摘要
鱼眼摄像机的广泛应用使得鱼眼视频目标跟踪受到了越来越多的关注,然而其特殊的成像原理造成的严重畸变给目标跟踪带来了不利的影响.为了降低鱼眼镜头视频图像的畸变对跟踪结果的影响,提出一种基于响应模板和特征组合的鱼眼视频目标跟踪方法.首先基于多个样本对应的响应值合成响应模板,构建基于响应模板的分类器;然后提取目标的HoG特征和Color Name特征分别训练分类器,综合考虑2个分类器的响应来确定目标位置;最后在分类器训练前用成像模型对变形的目标进行矫正.在鱼眼视频数据集上的实验结果表明,该方法能够在实时性的基础上很好地降低图像畸变以及目标变形的影响,获得良好的跟踪表现.
The wide use of fisheye cameras makes target tracking on fisheye video get increasing attention. However, the serious distortion caused by the special imaging principle of the fisheye lens brings the target tracking negative effect. Aiming at weakening the interference of the distortion, this paper proposes a novel fisheye video target tracking method based on response template and feature integration. Firstly, the proposed method synthesizes the response template based on the responses of multiple samples as well as constructs a classifier based on the response template, and then extracts the object’s HoG feature and Color Name feature respectively to train the corresponding classifiers. The responses of two classifiers are considered jointly to determine the target location. For further optimizing the tracker, imaging model is used to correct the deformed target before the training of the classifier. Finally, the evaluation results on the constructed fisheye video dataset validate that the proposed method can greatly reduce the negative impact of the image distortion and the target deformation while keeping the real-time performance.
作者
周小龙
黄诚斌
邵展鹏
陈胜勇
雷帮军
Zhou Xiaolong;Huang Chengbin;Shao Zhanpeng;Chen Shengyong;Lei Bangjun(College of Electrical and Information Engineering, Quzhou University, Quzhou 324000;College of Computer Science and Technology, Zhejiang University of Technology, Hangzhou 310023;School of Computer Communication and Engineering, Tianjin University of Technology, Tianjin 300384;Hubei Key Laboratory of Intelligent Vision Based Monitoring for Hydroelectric Engineering (China Three Gorges University), Yichang 443002)
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2019年第7期1067-1074,共8页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(61876168,61403342,U61325019,61603341)
浙江省自然科学基金(LY18F030020)
湖北省水电工程智能视觉监测重点实验室开放基金(2017SDSJ03)
关键词
目标跟踪
鱼眼视频
图像畸变
相关滤波
target tracking
fisheye video
image distortion
correlation filter
