Abstract: A new background modeling method based on Sample Consensus (SACON) was proposed for detecting foreground objects. The method, which has higher accuracy than other five popular methods in the experiment when seven typical video sequences were adopted, is suit for detecting and tracking the moving target under complex circumstances. On pedestrian detection and tracking, considering the target to be non-rigid, a new target appearance model containing spatial domain and color information was built based on SACON. The experimental results show that the method can describe pedestrian feature exactly and can detect or track targets with barriers and similar color and small targets with high accuracy.

Key words: moving target detection and tracking

摘要： 提出了一种基于抽样一致性(SACON)的背景模型。采用七种典型视频序列进行了实验,SACON背景模型较目前常用五种背景模型具有较高的准确性,适用于复杂场景下的目标检测与跟踪。应用于人体检测与跟踪时,将目标作为非刚性物体进行处理,结合颜色和空域信息建立一种新的目标外观模型。实验证明,该方法能较准确的描述人体特征,即使在有遮挡、颜色相近及小目标等情况下也均能准确的对人体目标进行检测和跟踪。

关键词: 运动目标检测与跟踪, 随机抽样一致性, 背景模型, sample consensus, background model