摘要
针对传统单目视觉里程计在特征提取过程中误匹配点过多,匹配精度低、运算量大、提出了一种基于改进SURF算法的单目视觉里程计模型,首先使用SURF算法对单目摄像头采集的图像的相邻两帧进行特征点的检测与匹配,然后用RANSAC算法对误匹配点进行剔除,提高匹配的精度,减少运算量,最终求出相邻两帧图像特征点匹配的旋转矩阵R和平移向量T,完成运动估计。实验结果表明,该模型在预估曲线运动和直线运动时的运算速度分别提高了11.2%和10.38%。
In traditional monocular visual odometries, there are many false match points, low match accuracy, large amount of computation during the feature extraction process of the traditional monocular visual odometry. This paper presents an improved monocular visual odometry model based on the SURF algorithm, The feature points between two adjacent frames are detected and matched with the SURF algorithm. The RANSAC algorithm is applied to remove the error feature points. Then, the rotation matrix R and shift vector T between two adjacent frames are calculated to accomplish the motion estimation. The experiment results show that the computing speed is accelerated by 11.2% and 10. 380/oo in the curve motion and the straight motion, respectively, with the proposed visual odometry estimating model.
出处
《电子测量技术》
2017年第5期185-188,200,共5页
Electronic Measurement Technology
基金
国家自然科学基金(60777018)资助项目
