【作者】 姜炜；

【导师】 刘建国；

【作者基本信息】 华中科技大学 ， 控制工程， 2017， 硕士

【摘要】 车轮的图像检测与跟踪在安防、智能交通等领域有着重要的应用价值,本文对基于视频图像的车轮检测与跟踪进行了深入研究。本文首先研究了车轮检测,由于Hough变换法可以很好的抑制图像噪声,因此本文以Hough变换法为车轮检测算法的基础,但目前基于离散边缘点的Hough变换法存在计算量大、内存占用多等缺陷。由于弧段比离散边缘点包含更加完整和准确的几何形状信息,可以排除其他无关边缘点的干扰,因而本文提出以弧段组合代替离散边缘点作为车轮检测数据的思想。首先需要从图像大量离散边缘点中提取弧段,为了消除图像的噪声、孤立点的影响,正确获得车轮检测需要的弧段,本文提出了一种简单高效的车轮轮廓的弧段提取方法。由于所提取的弧段通常是车轮轮廓的一部分,为了进一步提高算法的运行效率,本文设计了同属一个车轮轮廓的弧段组合方法。然后以弧段组合所包含的边缘点数据为基础,采用Hough变换法进行车轮检测,并根据可靠度指标和相似度指标对检测结果进行筛选。之后使用本文所设计的车轮检测算法在现实图像上进行实验验证,实验结果表明该算法具有较好的准确性和鲁棒性。检测到车轮的具体位置后,为了实现对运动车轮的行为理解,本文对车轮跟踪进行了深入研究。针对实际应用环境,本文选用核相关滤波器跟踪算法对车轮进行跟踪,并对核相关滤波器进行了适当改进。然后根据车轮跟踪过程的三种状态设计了整个跟踪流程。最后实验测试了该算法的跟踪效果,实验结果表明该算法能够实时准确的检测与跟踪实际应用环境中的运动车轮。更多还原

【Abstract】 Wheel image detection and tracking has important application value in the security,intelligent transportation and other fields,this paper studies deeply the wheel detection and tracking based on video image.This paper first studies the wheel detection.The Hough transform method can suppress the image noise very well.Therefore,Hough transform is the basis of wheel detection algorithm.But the main existing problems of Hough transform which based on the discrete edge point are long time consuming,big memory occupancy and so on.Since the arc contains more complete and accurate geometric information than the discrete edge points,it is possible to exclude interference from other extraneous edge points.Therefore,this paper proposes the idea of replacing the discrete edge points with the arc segment as the wheel detection data.In order to eliminate the influence of the noise and the isolated point of the image,correctly extract arc needed,this paper presents a simple and efficient method for extracting arc from wheel contours.Beacause arc extracted is generally part boundary of wheel contours,so in order to further improve the efficiency of the algorithm,this paper has designed arc grouping algorithm.Then,based on the edge points included in the arc group,the Hough transform method is used to detect the wheel and the detection results are screened according to the reliability index and the similarity index.Finally,with the designed wheel detection algorithm,experiments have carried out on the real images.The experimental results show that the algorithm has good accuracy and robustness.After the specific position of the wheel is detected,in order to realize the understanding of the behavior of the moving wheel,this paper has conducted in-depth study of the wheel tracking.Aiming at the practical application environment,this paper chooses Kernelized Correlation Filter tracking algorithm to track the wheel,and the Kernelized Correlation Filter has been improved.Then,the whole tracking process is designed according to the three states of the wheel tracking process.Finally,the experimental results show that the algorithm can detect and track the moving wheel in the real application environment in real time.更多还原