【摘要】 以选择性背景更新为基础,提出由相似度决定更新速度的新方法。综合颜色差异、连通性和车辆占空比等属性建立分割双阈值的能量函数,并利用模拟退火算法求取全局最优阈值。用最优分割双阈值对车辆图像进行阈值粗分割,再以条件随机域工具整合视频空间中的时间和空间信息建立随机标号域模型,并计算最大后验概率确定粗分割中各像素点的标号(细分割)。运用色彩空间变换,消除室外环境下多种干扰的影响,增大前景(车辆)与背景图像的颜色差异,提高图像分割效率。仿真试验表明,车辆视频图像通过粗细两次分割后能够得到较好的车辆区域,与其他方法相比本文方法具有更强的准确性、鲁棒性和实时性。更多还原

【Abstract】 Based on selective background updating,a new updating approach which determining updating speed by similarity was proposed.The energy function for dual-threshold segmentation was constructed based on color difference,connectivity and duty ratio,and the global optimal threshold was determined by simulated annealing algorithm.Coarse segmentation for the vehicle images was performed using optimal dual-threshold,a random label domain model was built by the time and space information in video spaces integrated with conditional random field tools,and refined segmentation,i.e.,fixing the labeling of pixel dots in coarse segmentation by computing the maximum posterior probability(MAP),was performed.Using the model of color space transformation,the problems of several disturbances in outdoor environment were dealt with,the color difference between foreground(vehicles) and background image was augmented,and the segmentation efficiency was improved.The simulation results show that good vehicle area can be obtained after the coarse and refined segmentation for the vehicle video image.Compared with other approaches,the proposed approach is more accuracy,robust and real time.更多还原