摘要
针对传统的特征点匹配算法对于纹理少的识别目标特征点检测不理想的问题,基于任何物体都具有边缘特征,通过匹配边缘特征点,实现物体的边缘特征点匹配,采用Canny检测算法,提取边缘特征点。在此基础上,增加FAST角点检测算法,采用BRIEF描述算法实现对特征点构建描述子,通过2种检测算法融合,解决了纹理少的物体检测特征少的问题,并增加了物体特征点匹配数量。采用YOLO网络模型实现物体识别和框出物体区域。实验结果表明,基于双目图像多特征点融合匹配算法,很好地解决了纹理少的物体匹配特征少的问题,构建的物体识别和定位系统可以实现对训练的物体识别与定位。
Traditional feature point matching algorithm is not ideal for detecting feature points with fewer textures.Based on the fact that any object has edge features,edge feature points are matched by matching edge feature points. In this paper,Canny detection algorithm is used to extract Edge feature points. Based on this,the FAST corner detection algorithm is applied,and the BRIEF description algorithm is used to construct the descriptor for the feature points.By combining the two detection algorithms,it can solve the problem of less texture detection and fewer feature matching points. In this paper,YOLO network model is applied to achieve object recognition and box out the object area.Experimental results show that the proposed algorithm can solve the problem of less matching of features with fewer textures based on binocular multi-feature point fusion algorithm. The object recognition and localization system constructed in this paper can realize the recognition and location of training objects.
作者
王霖郁
蒋强卫
李爽
WANG Linyu1 , JIANG Qiangwei1, LI Shuang2(1. Harbin Engineering University, College of Information and Communication Engineering, Harbin 150000, China 2. The StateKey Laboratory of Satellite Navigation System and Equipment Technology, Shijiazhuang 050081, China)
出处
《无线电工程》
2018年第8期628-633,共6页
Radio Engineering
基金
国家"十三五"重点研发计划基金资助项目(SQ2016YFGX040104)
科学技术部高科技研究发展中心所属专项基金资助项目
关键词
YOLO网络
双目视觉
双目图像特征点匹配
物体识别与定位
YOLO network
binocular vision
binocular image feature point matching
target identification and location
