基于随机聚类的复杂背景建模与前景检测算法

为了构建鲁棒的背景模型和提高前景目标检测的准确性, 综合考虑同一位置的像素点在时间上的关联性和与其相邻像素的空间关联性, 基于经典的ViBe算法中的随机聚类思想提出了一种复杂背景建模和前景检测方法. 利用样本一致性原理, 采用前n帧序列图像得到初始化背景, 避免了Ghost现象的发生; 根据实际复杂背景的动态反馈获取自适应聚类阈值和自适应更新阈值进行随机聚类, 从而实现了对动态背景的适应性; 通过全局扰动阈值和局部像素级判断阈值的结合, 实现了对光照缓慢变化、快速变化以及突然变化的免疫性, 准确地分割前景目标. 对多组数据集的测试结果表明, 本文算法较大地提高了背景模型对动态背景、光照变化及相机抖动的复杂背景的适应性和鲁棒性. 算法还能很好地适用于红外图像检测运动目标的场合, 扩展了本算法的应用范围. 在没有进行任何图像预处理和形态学后处理情况下, 得到的原始前景检测精度优于其他对比算法.

Complex background model and foreground detection based on random aggregation

In order to build a robust background model and improve the accuracy of the foreground object detection, we give a comprehensive consideration on the same location pixels of the relevance of time and the correlation of space with its adjacent pixels; and based on the classic ViBe of random algorithm ideas, a kind of complex background model and foreground detection method is proposed. Using the first n series of images to initialize the background model with the sample consistency principle, we can avoid the appearance of the “Ghost” phenomenon; and get the difference between each pixel and its multiple sample value in the background model, and then compute the sum and the average. The average shows the dynamic degree of the background point which is the corresponding pixel background of dynamic feedback information. We get the adaptive clustering threshold and adaptive updating threshold with the dynamic feedback to make random clusters realize the adaptability to dynamic background and combine the global disturbance threshold with the local pixel level judgment threshold to implement the immunity of illumination with slow changes, fast changes or sudden changes, so that we can segment the prospect target accurately. By selecting neighborhood pixels to update the neighborhood background randomly in terms of spatial information dissemination mechanism, a good detection effect is obtained in the case of camera shake. Through multiple sets of test data, experimental results show that this algorithm can significantly improve the adaptability and robustness of the background model such as dynamic backgrounds, illumination changes, and camera shake. The algorithm can well apply to the occasion of moving targets in infrared image detection, and expand its application range. Without any image preprocessing and morphological post-processing, the original detection accuracy of foreground is superior to other algorithms.