通信与信息系统， 2010， 博士

【摘要】 三维成像技术发展迅速,广泛应用于导航、监控、识别和虚拟现实等领域。其中,无扫描三维成像系统是目前的研究热点。基于直接飞行时间(Time of Flight,TOF)测量法的无扫描三维成像系统,探测距离远,能穿透树叶、尘埃等障碍物,适合军事应用领域。基于间接TOF测量法和像增强器耦合型CCD (Intensified CCD, ICCD)/CMOS传感器的无扫描三维成像系统,空间分辨率较高,适合机器人导航、智能视频监控、演播室和游戏控制等近距离应用。基于ICCD的三维成像系统,与CMOS专用传感器技术相比,具有高空间分辨率和高动态范围等优点,但因缺乏定量分析方法,无法对各种因素进行精确分析,给系统设计和优化造成了困难。为解决这些问题,优化系统设计,本文对基于ICCD传感器和间接TOF测量法的无扫描三维成像摄像机的机理和特性进行研究,分别从工作原理,系统建模,数据处理和系统实现这几个方面,提出了具有参考意义的结论和改进的算法。第一部分研究了TOF系统测距理论。从摄像机固定噪声和光子噪声出发,推导了几种间接TOF测量法的最小相对误差。对余弦波-方波法展开重点研究：根据采样原理,研究了调制光源谐波分量、光源调制度、摄像机解调深度、采样脉冲波形对称性和系统增益非线性等因素对测量精度的影响。理论分析结果表明,应提高摄像机动态范围和势阱容量,抑制余弦调制光源奇次谐波分量,提高光源调制度和系统解调度,提高采样脉冲波形对称性,抑制高于2阶的非线性。第二部分主要研究了基于ICCD的TOF无扫描三维成像系统模型。改进了ICCD噪声模型和增益饱和模型,并推导了测量误差估计公式。结合光源功率和像增强器增益非恒定的特点,提出了模型时间离散化的处理方法。结合光照模型和物体三维数据,采用蒙特卡洛仿真方法,在MATLAB环境下,对影响系统测量精度的因素进行定量分析。实验结果验证了第一部分的研究结论,同时表明应该降低MCP增益,选择时间常数较小的低阻抗MCP。提出了自动曝光控制和分段检测两种改进测量精度的方法。该模型为系统设计优化提供了依据和支持。第三部分以SR3000摄像机为例,对三维成像系统的数据处理方法进行了研究,提出了一种数据处理框架。介绍了摄像机标定和距离校正技术。提出了利用交替频率消除距离歧义性的方法和基于感兴趣区域的快速自动曝光控制方法。针对环境因素,提出了一种基于幅值的数据可靠性判断和噪声滤除方法。针对混合像素,提出了一种自适应阈值的边界混合像素滤除方法。针对散射等特殊情况,提出了根据偏移值与幅值比值进行数据有效性判断的方法。第四部分,针对无扫描三维成像系统实现进行研究,提出了一种系统参数设置和优化方法。首先根据各参数对测量结果的影响正向推导系统最优参数,然后结合具体器件性能反向确定相关参数预测值,最后根据仿真结果修正系统参数。提出了通过延时曝光来提高系统动态范围的方法。在此基础上,提出了系统硬件结构、软件结构和外形结构的设计方法。最后是全文总结和展望。更多还原

【Abstract】 The rapid development of 3D imaging technology, has been widely used in navigation, surveillance, identification and virtual reality fields, and non-scanning three-dimensional imaging system is the focus of research. Direct TOF measurement based non-scanning 3D imaging system is suitable for military applications for its long detection range, and high penetration ability into leaves, dust and other barriers. Non-scanning 3D imaging system based on indirect TOF measurement and ICCD/ CMOS sensor, for its high spatial resolution, is more suitable for robot navigation, intelligent video surveillance, studio, game control and other close range applications. ICCD-based 3D imaging system, compared with CMOS sensor technology, has high spatial resolution and high dynamic range. But it’s hard to design and optimize, since quantitative analysis method is not available. To address these problems, this dissertation studied on the principle and the characteristics of indirect TOF non-scanning 3D imaging camera based on ICCD and some conclusions for reference and improved algorithms were proposed on system principles, system modeling, data processing methods and system implementations.The first part is the ranging principle of TOF system theory. The minimum relative error of various indirect TOF methods commonly used was derived from the view of camera noise and photon noise. The cosine wave-square wave method was studied: according to sampling theory, aliasing, light modulation depth, demodulation depth of the camera, the symmetry of sampling pulse, non-linear system gain and other factors which affect measurement accuracy were studied. The results showed that dynamic range and potential well capacity of the camera should be increased. Odd harmonic of cosine modulated light and nonlinearity higher than 2nd -order had impact on accuracy. Additionally, light modulation depth and system demodulation depth, as well as the symmetry of sampling pulse should be improved.The second part studied the model of TOF scannerless 3D imaging system based on ICCD. An improved ICCD noise model was proposed, and a nonlinear gain model was used. As the light was cosine modulated and the gain of image intensifiers changed over time, the model was modified in time-discrete situation.. The system was simulated in MATLAB by Monte Carlo simulation method, together with illumination model and objects 3D data. The influence of the factors on measurement accuracy was analyzed quantificationally. The results verified the conclusion in the first part of research. Additionally, it showed that small MCP gain and low impedance MCP with small time constant should be chosen. Automatic exposure control and segmented detection method were suggested for improving the measurement accuracy. The model provides the basis and support for system design and optimization.The third part, data processing methods of 3D imaging system based on SR3000 camera were studied, and a framework for data processing was presented. Calibration and distance correction technologies were introduced. A method to eliminate range ambiguity by alternating frequency, and a rapid automatic exposure control method based on the region of interest were proposed. Considering environmental factors, a data reliability analysis method and a noise filtering method based on amplitude were proposed. Kinds of mixed pixel were discussed, and an adaptive threshold filter method for the border mixed-pixel was presented. Scatter and other special conditions were also analyzed, and a method for range data analysis based on the ratio of offset and amplitude was suggested.The last part talked about the design and optimization of non-scanning 3D imaging system. A method for system design and optimization was proposed:Firstly, the impact of parameters on the measurement results was derived, and then, the parameters were determined conversely according to the specific device performance. Finally, the parameters were corrected according to simulation results. A method for increasing the system dynamic range by exposure time delay control was proposed. Additionally, the hardware structure, software structure and structure design method were discussed.Finally the conclusion and perspective are given at the end of the dissertation.更多还原