基于可变形模型的目标跟踪算法

发布时间：2018-02-01 04:42

本文关键词： 可变形模型结构化分类器在线学习目标跟踪　出处：《红外与激光工程》2017年09期 　论文类型：期刊论文

【摘要】：近年来目标跟踪技术的研究已经有了很大的进展,但目标的遮挡和形变仍然是目标跟踪算法面临的重大挑战。针对这些问题提出了一种基于可变形模型的目标跟踪算法。首先,利用可变形模型对跟踪目标进行表达,该模型将目标分为若干子块,目标的特征由局部子块特征和全局特征共同构成。将目标的特征和子块之间的空间关系结合起来,给出了对目标的一个统一的相似度度量函数。然后,在线训练一个结构化输出支持向量机作为分类器,该分类器的输出是可变形模型中目标的结构化描述。利用该分类器可以在视频及图像序列中准确地检测到目标,完成跟踪。通过实验比较,该算法的跟踪性能优于其他主流跟踪算法,尤其在目标发生遮挡和形变的时候仍能准确跟踪。
[Abstract]:In recent years, the research of target tracking technology has made great progress. However, the occlusion and deformation of the target is still a major challenge for the target tracking algorithm. A deformable model-based target tracking algorithm is proposed to solve these problems. First of all, a new target tracking algorithm based on deformable model is proposed. The deformable model is used to express the tracking target, which divides the target into several sub-blocks. The feature of the target is composed of local subblock feature and global feature. The spatial relationship between the target feature and the sub-block is combined, and a uniform similarity measure function is given. Online training a structured output support vector machine as a classifier. The output of the classifier is the structured description of the target in the deformable model. Using the classifier, the target can be accurately detected in the video and image sequence, and the tracking is completed. The tracking performance of this algorithm is better than that of other mainstream tracking algorithms, especially when the target is occluded and deformed.
【作者单位】：中国科学院沈阳自动化研究所;中国科学院光电信息处理实验室;中国科学院大学;空军驻湖北地区军事代表室;空军装备部装备采购局;
【分类号】：TP391.41
【正文快照】： 0引言近年来计算机视觉飞速发展,其中智能视频监控、人机交互、行为识别、机器人视觉等技术已日趋成熟。目标跟踪算法是这些技术的重要组成部分,目标跟踪算法的精确性、鲁棒性对这些技术的发展、应用起着至关重要的作用。虽然目标跟踪算法层出不穷,但光照变化、部分遮挡、复杂

【相似文献】