| 摘 要: 对地面车辆目标的视觉跟踪任务首要是满足实时性，其次是在复杂背景下对目标跟踪的鲁棒性。KCF算 法作为经典的判别式跟踪算法，凭借其高效的跟踪器学习效率，一直作为主流的实时跟踪算法之一。其中，搜索区域的 大小选取在很大程度上决定了能否生成稳定的跟踪器，然而对于不同尺寸的车辆目标，其最优的搜索区域大小通常是不 同的。为此，本文以标准数据集OTB2015作为车辆目标视频源，通过分辨率降采样来模拟多组不同尺寸的目标运动场 景，论证在不同距离下实现最优车辆跟踪的KCF算法参数配置，为长距离的车辆跟踪任务提供了参数依据。 |
| 关键词: KCF算法；目标跟踪；地面目标 |
|
中图分类号: TP391
文献标识码: A
|
|
| Research on Parameter Configuration of KCF in Vehicle Target Tracking |
|
HUANG Nan,LU Feng,WANG Qingzhao1,2,3
|
1.( 1.Army Academy of Armored Forces, Beijing 100072, China;2.
2.China North Vehicle Research Institute, Beijing 100072, China;3.
3.Army Armored forces research institute, Beijing 100072, China)
|
| Abstract: The first requirement of vehicle target tracking is real-time,and the robustness of the tracker in complex environment is then followed.As a typical discriminative tracker,KCF has been brought into focus for its high computational speed,and the robustness of which is depended on the size of searching area to a great extent.However,the optimal sizes of searching area are usually different for vehicle targets,because the sizes of target vary greatly.Therefore,the paper uses benchmark OTB2015 as the video source,simulating multiple groups of moving target scenes of different sizes by reducing resolution of the videos.Experiment analysis is carried out on KCF tracker,which aims to demonstrate the optimal parameter configuration of KCF for vehicle tracking at different distances. |
| Keywords: KCF algorithm;target tracking;vehicle |
