一种利用跟踪信息的星载视频车辆检测方法

扫码查看

原文链接

NETL
NSTL
万方数据

中文摘要：卫星视频可快速提供城市级车辆动态信息,为城市信息学、智慧城市、城市发展规划等提供可靠数据.然而当前星载视频背景中包含大量噪声与伪运动目标,运动车辆呈现为暗弱微小的斑点,使得准确检测出运动车辆非常困难.为此,本文提出一种利用跟踪信息的星载视频车辆检测方法.该方法首先利用D-LinkNet网络提取道路掩膜,剔除背景中噪声与伪运动目标的干扰;然后利用前景检测算法进行车辆初步检测;之后对初步检测结果进行帧间运动车辆跟踪,最后对因漏检而跟踪失败的运动车辆目标进行基于跟踪信息的区域精检测,获得高准确度的运动车辆目标.为了验证本文方法的有效性,对已有数据集进行更为精确的重注,新增2120个运动车辆目标,得到重注数据集.在重注数据集与吉林一号视频数据上,将本文方法与多种当前成熟方法进行了对比试验.试验结果表明:①本文检测方法可显著提升星载视频运动车辆检测召回率;且在所有试验中,均获得了最高F值;②在本文方法分步骤性能评估试验中,印证了本文方法掩膜对背景区域误检的剔除作用以及跟踪信息对漏检目标再次检测的促进作用;③在跟踪阶段测试试验中,本文方法获得了更好的多目标跟踪准确度和跟踪器身份维持能力.因此,本文方法在卫星视频的运动车辆检测中具有较好的稳健性,对城市级车辆动态信息获取提供了一种新的方法与思路.

外文标题：An Onboard Video Vehicle Detection Method Using Tracking Information

外文摘要：Satellite video can quickly provide city-level information on vehicle dynamics,providing reliable data for urban informatics,smart cities,and urban development planning.However,the image background captured by current video satellites contains a large amount of noise and pseudo-motion targets due to perspective shifts.The moving vehicles are presented as dark,weak,and tiny spots in the on-board video,which makes it very difficult to accurately detect the moving vehicles.In this paper,we propose a motion vehicle detection method using tracking information in the on-board video.The method first extracts the road mask in the first frame of the video using the D-LinkNet to eliminate the interference of noise and pseudo-motion targets in the background.Then the foreground detection algorithm is used for the preliminary detection of moving vehicles in the road mask.The preliminary detection results are used for the inter-frame tracking of moving vehicles.Finally,the tracking information-based regional fine detection is performed for the moving vehicle targets that failed to be tracked due to omission of detection,to obtain highly accurate moving vehicle targets.In order to verify the effectiveness of the proposed method,we accurately reannotate the existing dataset by supplementing the missing targets and adding 2120 new moving vehicle targets in the dataset.Based on the reannotated dataset and two study areas with Jilin-1 video data,the proposed method is compared with a bunch of existing methods using different indexes.The results show that:(1) the proposed detection method can significantly improve the recall rate of on-board video motion vehicle detection and obtains the highest F-value in all tests;(2) by evaluating the step-by-step performance of the proposed method,we found that the proposed method's mask has the effect of eliminating the false detections in the background region and the tracking information has the effect of facilitating the redetection of omitted targets;and (3) in the test of the tracking phase,the proposed method obtains higher multi-target tracking accuracy and tracker identity maintenance capability.Therefore,the method proposed in this paper has good robustness in detection of moving vehicles from satellite video and provides a new method and idea for the acquisition of vehicle dynamic information at the city level.

外文关键词：

video satellitetracking informationroad constraintsvehicle detectionmulti-target trackingdata set re-injectionJilin-1KNN

作者：

李明、范大昭、董杨、纪松、李东子

展开 >

作者单位：

中国人民解放军战略支援部队信息工程大学,郑州 450001

关键词：

视频卫星跟踪信息道路约束车辆检测多目标跟踪数据集重注吉林一号 KNN

基金：

国家自然科学基金项目国家自然科学基金项目嵩山实验室项目高分遥感测绘应用示范系统(二期)

项目编号：

4197142742371459221100211000-542-Y30B04-9001-19/21

出版年：

2024

DOI：

10.12082/dqxxkx.2024.230372

地球信息科学学报

中国科学院地理科学与资源研究所

地球信息科学学报

CSTPCD北大核心

影响因子：1.004

ISSN：1560-8999

年,卷(期)：2024.26(10)