高级检索
    周良毅, 王 智, 王营冠. 基于动态遮挡阈值的多视角多目标协作追踪[J]. 计算机研究与发展, 2014, 51(4): 813-823.
    引用本文: 周良毅, 王 智, 王营冠. 基于动态遮挡阈值的多视角多目标协作追踪[J]. 计算机研究与发展, 2014, 51(4): 813-823.
    Zhou Liangyi, Wang Zhi, Wang Yingguan. Multi-View Cooperative Tracking of Multiple Mobile Object Based on Dynamic Occlusion Threshold[J]. Journal of Computer Research and Development, 2014, 51(4): 813-823.
    Citation: Zhou Liangyi, Wang Zhi, Wang Yingguan. Multi-View Cooperative Tracking of Multiple Mobile Object Based on Dynamic Occlusion Threshold[J]. Journal of Computer Research and Development, 2014, 51(4): 813-823.

    基于动态遮挡阈值的多视角多目标协作追踪

    Multi-View Cooperative Tracking of Multiple Mobile Object Based on Dynamic Occlusion Threshold

    • 摘要: 多移动目标相互运动过程中产生的相互遮挡造成目标间的复杂空间关系,为视觉传感器网络的多目标信息融合、协同处理和协作追踪带来困难,提出动态遮挡阈值的多视角多目标协作追踪的多视角多目标协同追踪算法.为克服多视觉传感器中的信息融合及目标一致性表征问题,在基于贝叶斯理论的多视角移动追踪中引入遮挡变量,来描述多目标间的空间关系;通过分析不同视角包含的目标信息量及其对于融合的贡献,给出了遮挡阈值的动态表达式;通过动态调整和比较遮挡阈值来判断目标间的遮挡状态,改进了目标遮挡的判决标准和公共平面中的目标融合特征;并通过结合改进粒子滤波得到基于遮挡变量的多视角目标协作追踪算法,保证了追踪系统的稳定性.实验结果表明,引入遮挡变量以及动态遮挡阈值有效解决了传统追踪算法中的目标不一致性、尺寸变化等难题,提高了目标追踪精度,在目标被遮挡的情况下仍然能够保持良好的追踪效果.

       

      Abstract: The occlusion among mobile objects during movement causes a complicated space relationship, which makes multi-view information fusion, cooperative processing and tracking difficult problem. A multi-view object tracking algorithm is proposed by combining modified fusion feature with dynamic occlusion threshold and the improved particle filter in this paper. To solve the objects characteristic uncertainty in multi-view information fusion, occlusion variable is introduced to describe space relationship among mobile objects. Through analyzing the contribution on information fusion of objects in different scene planes with the help of the positions and scales of objects, and the homography transform and sensor model, the expression of dynamic occlusion thresholds are given. After dynamically adjusting and comparing occlusion threshold, an accurate occlusion state among multiple mobile objects is obtained, which is utilized in objects characteristic fusion within common scene. Then an improved particle filter tracking algorithm based on Bayesian theory is proposed, which utilizes the modified characteristic fusion and makes the tracking system more robust to object occlusion. Experiments show that the proposed occlusion variable and dynamic occlusion threshold can effectively solve the problems of objects characteristic uncertainty and scale variation, and good tracking precision is maintained even when objects are occluded.

       

    /

    返回文章
    返回