基于泛化傅里叶描述子的运动目标检测

Moving Object Detection Based on Generic Fourier Descriptor

提出了一种两层运动目标检测算法.基于普通模型的第一层检测从当前帧中粗略地分割出运动目标.第二层检测包括两部分:首先,从粗略分割和所有历史分割中提取运动目标的泛化傅里叶描述子,然后基于描述子相似性度量,从历史分割中提取和粗略分割相似程度较高的部分组成新模型,并基于新模型得到第二层检测结果.普通模型与新模型均使用概率建模方法,两层检测均使用图分割技术.实验结果表明了该方法的有效性.

This paper proposes a two-stage moving object detection algorithm.Rough detection of moving object is obtained in the first stage based on an ordinary probabilistic model in the current frame.There are two steps in the second detection stage.First,the generic Fourier descriptor is extracted from both the rough detection and past detections to describe the silhouette of the moving object.And then by comparing the silhouettes between current and past frames,silhouettes most similar to current frame are selected to form a new probabilistic model.Finally,the detection result is obtained according to the new probabilistic model in the second stage.Moreover,graph cut algorithm is used during the two-stage detection process.Experiment results show that this method is effective.