基于X射线的计算机断层成像（Computed Tomography，CT）技术是目前工业无损检测领域中的重要检测手段。在工业CT系统的实现当中，机械运动系统需要保证扫描过程中较高的运动精度，在对大型物体进行检测时，这一问题限制了大型工业CT系统的适用范围。为使工业CT能适用于超大型的物体，同时降低系统对机械精度和环境条件的要求，本论文提出了一种基于视觉定位技术的工业CT系统设计。通过对被检测物体扫描过程中的运动轨迹进行实时测量，可以在机械运动存在偏差的情况下实现工业CT的扫描和重建，提高了工业CT系统的适用性。基于以上设计思路，本论文将视觉定位系统与多段直线CT扫描方式相结合，给出了一种适用于大型物体的工业CT检测方案。系统中需要解决的关键技术主要有：1、通过视觉定位系统测量被检物体的运动信息；2、设计结合定位信息的非确定扫描轨迹重建算法。视觉定位方面，本论文根据CT扫描的特点确定了系统需要测量的信息及对应的精度，设计了相应的视觉定位方案。针对定位系统布局及硬件参数的选择，本论文分析了影响定位系统精度的各个因素，并通过实验验证了定位系统的可行性。此外，本论文还提出了一种基于图像不变矩的投影数据自校正方法，在不引入新的设备的情况下实现了位置信息的估计，为定位系统提供了一种低成本的备选方案。重建算法方面，本论文在现有重建算法的基础上，利用定位信息得到系统的实际扫描轨迹，并针对出现的轨迹误差给出了对应的结合定位信息的多段直线CT重建方法。为缩短扫描时间并提高检测效率，实际系统可能不进行全部角度的扫描，这种情况下投影数据存在有限角问题，本论文分析了多段直线轨迹CT扫描中可能出现的有限角度数据形式，并利用压缩感知的方法给出了对应的有限角度重建算法。为扩展视觉定位系统在工业CT中的适用范围，本论文将非确定轨迹图像重建算法扩展到了一般轨迹中，提出了基于扇束滤波反投影方法的非确定轨迹重建算法，为基于视觉定位的CT系统设计提供了便利。论文研究期间在现有实验室多方式扫描平台的基础上添加了视觉定位系统，并通过实验验证了视觉定位CT系统的可行性。

X-ray computed tomography (CT) is an important approach in industrial non-destructive inspection. In industrial CT applications, mechanical system needs to provide accurate movement during the scanning process, which has constrained the application of industrial CT when the object to be scanned is very large. To lower the requirement of mechanical system design and make industrial CT applicable for extreme large object, this dissertation proposes a new CT system design with stereo vision tracking subsystem. By measuring the movement during scanning process, it can realize industrial CT scan and reconstruction under biased mechanical condition, which improve the feasibility of industrial CT.Based on above approach, this dissertation adds vision tracking system on multi-segments straight line CT and gives a system design for the inspection of large object. There are two key techniques need to be solved in this system: 1. measure the movement of object by vision based tracking; 2. design the reconstruction algorithm for dynamic orbit CT.In realizing tracking system, this dissertation analyses the variable need to be measured for CT reconstruction and proposes corresponding implementation. We analyses the factor that affect the accuracy of tracking system and validate the feasibility of this method. In addition, we proposes another method to measure the movement by calculating the invariant moments from projection data, which require no additional equipment and provide an alternative solution for tracking system.In the aspect of reconstruction algorithm, we extend current reconstruction algorithm for multi-segments straight line CT by considering orbit distortion by inaccurate movement. To improve the efficiency of inspection and save inspection time, the scanning process may not cover full range of projection angle. In this condition, the projection data in real system may have limited angle problem. We analyses the style of limited angle data in multi-segments straight line CT and proposes an iterative reconstruction algorithm based on compressive sensing. To extend the application field of vision tracking CT system, we develop a filtered-backprojection type reconstruction algorithm for general dynamic orbit, which gives flexibility for CT system design. We implement the vision based tracking system on an experimental cone-beam CT system with flat panel detector. Experiment results demonstrate the effectiveness of this CT system design with stereo vision tracking.