改进视觉背景提取ViBe算法的目标检测

Improved visual background extractor Vi Be algorithm for detecting objects

针对视觉背景提取Vi Be算法消除鬼影时间长、对动态场景适应性弱、对光线变化敏感等问题,提出一种适应性强的改进算法。在背景模型初始化时,通过对多帧图像随机选取像素点并累加判断后,形成与实际吻合度达86.78%的背景模型;提出衡量背景动态程度因子,根据其值获取图像的自适应聚类、更新阈值,提高了算法在动态背景下的检测精度;考虑到光线变化对检测结果的影响,提出衡量图像亮度因子并应用于聚类检测,增强了算法对光照的鲁棒性。与其他算法进行对比实验后表明,改进算法在不同场景中能有效检测目标物体,具有较好的适应性。

As the ViBe algorithm has some ghost which will be a long time, weak adaptability to dynamic scene and strong sensibility to light changes, it proposes an improved algorithm which is more adaptable in this paper. In the background model initialization, the background model with the actual degree of agreement reaches 86.78%, it is formed after the multi frame images are randomly selected and accumulated judgment. Then it puts forward the measure background dynamic degree factor, according to its value, it can obtain the image’s adaptive clustering and threshold update, in order to improve the detection accuracy in complex background. Considering the effect of the change of light on the detection results, it puts forward the image brightness factor and uses it in the cluster detection and increases the robustness of the algorithm to the illumination. Compared with other algorithms, improved algorithm can detect target object accurately in different scenarios and has better adaptability in kinds of scenes.