Abstract: Due to the difference and the separation of the multiple disjoint camera views, the same object’s appearance in different cameras may be very different, which result in matching moving objects across disjoint camera views is a challenging task. This paper proposes an illumination-tolerant appearance representation, which is capable of matching objects in disjoint camera views. This method is based on a first cluster and K-means cluster to abstract object’s major color spectrum histogram, which is introduced to represent a object. Then it uses EMKM to weaken K-means sensitive to the initialization of center values. Finally a feature similarity measurement algorithm is proposed to measure the similarity of any two objects. The paper also introduces SIFT to match objects when can’t tell one from another. Meanwhile, this method can be used in matching objects across overlapping camera views, by means of integrating with other methods to improve the accuracy. Experimental results show the accuracy of the proposed method.

Key words: disjoint camera views, major color spectrum histogram, similarity measurement, Scale Invariant Feature Transform（SIFT）

摘要： 在无重叠视域的多摄像机监控中，由于不同摄像机的视域差别和视域分离，同一运动目标在不同的视域中的成像可能会非常不同，因此在这种情况下对运动目标进行匹配是一项具有挑战性的工作。提出了一种可以容忍光照的不同，在无重叠视域的多摄像机下进行目标匹配的方法。该方法经过初始聚类和K-means聚类对目标进行主颜色谱的提取，利用EMKM算法改善K-means对初始中心点的依赖性，把提取出来的主颜色谱直方图作为目标的特征，然后利用特征相似度测量来判定任意两个物体之间是否匹配；当无法对某些物体进行准确匹配时，再利用SIFT特征进行下一步匹配。该方法也可以用于有重叠视域的多摄像机目标匹配中，通过与其他匹配方法相结合，提高匹配的准确度。实验结果证实了该方法具有较高的准确度。

关键词: 无重叠视域, 主颜色谱直方图, 相似度测量, 尺度不变特征变换（SIFT）