基于LineMod的弱纹理多目标遮挡检测方法  

Weak Texture Multi-object Occlusion Detection Method Based On LineMod

作　　者：蒋云飞 柴琦 杨杰[1] JIANG Yun-fei;CHAI Qi;YANG Jie(College of Mechanical and Electrical Engineering, Qingdao University, Qingdao 266071, China)

机构地区：[1]青岛大学机电工程学院,青岛266071

出　　处：《青岛大学学报（自然科学版）》2022年第3期45-50,共6页Journal of Qingdao University(Natural Science Edition)

摘　　要：在遮挡情况下,LineMod算法对物体的识别能力较弱,为此提出一种基于LineMod的聚类匹配识别方法CMRL。首先利用特征点的4D附加特征与3D空间特征共同形成更深层次的7D特征块,分析各特征点的内在逻辑关联性以获取新的特征点分类方式,增加独特特征(物体边缘、角、弧等)在特征点分类过程中的影响比重;然后利用数据降维算法与K_(means)方法相结合的方式,将7D特征信息降维为新的3D特征信息,从而实现对特征块的重新分配,得到具有更多匹配信息的模板作为最终模板。实验结果表明,该方法在多目标遮挡的复杂场景下的鲁棒性、识别率以及准确率都有较大的提高。In the case of occlusion,LineMod algorithm is weak in object recognition,so a linemod-based clustering matching recognition method(CMRL)is proposed.Firstly,4D additional features and 3D spatial features of feature points are used to form deeper 7D feature blocks.Then new feature point classification is obtained by analyzing correlation of 7D feature of each feature point.The proportion of unique features(object edge,angle,arc,etc.)is increased in feature point classification process.After classification,data dimension reduction algorithm and Kmeans method are combined to reduce the dimension of 7D feature information to the new 3D feature information,so as to realize the reallocation of feature blocks and get the template with more matching information as the final matching template.Experimental result shows that the robustness,recognition rate and accuracy of the proposed method are improved greatly in complex scenes with multi-target occlusion.

关 键 词：聚类匹配识别 4D附加特征 7D特征块 独特特征 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]