中文摘要：

针对估计相对深度的传统方法，易受噪声影响的问题，提出了一种基于光流的障碍物检测方法．基于摄像机前方局部地平面假设，通过尺度空间不变（SIFT）特征匹配得到单目图像序列前后帧的匹配点集，用随机抽样一致算法（RANSAC）鲁棒地估计出相机前方近似平面的单应性矩阵，并计算得到光流场，进而恢复出相对深度并建立障碍物图．由于避免了计算光流的一阶微分，该方法具有较好的鲁棒性．室内和室外环境的实验结果都表明，该算法能够恢复出相对深度，并对障碍物进行有效检测．

英文摘要：

Since the common approaches for estimating the time-to-contact employ the first order derivatives of the optical flow, such methods have the drawback of being sensitive to error in the estimates of optical flow. Based on the assumption that the robot was moving on a locally planar ground, a novel method for estimating the time-to-contact of obstacle was presented. Scale invariant feature transform （SIFT） matching of features between the image sequences captured from monocular camera was firstly applied. Then, the homography for the planar ground was robustly estimated based on the random sample consensus （RANSAC） algorithm, and the optical flow was estimated. After that, the time-to-contact was recovered and the obstacle map was finally constructed. Since the computation of the first order derivatives Of optical flow is completely avoided, this method is robust to image noise. Experimental results of indoor and outdoor scenes show that the method can effectively detect obstacles based on the recovered time-tocontact.