光学学报, 2018, 38 (5): 0515001, 网络出版: 2018-07-10   

特征融合的尺度自适应相关滤波跟踪算法 下载: 1038次

Scale Adaptive Correlation Filtering Tracing Algorithm Based on Feature Fusion
作者单位
上海交通大学电子信息与电气工程学院, 上海 200240
摘要
为提高相关滤波(CF)跟踪算法的稳健性,并克服传统CF方法无法处理目标尺度变化以及未利用图像颜色特征等问题,提出了一种基于融合颜色特征的尺度自适应相关滤波改进跟踪算法。首先,将目标搜索区域从3原色(RGB)颜色空间转换到Lab颜色空间,提取搜索区域的Lab 3通道颜色特征;然后,融合Lab颜色特征与方向梯度直方图(HOG)特征得到多通道特征,利用核相关滤波(KCF)计算输出响应图并寻找图中最大响应位置即目标位置;最后,基于Lab颜色特征建立尺度模型,从当前帧的目标位置处截取不同尺度图像块,通过将其与尺度模型比较得到目标尺度最优估计。实验选取35段公开彩色视频序列进行测试,并将所提算法与其他5种跟踪性能较好的跟踪方法进行对比。实验结果表明,所提方法对彩色视频序列中的目标遮挡、变形、尺度变化等现象具有良好的适应性,其平均性能优于对比方法,同时具有76 frame·s -1的实时跟踪速度。
Abstract
In order to improve the robustness of correlation filtering (CF) tracking algorithm, and overcome the problems that the traditional CF method cannot handle target scale change and does not use image color feature, a scale adaptive tracking algorithm is proposed based on correlation filtering improvement with fused color features. Firstly, the target searching area of the image is transferred from the color space of the three primary colors (RGB) to the Lab color space to obtain the Lab three channel features of the search area. Then, Lab color features and histogram of oriented gradients (HOG) feature are fused to obtain the image feature of multi-channel. The kernelized correlation filtering (KCF) is used to get the output response chart and find the position of maximum response, namely target location. Finally, the scale model is established through the Lab color feature, and the different scale image blocks are intercepted from the current frame target position. Optimal estimation of the target scale is obtained when we compare the scale image blocks with scale models. 35 pieces of open color video sequences are selected in experiments for testing, and the proposed method is compared with five other tracking methods with excellent performance. Experimental results show that the proposed method is well adapted to the phenomena of target occlusions, deformation and scale change in color video sequences,and its average performance outperforms the other compared methods. At the same time, the real-time tracking speed of the proposed method is 76 frame·s -1.

李聪, 鹿存跃, 赵珣, 章宝民, 王红雨. 特征融合的尺度自适应相关滤波跟踪算法[J]. 光学学报, 2018, 38(5): 0515001. Cong Li, Cunyue Lu, Xun Zhao, Baomin Zhang, Hongyu Wang. Scale Adaptive Correlation Filtering Tracing Algorithm Based on Feature Fusion[J]. Acta Optica Sinica, 2018, 38(5): 0515001.

本文已被 12 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!