张寅, 蔡旭阳, 闫钧华, 苏恺, 张琨. 联合时空信息和轨迹关联的空中多目标检测[J]. 武汉大学学报 ( 信息科学版), 2020, 45(10): 1533-1540. DOI: 10.13203/j.whugis20190359
引用本文: 张寅, 蔡旭阳, 闫钧华, 苏恺, 张琨. 联合时空信息和轨迹关联的空中多目标检测[J]. 武汉大学学报 ( 信息科学版), 2020, 45(10): 1533-1540. DOI: 10.13203/j.whugis20190359
ZHANG Yin, CAI Xuyang, YAN Junhua, SU Kai, ZHANG Kun. Aerial Multi-target Detection Based on Spatial-Temporal Information and Trajectory Association[J]. Geomatics and Information Science of Wuhan University, 2020, 45(10): 1533-1540. DOI: 10.13203/j.whugis20190359
Citation: ZHANG Yin, CAI Xuyang, YAN Junhua, SU Kai, ZHANG Kun. Aerial Multi-target Detection Based on Spatial-Temporal Information and Trajectory Association[J]. Geomatics and Information Science of Wuhan University, 2020, 45(10): 1533-1540. DOI: 10.13203/j.whugis20190359

联合时空信息和轨迹关联的空中多目标检测

Aerial Multi-target Detection Based on Spatial-Temporal Information and Trajectory Association

  • 摘要: 为解决空中多目标检测中目标在图像中所占像素数少、目标相互毗邻和相互遮挡的难题,提出了联合时空信息和轨迹关联的空中多目标检测算法。该算法首先利用基于像素的背景建模算法得到目标空间信息,结合邻近帧差算法提取的目标时间信息,联合得到目标的时空信息图;然后,利用卡尔曼预测器对目标位置进行预测,利用匈牙利匹配算法关联目标得到目标轨迹,根据目标轨迹补充漏检目标, 提高目标的查全率;最后,分段提取目标轨迹特征,根据轨迹特征进行判断,滤除虚警目标, 提高目标的查准率。采用空中多目标视频进行实验验证,结果表明,本文算法检测性能优良,查全率高于96%,查准率高于98%,F测度高于97%。

     

    Abstract: To tackle the problem that the targets in the image have few pixels and the targets are adjacent to each other and occluded each other, an aerial multi-target detection algorithm is proposed based on spatial-temporal information and trajectory association. Firstly, the pixel-based background modeling algorithm is used to obtain the target space information, and the neighboring frame difference algorithm is used to obtain the target time information. The time and space information are combined to get a spatial-temporal information map. Secondly, the Kalman predictor is used to predict the target position, and the Hungarian matching algorithm is used to correlate the target to obtain the target trajectory. Based on the target trajectory, the missed detection targets are supplemented to improve the target recall rate. Finally, based on the target trajectory features which are extracted in segments, the false alarm targets are filtered out to improve the target precision rate. Experimental videos with aerial multi-target are adopted for experimental verification, and the experimental results show that the proposed algorithm has good detection performance, with the recall rate higher than 96%, the precision rate higher than 98%, and the F-measure higher than 97%.

     

/

返回文章
返回