Abstract: Multiple objects tracking is the basic and important problem in intelligent visual surveillance system. It is difficult to track multiple objects under occlusion only using two-dimensional visual features. Recently, the three-dimensional tracking system becomes the hot point, which can utilize three-dimensional features to recognize and track objects even under occlusions, and improve the accuracy of intelligent visual surveillance system entirely. The methods of tracking multiple objects based on three-dimensional visual system are introduced, which can be divided into three classifications described in detail. The merits and drawbacks of representative methods of each classification are introduced, and the future application and research trend is discussed.

Key words: multiple objects tracking, visual surveillance, three-dimensional visual modeling system

摘要： 多运动目标跟踪是智能视觉监控系统中的关键性的亟待解决的问题，采用二维视觉特征跟踪会在目标相互遮挡时丢失目标特征造成跟踪困难，近年来三维视觉跟踪系统越来越成为热点，利用三维特征能在多目标相互遮挡时更好地识别、跟踪目标，实现多目标遮挡时的精确跟踪，从而全面提高智能视觉监控系统的精确性。总结了近年来基于三维视觉系统的多运动目标跟踪方法，根据采用的不同三维视觉系统分为三类，将每类中具有代表性的方法进行了论述，分析了各典型方法的优缺点，最后提出了进一步研究的主要发展内容和趋势。

关键词: 多目标跟踪, 视频监控, 三维视觉建模