复杂背景下的运动目标检测一直是计算机视觉领域中一个具有挑战性的问题，本文提出一种基于多层背景模型的运动目标检测算法。该算法首先从视频序列的第一帧中提取每个像素的邻域样本，用于初始化背景模型，只需一帧图像即可完成背景模型的初始化；其次，为实现背景模型的自适应更新，引入随机采样技术，随机选取一个不匹配的码字，采用新的背景像素取而代之，避免错误分类的码字长时间驻留在背景模型中；为处理动态场景中多种干扰因素的影响，提出了多层背景模型策略，每个像素经过多层背景模型的逐层验证，保证了背景模型的精确性。实验结果表明，该算法能够有效克服复杂背景下的多种干扰因素影响，且检测率和识别率均高于现有经典算法。%Moving object detection under complex-background is always a challenging issue,and in order to defend these challenges,this paper proposed an algorithm named MMBM (Moving object detection based on Multi-layer Back-ground Model).First,samples are selected from neighbors of each pixel of the first frame to initialize background model. Only one frame image is needed for initialization.Second,in order to update the background model adaptively,random sam-pling technique is introduced,i.e.,selecting one code word randomly from the background model and then updating it with new background pixel,which overcomes the deficiency of the wrong classified code word overstaying in the background model.Multi-layer background model is proposed in order to overcome the influence of multi-disturbance in dynamic back-ground,in which one pixel is tested through multi-layers,so as to guarantee and improve the accuracy of background pixels. Finally,Experimental results show that this algorithm can overcome the influence of multi-disturbance existing in dynamic outside scenes effectively,and at the same time,achieve a higher detection rate and recognition rate over the existing classi-cal algorithms.
展开▼
