激光与光电子学进展, 2018, 55 (5): 051102, 网络出版: 2018-05-16
基于靶标区域分割的双目定位系统研究与实现 下载: 710次
Research and Implementation of Binocular Location System Based on Region of Interest Segmentation
成像系统 投影法 图像分割 加速稳健特征算法 特征点匹配 imaging systems projection method image segmentation speed up robust features algorithm feature point matching
摘要
针对传统双目视觉定位系统立体匹配困难、计算量大、效率低的问题,设计了一种基于靶标区域(ROI)提取的双目定位系统,以改善定位精度。首先,利用张氏标定法标定双目摄像机,并通过Bouguet算法对整个系统进行立体校正;然后利用直方图阈值法和投影法分割图像中的ROI,以大幅降低后续特征匹配的计算量;最后采用加速稳健特征算法对左右摄像机图片进行亚像素级角点的提取与匹配,并结合标定结果获得精确的定位。实验结果表明,ROI内特征点的匹配准确度可以达到90%以上,系统的定位时间在700 ms以内,可以满足系统实时性的要求。该方法主要针对ROI进行特征点提取和匹配,避免了不必要的全局图像处理,从而将匹配速度从秒级缩短至毫秒级。
Abstract
Aiming at the problems of difficulty in stereo matching, large computation amount, and low efficiency of the traditional binocular vision positioning system, a binocular positioning system based on the region of interest (ROI) extraction is designed to improve the positioning accuracy. The binocular camera is calibrated by calibration method of Zhang, and the system is calibrated by Bouguet algorithm. The ROI in image is divided by the histogram thresholding method and projection method, and the computation amount of the feature matching is reduced greatly. The speed up robust features algorithm is used to extract and match the camera image subpixel corners, and accurate positioning results are obtained when the calibration results are combined. Experimental results show that the matching accuracy of feature points in ROI can reach more than 90%, and the positioning time of the proposed system is less than 700 ms, which can meet the real-time requirements of the system. This proposed method is mainly used for ROI feature point detection and matching. The global image processing is avoided, and the feature points matching speed can be reduced from seconds to milliseconds magnitude.
刘远远, 冯鹏, 龙邹荣, 俞鹏炜, 李鑫韬, 魏彪. 基于靶标区域分割的双目定位系统研究与实现[J]. 激光与光电子学进展, 2018, 55(5): 051102. Liu Yuanyuan, Feng Peng, Long Zourong, Yu Pengwei, Li Xintao, Wei Biao. Research and Implementation of Binocular Location System Based on Region of Interest Segmentation[J]. Laser & Optoelectronics Progress, 2018, 55(5): 051102.