中文摘要：

针对多摄像机之间的光照变化、环境差异以及视角变化，文章提出基于外观模型和时空模型相结合的方法进行无重叠视域多摄像机间的目标跟踪。首先使用颜色标准化的方法减弱摄像机之间的颜色差异、学习目标的外观模型。颜色标准化过程无需人工标定目标间的对应关系，因此当摄像机数目增多时可以自动学习新的外观模型。然后由目标的外观相似度无监督学习得到与摄像机的环境、光照和视角无关的时空模型，即不同摄像机下进出口的分布模型、路径关系和转移时间概率分布。最后计算目标之间的外观相似度和时空相似度，实现多摄像机间的目标跟踪。在实际摄像机网络场景中应用该方法，结果表明了其有效性，并且具有较高的准确率。

英文摘要：

In view of the changes of the illumination, environment and viewpoint between different cameras, a tracking method across networked cameras with non-overlapping views based on appear- ance models and spatio-temporal models is proposed. Color normalization method which can weaken the color difference between different cameras is used to learn the appearance model, and when the number of the cameras increases, new appearance model can be learned automatically as this method does not require hand-labeled correspondence. And an unsupervised method utilizing people＇s appear- ance similarity is used to learn the spatio-temporal model which is irrelevant to the camera＇s lighting, environment and viewpoint. The spatio-temporal model includes the location distributions, the con- nected relationship and the transition time probability distributions of the entry/exit zones under dif- ferent cameras. Then the appearance similarity and the spatio-temporal similarity are calculated to re- alize the multi-camera object tracking. The proposed method is effective in the real camera network scenarios and achieves high tracking accuracy.