特征融合的尺度估计颜色名跟踪算法

针对颜色名跟踪器在跟踪过程中运动目标在遮挡、尺寸变化等干扰下不能有效跟踪的问题,提出一个在颜色名(color name,CN)跟踪器框架下融入LBP纹理特征的快速尺度估计算法。运用自适应融合CN颜色特征和LBP纹理特征,利用特征融合后的最高输出响应作为预测目标位置信息;利用尺度滤波器对预测到的目标信息做出尺度估计;结合响应图的峰值旁瓣比,对目标模型做出更新调整。将改进算法在OTB-2013视频集中和近年来流行的跟踪算法做对比实验,实验结果表明,改进算法在尺度适应性和遮挡方面优于大部分算法。

Scale estimation color name tracking algorithm for feature fusion

To solve the problem that the color name tracker can not effectively track the moving target under the disturbance of occlusion and size change during the tracking process,a fast scale estimation algorithm incorporating LBP texture features under the color name(CN)tracker framework was proposed.Adaptively fusion of CN color features and LBP texture features was used,and the highest output response after feature fusion was taken as the predicted target position information.The scale filter was used to make scale estimation of the predicted target information.Combined with the peak to sidelobe ratio of the response map,the target model was updated.The improved algorithm was compared in the OTB-2013 video set with the popular tracking algorithms in recent years.The experimental results show that the improved algorithm is superior to most of those algorithms in scale adaptability and occlusion.