交通路口监控视频跨视域多目标跟踪的可视化
发布时间:2018-05-31 02:31
本文选题:跨视角 + 多目标跟踪 ; 参考:《计算机学报》2018年01期
【摘要】:跨视域大场景的多目标跟踪与展示是智能监控的基本需求之一.该文设计了一种基于视域拼接的跨视域多目标跟踪的可视化算法,借助于视频场景中几何信息实现视域拼接,从而实现将交通路口不同视角监控视频中的跟踪目标在统一的视场下展示.算法主要包含四个步骤:视域背景拼接、目标检测、跨视域多目标跟踪以及可视化显示.其中,视域背景拼接步骤利用交通场景背景图像几何信息辅助的半交互方式确定特征点对,计算不同视角到参考视域平面的单应变换矩阵,并利用SPHP算法保形后对所有配准图像线性融合以完成背景拼接;目标检测步骤利用ViBe背景建模算法分离目标,并进行阴影消除以提高检测准确性;跨视域多目标跟踪则结合各个视角到拼接视域平面的映射关系获得目标的定位信息,采用Kalman滤波和最小均方的轨迹匹配实现跨视域多目标的一致性跟踪;最后可视化显示步骤则在拼接的视域背景上对跟踪目标进行动态可视化展示.实验结果表明,该算法能够在统一视场下展现多个视域的监控场景信息,更方便于交通路口的监控.
[Abstract]:Multi-target tracking and display is one of the basic requirements of intelligent monitoring. In this paper, a visual algorithm of multi-object tracking based on visual mosaic is designed, which realizes visual mosaic by means of geometric information in video scene. In order to achieve the traffic intersection in different visual angle surveillance video tracking targets in the unified field of view display. The algorithm consists of four steps: background mosaic, target detection, multi-target tracking and visual display. Among them, the visual background stitching step uses the semi-interactive mode assisted by the geometric information of the traffic scene background image to determine the feature point pairs, and calculates the monoclinic transformation matrix from different angles of view to the reference horizon plane. The SPHP algorithm is used to preserve the shape of all the registered images to complete the background stitching, the target detection step uses the ViBe background modeling algorithm to separate the target, and the shadow is eliminated to improve the accuracy of the detection. Based on the mapping relationship from each visual angle to the spliced horizon plane, the location information of the target is obtained, and the Kalman filter and the minimum mean square trajectory matching are used to realize the consistent tracking of the multi-targets across the horizon. Finally, the visual display step is used to visualize the tracking object on the background of the splicing field of view. The experimental results show that the algorithm can display the scene information of multiple visual fields under the unified field of view, and it is more convenient for traffic intersection monitoring.
【作者单位】: 北京理工大学计算机学院;
【基金】:国家杰出青年科学基金(61425013) 国家自然科学基金面上项目(61472035)资助
【分类号】:TP391.41
【相似文献】
相关期刊论文 前10条
1 路红;费树岷;郑建勇;张涛;;基于行为和部分观测的多目标跟踪(英文)[J];Journal of Southeast University(English Edition);2008年04期
2 蒋恋华;甘朝晖;蒋e,
本文编号:1957955
本文链接:https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/1957955.html