红外与激光工程, 2020, 49 (S2): 20200109, 网络出版: 2021-02-05  

基于Mean Shift点法向量分类的目标三维姿态估计

3D pose estimation of target based on Mean Shift point normal vector classification
作者单位
1 哈尔滨工业大学 光电子技术研究所 可调谐(气体)激光技术重点实验室,黑龙江 哈尔滨 150001
2 哈工大(北京)军民融合创新研究院有限公司,北京 101300
3 复杂系统控制与智能协同技术重点实验室,北京 100074
4 中国空空导弹研究院,河南 洛阳 471009
摘要
目标三维姿态信息在目标运动分析、目标识别和目标跟踪等领域的应用越来越广泛。现有的OPDVA算法采用基于距离的K-means算法对点法向量进行分类后确定目标坐标系MCS的正方向,求取目标三维姿态角。针对点法向量分类效果不理想的情况,提出了基于Mean Shift点法向量分类的目标三维姿态估计算法(PEMSPNC)。该算法利用不依赖初始参数设定、基于密度聚类的Mean Shift算法,对密度分布不同的不同平面点法向量分类,寻找密度最大处点法向量做为每类代表法向量确定MCS的正方向,然后计算目标姿态角,并根据目标姿态估计结果计算目标尺寸。采用矩形拟合法、OPDVA和PEMSPNC算法分别对仿真和实测目标距离像进行实验。实验结果表明:采用PEMSPNC算法得到的姿态估计结果误差最小,相比于OPDVA算法,平均误差降低了0.443 4°,且对实测数据有较好的处理结果。
Abstract
The three-dimensional pose information of target is more and more widely used in the fields of target motion analysis, target recognition and target tracking. The K-means algorithm based on distance was used to classify the point normal vector by the existing OPDVA algorithm, then the positive direction of the target coordinate system MCS was determined, and the three-dimensional pose angle of the target was obtained. In view of the unsatisfactory effect of point normal vector classification, a three-dimensional pose estimation algorithm based on mean shift point normal vector classification (PEMSPNC) was proposed. The Mean Shift algorithm was used, which did not depend on the initial parameter setting and based on density clustering, to classify the normal vectors of different plane points with different density distribution, and to find the normal vector of the point with the maximum density as the representative normal vector of each class to determine the positive direction of MCS, then the pose angle of the target was calculated, and the target size according to target pose estimation results was computered. The rectangle fitting method, OPDVA and PEMSPNC algorithm were used to test the simulated and measured range profiles. The results show that the pose estimation error obtained by using PEMSPNC algorithm is the smallest, and compared with OPDVA algorithm, the average error is reduced by 0.4434 °, and has a good processing result for the measured data.

张欣, 李思宁, 孙剑峰, 姜鹏, 刘迪, 王鹏辉. 基于Mean Shift点法向量分类的目标三维姿态估计[J]. 红外与激光工程, 2020, 49(S2): 20200109. Zhang Xin, Li Sining, Sun Jianfeng, Jiang Peng, Liu Di, Wang Penghui. 3D pose estimation of target based on Mean Shift point normal vector classification[J]. Infrared and Laser Engineering, 2020, 49(S2): 20200109.

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!