人物越界检测中的自适应背景建模

Adaptive Background Modeling for People Cross-border Detection

提出一种面向人物越界检测的自适应背景建模算法。首先为当前场景建立一个自适应的背景模型。然后,用去除背景图像的方法得到前景图像,再利用连通区域检测得到图像的运动区域。最后,采用跟踪与分类交互工作的方法达到人物检测的目的。跟踪所采用的是均值漂移(Mean Shift)算法,分类采用的是方向梯度直方图(Histogram of Oriented Gradient)和支持向量机(Support Vector Machine)的方法。实验结果表明,在各类不同场景下,文中方法比常用背景建模方法相比具有更好的适应能力,同时对场景中的光照变化、树叶剧烈摆动等问题也有较好的处理结果。采用此方法在得到的运动区域进行跟踪与分类,可以对人物进行准确的检测。

Propose an adaptive background modeling method for people cross-border detection. Firstly, construct an adaptive background model for existing scene. Secondly, the background image method of elimination is used to obtain the foreground image, and the connect- ed area detection is applied to get the motion are of image. Finally, the method of interaction work of tracking and classifying is used to reach the purpose of human detection. As the motion area is taken out, motion tracking can be achieved by Mean Shift method. Then HOG descriptors and SVM method are used to accomplish motion classification subsequently. According to the experimental results,in all kinds of different scenarios, compared with the common background modeling method, the method proposed has better ability to adapt, at the same time for the Scene problems of illumination change,leaves violent oscillation, also have good processing results. Movement area ob- tained by using this method for tracking and classification, the character can be accurately detected.