基于机器视觉的模型测量靶标的跟踪研究

发布时间：2018-04-23 18:27

本文选题：编码靶标 + 识别定位　；参考：《北京交通大学》2016年硕士论文

【摘要】：随着信息技术迅速发展,基于机器视觉的目标跟踪技术在很多领域具有广泛的应用价值。目标跟踪涉及到图像处理、模式识别以及人工智能等诸多领域,是一门交叉性很强的学科。在对目标进行跟踪的过程中,目标自身特征的多样性、运动场景的复杂性和跟踪过程遇到的问题(如遮挡、光照、防射变换、相似目标物干扰),都构成了限制目标跟踪算法性能的影响因素。因此研究一种具有实时性、准确性、鲁棒性的运动目标跟踪方法依然是该领域面临的一个巨大挑战。本文对基于机器视觉的模型测量靶标的动态跟踪算法进行研究。根据跟踪系统的特点,首先以圆作为基本结构设计测量靶标,并对靶标识别定位。根据圆的特性,采用圆参数优化方法提取靶标亚像素级的中心坐标,并通过不同的仿真实验和评价指标对提取精度进行分析,表明设计靶标的适用性和中心定位的精确性。为了满足目标跟踪系统的实时需求,本文选择了计算量较小的Mean shift跟踪算法,分析该跟踪算法容易出现跟踪失效等不足。针对光照不均匀影响问题,本文提出融合顶帽变换的Mean shift跟踪算法,提高了运动目标检测的完整性,进而增加了目标跟踪的准确性。针对Mean shift目标跟踪算法存在尺度更新策略缺陷,以及在跟踪过程中出现遮挡、防射变换等复杂情况下跟踪不稳定,本文采用基于SIFT尺度不变特征的Mean shift跟踪算法,在跟踪过程中,采用SIFT算子的局部描述和RANSAC算法实现对错误匹配点的剔除。通过在复杂环境下进行目标跟踪实验,并提取目标运动轨迹曲线进行对比分析说明,实验证明本文采用方法能够有效地适应跟踪过程中出现的尺度变换和遮挡等情况,提高了跟踪算法的适用性、准确性和鲁棒性。
[Abstract]:With the rapid development of information technology, target tracking technology based on machine vision has wide application value in many fields. Target tracking involves many fields, such as image processing, pattern recognition and artificial intelligence. In the process of tracking the target, the diversity of the target's own characteristics, the complexity of the moving scene and the problems encountered in the tracking process (such as occlusion, illumination, anti-shooting transformation, etc.) Similar objects interfere with each other, which are the factors that limit the performance of the target tracking algorithm. Therefore, studying a real-time, accurate and robust moving target tracking method is still a great challenge in this field. In this paper, the dynamic tracking algorithm of model measurement target based on machine vision is studied. According to the characteristics of the tracking system, the circle is first used as the basic structure to design the measuring target, and the target identification and positioning. According to the characteristics of the circle, the circular parameter optimization method is used to extract the central coordinates of the sub-pixel level of the target, and the accuracy of the extraction is analyzed through different simulation experiments and evaluation indexes, which shows the applicability of the design target and the accuracy of the center location. In order to meet the real-time requirement of the target tracking system, this paper chooses the Mean shift tracking algorithm which has less computation, and analyzes the shortcomings of the tracking algorithm, such as the easy occurrence of tracking failure and so on. In order to solve the problem of uneven illumination, this paper proposes a Mean shift tracking algorithm based on top cap transform, which improves the integrity of moving target detection and improves the accuracy of target tracking. Aiming at the defect of scale updating strategy in Mean shift target tracking algorithm, and the instability of tracking in complex cases, such as occlusion and anti-shooting transformation, the Mean shift tracking algorithm based on SIFT scale invariant feature is adopted in this paper. In the tracking process, the local description of SIFT operator and the RANSAC algorithm are used to eliminate the error matching points. Through the target tracking experiment in the complex environment and the comparison and analysis of the target trajectory curve, it is proved that the proposed method can effectively adapt to the scale transformation and occlusion in the tracking process. The applicability, accuracy and robustness of the tracking algorithm are improved.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】