光学学报, 2017, 37 (5): 0515005, 网络出版: 2017-05-05   

基于特征融合和尺度自适应的干扰感知目标跟踪 下载: 786次

Distractor-Aware Object Tracking Based on Multi-Feature Fusion and Scale-Adaption
作者单位
南京理工大学自动化学院, 江苏 南京 210094
摘要
针对复杂场景下单一颜色特征稳健性差、存在类目标干扰及目标尺度变化的问题, 提出了一种基于特征融合和尺度自适应的干扰感知目标跟踪方法。首先, 综合目标、邻域背景、类似干扰区域的三原色(RGB)特征和改进的方向梯度直方图(HOG)特征计算得到干扰感知目标模型; 在搜索区域内逐像素点计算目标概率图, 然后进行密集采样得到候选目标, 利用目标概率图的概率值与距离值进行加权, 同时定位目标和类似干扰, 并更新目标模型; 采用RGB直方图建立尺度模型, 从当前帧图像上截取不同尺度的图像块并计算其RGB直方图, 通过与尺度模型比较, 获得最优尺度估计并更新尺度模型。实验结果表明, 提出的方法对复杂场景下的类目标干扰、局部遮挡、尺度变化等均具有很好的适应性, 同时距离精度、重叠精度等指标优于对比算法。
Abstract
Aiming at the tracking drift problem caused by the RGB feature, similar appearance and scale change in complex scenes, an improved method of distractor-aware object tracking based on multi-feature fusion and scale-adaption is proposed. Firstly, the distractor-aware object models are established base on the RGB feature and the modified (histogram of oriented gradient) HOG feature, which are extracted from object, surrounding background, and distractors. Secondly, the candidates are extracted by dense sampling in likelihood maps, which are obtained by calculating every pixel in the search region. The locations of the target and the distractor are obtained by vote score and distance score, also the model updating method is given. The RGB feature is extracted to establish a model scale, and the multi-scale feature pyramid method is used to get templates at different scales. The optimal scale is obtained by comparison between the model scale and the template scales. The experimental results indicate that the proposed algorithm can well adapt to environmental variation including distractors, partially blocking and scale variation and outperforms the compared tracking methods in terms of the distance precision and overlap precision.

李双双, 赵高鹏, 王建宇. 基于特征融合和尺度自适应的干扰感知目标跟踪[J]. 光学学报, 2017, 37(5): 0515005. Li Shuangshuang, Zhao Gaopeng, Wang Jianyu. Distractor-Aware Object Tracking Based on Multi-Feature Fusion and Scale-Adaption[J]. Acta Optica Sinica, 2017, 37(5): 0515005.

本文已被 18 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!