热红外图像序列中基于KCF和Mean-Shift定位的目标跟踪方法

Target Tracking Method Based on KCF and Mean-Shift Positioning in Thermal Infrared Image Sequence

针对热红外图像低信噪比(SNR)特性,提出了一种判别式热红外目标跟踪方法。首先,通过自适应组合核化相关滤波器(KCF)来获取目标位置,使用最有区别的特征集梯度和信道编码强度特征训练滤波器;然后,将经过训练的滤波器与感兴趣区域相关联,并将输出响应自适应地组合在一起,基于峰值定位目标。使用AdaBoost分类器对包含目标像素和背景像素的图像块进行训练,以分割连续帧中的对象;最后,通过Mean-Shift均值漂移算法寻找峰值以获得最优位置。对LTIR数据集中13个具有挑战性的红外视频进行了实验,结果显示提出的跟踪器在平均中心位置误差、距离精度和重叠精度等方面均优于其他跟踪器。

Aiming at the low signal-to-noise ratio(SNR)of thermal infrared images,a discriminant thermal infrared target tracking method is proposed.First,the target position is obtained by adaptively combining kerfization correlation filters(KCF),and the filters are trained using the most distinctive feature set gradients and channel coding strength characteristics.Then,the trained filter is associated with the region of interest and the output responses are adaptively combined together to locate the target based on the peak.The AdaBoost classifier is used to train image blocks containing target pixels and background pixels to segment objects in successive frames.Finally,the Mean-Shift mean shift algorithm is used to find the peak value to obtain the optimal position.The experiments were conducted on 13 challenging infrared videos in the LTIR dataset.The results show that the proposed tracker is superior to other trackers in terms of average center position error,distance accuracy,and overlay accuracy.