多特征融合的尺度自适应目标跟踪  

Scale adaptive target tracking based on multi⁃feature fusion

作　　者：张原园 艾斯卡尔·艾木都拉[1] 玛依热·依布拉音[1] ZHANG Yuanyuan;Askar Hamdulla;Mayire Ibrayim(School of Information Science and Engineering,Xinjiang University,Urumqi 830046,China)

机构地区：[1]新疆大学信息科学与工程学院,新疆乌鲁木齐830046

出　　处：《现代电子技术》2022年第23期33-40,共8页Modern Electronics Technique

基　　金：新疆维吾尔自治区自然科学基金面上项目(2020D01C045);国家自然科学地区基金项目(62166043);教育厅青年基金项目(XJE⁃DU2019Y007)。

摘　　要：针对目标跟踪过程中出现的平面旋转、快速运动、背景杂乱、遮挡等复杂状况导致的跟踪失败问题,提出一种基于背景感知相关滤波跟踪器的多特征融合的尺度自适应目标跟踪算法。首先在特征提取过程中通过在原有方向梯度直方图特征下加入了颜色特征和灰度特征以提高目标的特征识别;然后在滤波器更新阶段,利用峰值旁瓣比来减弱遮挡等复杂环境对跟踪的影响;最后通过尺度估计方法,在以目标位置为中心以不同尺度所产生的图像块中选择响应值最高的来更新目标。主要思想就是对目标跟踪流程两端进行改进,通过加入特征提取类型来突出目标特征,通过加入峰值旁瓣比和尺度估计来选择最好的响应结果进行更新,从而达到提高跟踪准确率的要求,同时跟踪速度也要满足实时要求。通过在OTB2015数据集中测试,算法整体的精确率和成功率都有了很好的提升,算法精确率达到84.1%,成功率达到79.5%,45 f/s能够满足实时跟踪的要求。在平面旋转、快速运动、遮挡等多个复杂环境下都有了很好的改善,结果表明该方法具有较高的理论价值和工程应用价值。In allusion to the tracking failure caused by complex phenomena,such as plane rotation,rapid movement,background clutter and blocking in the process of target tracking,a scale adaptive target tracking algorithm based on multi⁃feature fusion based on background perception correlation filtering tracker is proposed.In the process of feature extraction,color feature and gray feature are added to the original histogram of oriented gradient(HOG)feature to improve the feature recognition of the target.In the stage of filter update,the peak to sidelobe ratio(PSR)is used to reduce the influence of blocking and other complex environments on tracking.Among the image blocks which are generated by different scales and take the target position as the center,the one with the highest response value is selected to update the target by the scale estimation method.The main idea is to improve the target tracking process at both the beginning and the end,highlight the target features by adding the types of feature extraction,and select the best response result for update by adding the PSR and scale estimation,so as to meet the requirements of improving the tracking accuracy.Meanwhile,the tracking speed should meet the real⁃time requirements.By the test in dataset OTB2015,the results show that the overall accuracy and success rate of the algorithm have been greatly improved.The accuracy rate of the algorithm has reached 84.1%,and the success rate has reached 79.5%.45 frames per second can meet the requirements of real⁃time tracking.Its tracking effect has been improved in many complex environments,such as plane rotation,rapid motion and blocking.The results show that the work of this paper has high theoretical value and engineering application values.

关 键 词：目标跟踪 尺度自适应 多特征融合 相关滤波 特征提取 峰值旁瓣比 

分 类 号：TN911.73-34[电子电信—通信与信息系统] TP391.1[电子电信—信息与通信工程]