全景视觉非线性核相关滤波目标跟踪技术  

Omni-directional visual object tracking using nonlinear kernelized correlation filters

作　　者：朱齐丹[1] 韩瑜[1] 蔡成涛[1] ZHU Qidan;HAN Yu;CAI Chengtao(College of Automation,Harbin Engineering University,Harbin 150001,China)

机构地区：[1]哈尔滨工程大学自动化学院,黑龙江哈尔滨150001

出　　处：《哈尔滨工程大学学报》2018年第7期1220-1226,共7页Journal of Harbin Engineering University

基　　金：国家自然科学基金项目(61673129);黑龙江省自然科学基金项目(F201414);中央高校基本科研业务费专项资金项目(HEUCF160418)

摘　　要：为了实现全景视觉下目标跟踪,本文提出了基于非线性核相关滤波的全景视觉目标跟踪算法,该算法将岭回归与循环样本矩阵和经典的相关滤波联系在一起。通过核技巧,建立了非线性回归函数,解决了全景成像畸变的影响。通过在极坐标系下构造傅里叶形式的梯度方向直方图特征描述子(HOG),保证了全景目标的旋转不变性。针对全景成像特点设计自适应机制和基于极坐标表示的目标搜索机制。以OTB100库中存在平面内旋转情况的图像序列与原方法进行对比实验,本文算法跟踪精度提高13.5%。并在全景图像上进行实验,相比于原方法,本文算法能够实现有效全景视觉目标跟踪。To achieve an omni-directional visual tracking,an improved object tracking method based on nonlinear kernelized correlation filters is presented in this paper. This method combines ridge regression with circulant data matrix and correlation filter. Using the kernel trick,a nonlinear regressive function is established to resolve the distortion in omnidirectional vision. By building rotation-invariant histograms of oriented gradients descriptors using Fourier in polar coordinate,the target rotating invariance is guaranteed. Adaptive kernel correlation filter and a polar coordinate representation based detection method are designed considering the features of omni-directional vision. According to experiments on images with in-plane rotation in OTB100,the tracking precision of the proposed method is 13. 5% higher than that of the original method. According to experiments on omnidirectional videos,the proposed method is more effective than the original method for omni-directional visual object tracking.

关 键 词：全景视觉 目标跟踪 非线性 核相关滤波器 旋转不变性 HOG 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]