【作者】 郭敏； 严余松； 邹涛； 廖雪花；

【Author】 Guo Min;Yan Yusong;Zou Tao;Liao Xuehua;Southwest Jiaotong University;Sichuan Normal University;

【机构】 西南交通大学； 四川师范大学；

【摘要】 针对高速公路这一特定场景,提出一种基于网格时空背景差的运动车辆检测算法。该算法融合背景差分信息、基于时间信息的帧间差分信息、基于空间信息的背景差分信息,得到真实运动物体的运动种子点,将包含种子点的连通区域作为真实的前景目标。网格背景模型中将持久不变的背景图像作为长效背景,将含有新进入并停留或原来停留忽然运动的目标的图像背景作为短效背景,并不断地将短效背景更新到长效背景中去,来完善背景的更新和实现停留目标的检测。在背景差分里考虑了颜色分量值的变化方差,将此方差因子加入差分函数,来减少光照变化。通过仿真实验表明,该算法可以避免背景模型对场景的表征不足以及光照变化引起的干扰杂点和车道线误检测,并实现了对静止车辆的有效输出,及时发现高速公路上的交通事件。更多还原

【Abstract】 For the scene of highway,this paper proposes a highway vehicles detection algorithm based on grid space-time background difference.This algorithm combines background difference information,time information of the inter frame difference information,spatial information based on background difference based on information,get the moving seeds of real moving objects,will contain as a foreground object real connected region of the seed point.At the same time,using the grid background model,the background is divided into long and short acting backgrounds,the background image is constant as a long time background,a new image as the background of short acting background,and keep a short-acting background update into long-term background,to perfect the background update and implement for target detection.In the background difference in the variance for color component values,will this variance factor adding differential function,to reduce the illumination change.Simulation results show that the algorithm can avoid the background model for characterization of scene and illumination change caused by interference noise points and lane line error detection,and implements the effective output of stationary vehicles,found on the highway traffic incident in a timely manner.更多还原