Abstract: The analysis about the theory of Generalized HoughTransform（GHT） and the existing research in the field of object detection is presented. Based on the original single object detection, the ability of multiple similar objects detection and counting is added. According to the threshold setting, many objects are identified quickly and accurately from the image, and the related position information of these objects is recorded. GHT algorithm is accelerated on GPU by exploiting the potential parallelism in the various steps of the algorithm and reasonably optimizing the memory using of GPU. The experimental results indicate that, GPU greatly enhances the running speed of the algorithm.

Key words: object detection, object counting, Generalized Hough Transform（GHT）, Graphics Processing Unit（GPU）

摘要： 分析了广义霍夫变换（GHT）的原理及在目标检测领域的已有研究，在原单目标检测的基础上，加入了多个相似目标物体检测与计数的能力，根据阈值的设定，能够快速准确地从图像中同时识别出多个目标物体，并记录下各目标物体的相关位置信息。利用GPU对GHT算法进行加速，充分挖掘算法各个步骤的潜在并行性，并结合GPU中的各种存储器优化方法，合理安排各存储器的存储数据，实验结果表明，GPU极大提升了算法的运行速度。

关键词: 目标检测, 物体计数, 广义霍夫变换（GHT）, 图像处理器（GPU）