摘要
针对核相关滤波的算法在复杂背景下的不足之处,提出一种多特征融合和尺度适应的相关滤波跟踪算法。分别使用HOG、CN2、HSV特征训练获得多个滤波器,在检测环节对多个响应图的结果自适应加权预测出目标位置,提高视觉跟踪算法在复杂背景下的适应能力;采用图像感知哈希算法快速匹配合适的目标尺度,针对模型更新环节,依据响应图的震荡程度优化模型更新策略,有效降低模型漂移发生的概率,减少模型更新的次数。通过OTB数据集与多种流行算法对比的实验结果表明了该算法的优越性。
To deal with the shortcomings of the kernel correlation filtering algorithm in the complex background,a multi-feature fusion and scaling adaptive correlation filtering tracking algorithm based on the kernel correlation filtering was proposed.HOG,CN2 and HSV feature training were used to obtain multiple filters respectively.In the detection link,the target position was predicted by adaptive weighting according to the results of multiple response graphs,which improved the adaptability of visual trac-king algorithm in complex background.The image perception hash algorithm was used to quickly match the appropriate target scale.In addition,the model update strategy was optimized according to the degree of oscillation in the response graph for the model updates,which effectively reduced the probability of model drift and the number of times of model updates.Experimental results of OTB data set and various popular algorithms show the superiority of the tracking algorithm.
作者
程语嫣
张九根
杨圣伟
CHENG Yu-yan;ZHANG Jiu-gen;YANG Sheng-wei(College of Electrical Engineering and Control Science,Nanjing Tech University,Nanjing 211816,China)
出处
《计算机工程与设计》
北大核心
2020年第12期3444-3450,共7页
Computer Engineering and Design
关键词
目标跟踪
多特征融合
相关滤波
图像感知哈希
尺度适应
target tracking
multi-feature fusion
correlation filtering
image aware hash
scale adaptation
