基于序列图像的运动目标检测和跟踪-检测技术与自动化装置专业论文.docxVIP

中墨辩学技术大学硕士学位论文 中墨辩学技术大学硕士学位论文 杨擎字 摘要 为了鳃次复杂背襞下，I、爨标运动豹检测翻跟踪阕题，本文提出了一莉基于图 像差分和数掭关联静运动蟊栎检测帮跟踪算法。 酋先，因为在图像序列的采集过程中，摄像杭的褫点彼置在不断酌发生交亿。 所以必须使用图像配准的方法。我们通过使用寻优的方法得到连续两幅圈像的空 间透视投影变换关系后，将配准阔像变换到参考阉像坐标系下。然后使用图像差 分的技术去处运动物体的背景信息，提取出W能的运动目标轨迹点。 在j墩基硝上，利鬟运动晷标农空闼寒避阉上的相关性以及运动最标豹轨迹所 其有的连续健，采魇一种特豫懿紧类方法，遴过转述起始，鞔迹延续，孰迹终止 三个过程将运动舀标轨述扶嗓声中摄取蠢来。在算法率，我们使耀卡尔爨滤波器 物体的运动参数信息滤波，预测，掇高了跟踪的鲁棒性和准确髋。 实验表明该算法能地检测出复杂背景下的小目标的运动信息，并能有效地处 理轨迹棚交葶Ⅱ检测过程巾出现飘目标的情况。 荚键调图像配准，佐化方法，装类算滋，运动轨迹跟踪，数据关联 IV 中国科学技术丈学碗士学位论文 中国科学技术丈学碗士学位论文 杨擎宇 Abstract To solve the moving point target detection problem from the image sequence under the condition of complex background，a new algorithm based on image di fference and data correlation is presented in this paper． First，since the camera viewpoint of every image is different，image registration must be used to get the spatial and perspective relations among the sequenced images．详e calculate the parameters by using optimization technology and several optimization methods are appl ied in our work．After that，by transforming the region image into the coordinate of the reference image，the successive two images are differenced to find the possible moving point targets． Then，a clustering method base on data correlation is introduced to pick out the targets from the noise．7he detection of each track is di vided into three steps：tracking initiation，tracking maintenance and tracking termination．Furthermore we use a Kalman—filtering to predict the movement parameters，which can highly improve the efficiency and accuracy of the tracking process． The experimental results show that the atgorithm c姐effectively and reliably detect the targets under the complex background。The algorithm can also work very well with some special situations}such as the situations of crossed tracks，new tracks during the whole detecting and tracking process． V 中国科学技术大学硕士学位论文 中国科学技术大学硕士学位论文 杨擎宇 KeyWords：image registration，optimization technology，elusteringmethod， moving target detection，data association 艟 串国辩学技术大学硕士学位