| 基于滤波器自适应更新的机场目标跟踪算法 |
| |
| 引用本文： | 杨临风,牟睿,黎新,李炜.基于滤波器自适应更新的机场目标跟踪算法[J].交通信息与安全,2022,40(1):72-79. |
| |
| 作者姓名： | 杨临风 牟睿 黎新 李炜 |
| |
| 作者单位： | 1.中国民用航空飞行学院民机火灾科学与安全工程四川省重点实验室 四川 广汉 618307 |
| |
| 基金项目： | 四川省科技计划重点研发项目;国家重点研发计划 |
| |
| 摘 要： | 机场场面目标跟踪常面临目标遮挡、背景干扰、低分辨率等因素的影响，导致跟踪准确性降低甚至丢失跟踪目标。针对以上问题，研究了基于滤波器自适应更新的机场目标跟踪算法。选取跟踪目标的颜色特征和深度特征，通过插值算子进行多特征融合，再将融合特征与之对应的滤波器进行卷积求和计算各区域置信度，置信度高的区域即为跟踪目标位置。为提高跟踪准确性，利用峰值旁瓣比与平均响应峰值能量建立了跟踪结果校验机制，并设计了1种滤波器自适应更新策略，使滤波器能够自适应调整学习速率，仅在结果可靠时更新。在西南某机场采集的视频数据集上进行测试，结果表明:算法在目标特征不明显或发生变化时具有更好的性能，在目标遮挡和背景干扰等9种因素下的跟踪性能有较大提升，整体精确度和成功率分别达到0.834和0.828，较原ECO算法分别提升了11.35%和11.29%，且均优于文中提到的其他5种经典算法。
|
| 关 键 词： | 智能交通 场面监视 自适应更新 相关滤波 卷积神经网络 多特征融合 |
| 收稿时间： | 2021-08-11 |
Development of an Object Tracking Algorithm for Airports Using Adaptive Filter Update Technique |
| |
| YANG Linfeng,MOU Rui,LI Xin,LI Wei.Development of an Object Tracking Algorithm for Airports Using Adaptive Filter Update Technique[J].Journal of Transport Information and Safety,2022,40(1):72-79. |
| |
| Authors: | YANG Linfeng MOU Rui LI Xin LI Wei |
| |
| Abstract: | Tracking objects over airport surface is often hindered by the factors such as occlusion, background clutter and low resolution, which often result in reduced tracking accuracy or even loss of tracked objects. In order to mitigate the above problems, an object tracking algorithm for airports based on adaptive filter update is developed. First, the color and convolutional neural network feature of the tracking object are selected. Based on these features, multi-feature fusion is performed through an interpolation operator. Then, the fusion feature and its corresponding filter are convolved and summedin order to calculate the confidence level of each region.Theregion with a high confidence level is then identified as thelocation of the tracked object. By using the peak to side-lobe ratio and the average peak-to-correlation energy, a verification method is developed to improve the tracking accuracy. Furthermore, a self-adaptive updating algorithm is designed to adjust the learning rate of the filter and updated only when the results are reliable. According to the results obtained using a video dataset collected at an airport in Southwest China, the proposed algorithm has a better tracking performance when the object features change or are unclear, and the results also indicate the tracking performance is significantly improved under 9 different factors, such as occlusions and background clutter. The overall accuracy and success rate are 0.834 and 0.828 respectively, which are higher than that of the original ECO algorithm by 11.35% and 11.29%, and are superior to the other five classical algorithms. |
| |
| Keywords: | |
| 本文献已被 万方数据 等数据库收录！ |
| 点击此处可从《交通信息与安全》浏览原始摘要信息 |
|
点击此处可从《交通信息与安全》下载全文 |
|
