由于现有以大数据量和计算量为基础的大尺寸动态视觉测量系统处理速度较慢，本文建立了一个高速大尺寸动态视觉测量系统，并对该系统涉及的特征点中心定位、编码点识别、相机定向等算法进行了并行化研究。首先，分析了在不同测量条件下各个主要算法的时间消耗情况及每个主要算法的并行性; 然后，对常规的特征点中心定位和编码点识别算法做了介绍，分别提出了特征点中心并行快速定位和编码点并行快速识别算法，并详细说明了这两种并行快速算法的实现原理。最后，针对大量原子操作的问题，提出了线程束集体原子操作的优化方法。实验结果表明: 在不损失定位精度和识别率的前提下，图像中包含300个点时的并行方案比串行方案的时间开销减少了42%，当点数达到20 000时，时间开销减少91%以上。实验显示提出的并行设计方案有效地提高了处理速度，解决了大尺寸动态视觉测量系统实时性差的问题。

As existing large scale dynamic vision measurement systems based on large amount of data processing and computing have a lower data processing speed,a high speed and large scale dynamic vision measurement system was proposed. The parallel optimization techniques involving in the system such as target locating, code detecting, camera orientation and other systematic algorithms were researched. Firstly, time consumptions of corresponding key algorithms under different considerations were analyzed. Then, the traditional target centroid and code detecting algorithms were introduced, their parallelisms were analysis, and the fast target center location algorithm and code recognition algorithm under the general parallel architecture were put forward. Moreover, the implement methods of the two parallel algorithms were explained in detail. Finally, the Warp Atom Operation Optimization (WAOO) method for massive atomic operations was proposed. The experimental results under the same location precision and recognition rate show that the processing time is reduced by 42% and 91% for 300 and 20 000 targets respectively when the parallel algorithm is used to replace the traditional serial algorithm. The algorithms proposed in this paper are verified effective in accelerating the processing speed and improving the real-time problem in large scale vision measurement systems.