中文摘要：

针对摄像机与被检测目标同时运动时的目标检测问题,提出一种立体视觉与光流融合的运动目标检测算法。结合立体视觉技术设计了光流与自运动估计模型,运用车辆的运动信息和场景的深度信息估计因摄像机运动产生的自运动光流;采用多分辨率细化的Horn算法估计场景的混合光流;对混合光流和自运动光流进行差分运算,剔除背景中静态目标的运动干扰。经过一系列形态学滤波处理获得运动目标完整区域,依据光流的连通性对运动目标标号,并确定位置信息。以典型的交通场景为对象进行分析,实验结果表明该算法能有效地检测出动态背景下的运动目标。

英文摘要：

To aim at detecting moving objects with a dynamical background,a novel method based on fusion of stereovision and optical flow for detecting moving objects is proposed.Ego-motion optical flow caused by camera motion is estimated by establishing the optical flow and ego motion models which integrates vehicle motion and depth information.Mixed optical flow of the scene is calculated from improved Horn algorithm.The difference between mixed optical flow and ego-motion optical flow yields residual object flow,which is caused by object motion.Initial target area is extracted accordingly.An intact object and the distance information are detected by series of morphological operation in the connecting area.The approach is tested on typical traffic images and the results shows that the approach can effectively detect the moving targets under dynamic background.