基于视差空间V-截距的障碍物检测  

Obstacle detection based on V-intercept in disparity space

作　　者：曹腾[1] 项志宇[1] 刘济林[1] 

机构地区：[1]浙江大学信息与电子工程学系浙江省综合信息网重点实验室,浙江杭州310027

出　　处：《浙江大学学报（工学版）》2015年第3期409-414,共6页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目(61071219)

摘　　要：为了更加高效地完成对基于双目立体视觉的驾驶系统的环境分析,提出新的应用于视差空间中分析障碍的"V-截距"方法.与传统在三维空间中的障碍检测方法不同,该方法直接在视差空间中进行检测,通过将障碍坡度信息转换为视差空间中V轴的截距实现检测.推导视差空间中的坡度-截距转换关系,划定合理的阈值区间.整个检测算法具有快速高效的特点,在原理上不受平面道路的假设约束,具有很强的实际应用价值.多种环境下的实验证明:该方法的障碍物检测效果可靠、稳定,检测速度是基于三维空间的方法的3.9倍.A new analysis method of obstacles in the disparity space with ＂V-intercept＂concept was proposed in order to analyze the environment around stereo vision based driving systems more effectively,Unlike the traditional methods which detected the obstacles in 3Dspace,this method worked directly in the disparity space,converting the obstacle＇s slope information into intercept on V axis in disparity space to achieve detection.Conversion relationship between slope and intercept was derived in the disparity space,and reasonable threshold interval was determined.The whole detection algorithm was fast and efficient,and was not affected by the assumption of flat road from the principle,giving a strong practical value.Experiments in multiple environments showed that result of this method was reliable and stable,which was3.9times faster than that of 3Dspace based method.

关 键 词：视差空间 立体视觉 障碍物检测 V-截距 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]