基于超像素的压缩感知跟踪

发布时间：2018-02-28 17:03

本文关键词： 压缩感知置信图超像素目标跟踪　出处：《天津大学》2016年硕士论文　论文类型：学位论文

【摘要】：复杂场景下的目标跟踪是计算机视觉领域最热点的课题之一。经过几十年的研究,目标跟踪技术有了长足的发展,并在视频监控、智能交通、人机交互等民用和军事领域上都有广泛的应用。但在实际应用中,目标跟踪依然是很有挑战的问题,例如光照变化、目标外观变化,目标被遮挡和复杂背景干扰等众多因素。这些因素对目标跟踪算法的鲁棒性和实时性提出很高的要求。当前,基于压缩感知理论的跟踪算法通过应用随机测量矩阵去压缩图像信号来提取低维特征,极大地提高跟踪算法的实时性且越来越引起人们注意。然而当前景目标和背景在形状或者纹理相似时,跟踪结果可能并不准确。针对此,本文提出基于超像素的压缩感知跟踪(Superpixel-based compressive tracking,SCT)算法,该算法根据新来的帧和目标在超像素之间的相似性来构建置信图。超像素块能把像素聚合成有意义原子区域,SCT算法吸收其优点。置信图提供很强的证据用来度量目标出现的可能性,这能够捕捉到在超像素级别目标和背景局部外观颜色的不同,同时改进实时压缩感知跟踪(Fast compressive tracking,FCT)算法的粗粒度到细粒度搜索策略。综上,本文提出基于超像素的压缩感知跟踪算法,该算法不仅考虑到目标和背景在形状或者纹理的不同,而且充分利用超像素级别判别性强的颜色描述子构建的置信图提供指导。在具挑战性视频序列上的实验结果表明就准确性和鲁棒性而言提出的算法优于最新水平的算法。
[Abstract]:Target tracking in complex scenes is one of the hottest topics in the field of computer vision. After decades of research, target tracking technology has made great progress, and in video surveillance, intelligent transportation, It is widely used in civil and military fields, such as human-computer interaction. But in practical application, target tracking is still a challenging problem, such as illumination change, target appearance change, There are many factors, such as target occlusion and complex background interference. These factors require high robustness and real-time performance of target tracking algorithm. The tracking algorithm based on compressed sensing theory extracts low-dimensional features by using random measurement matrix to compress image signals. It greatly improves the real-time performance of the tracking algorithm and attracts more and more attention. However, when the foreground target and background are similar in shape or texture, the tracking results may not be accurate. In this paper, a super-pixel based compressive tracking algorithm is proposed. According to the similarity between the new frame and the target, the algorithm constructs the confidence chart. The superpixel block can aggregate the pixels into a meaningful atomic region and the SCT algorithm absorbs its advantages. The confidence chart provides a strong evidence for measurement. The possibility of a target, This can capture the difference in local appearance colors between targets and backgrounds at the super-pixel level, while improving the coarse-grained to fine-grained search strategy of the Fast compressive tracking algorithm for real-time compression awareness tracking. In this paper, a compression sensing tracking algorithm based on hyperpixel is proposed. This algorithm not only takes into account the difference of object and background in shape or texture. The experimental results on challenging video sequences show that the proposed algorithm is superior to the latest algorithm in terms of accuracy and robustness.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】