基于Hausdorff距离的3维模型匹配的改进方法  

Improved 3D Model Matching Based on Hausdorff Distance

作　　者：陈莹[1] 韩崇昭[1] 

机构地区：[1]西安交通大学电信学院综合自动化研究所,西安710049

出　　处：《中国图象图形学报（A辑）》2005年第3期326-331,共6页Journal of Image and Graphics

基　　金：国家重点基础研究发展规划资助项目(2001CB309403)

摘　　要：通过单目灰度图像来实现已知3维模型移动对象的精确定位,是基于3维模型的交通视觉检测与目标跟踪系统的首要环节,也是机器视觉领域的一个重要问题。为了更好地进行图像匹配,提出了一种带权值的Hausdorff距离作为3维模型投影和图像中物体轮廓相似性的测度,以避免建立图像特征与模型之间的点点对应,这样既可减少计算量,也可提高匹配精度。为了避免陷入局部最优,可将一种带记忆功能的模拟退火(SA)算法引入图像模型匹配,这样可提高匹配参数的搜索精度。实验证明,由于SA算法和改进的Hausdorff距离相结合能有效地对3维模型和图像进行匹配,从而可对具有平移、旋转的物体实现精确定位。The localization and recognition of known three dimensional (3D) objects from single monocular intensity images is a principle component in vision based traffic scene, and is also one of the fundamental problems in computer vision. For better image matching, the improved modified Hausdorff distance is proposed to measure the degree of similarity between 3D model and image’s contour, which avoids establishing one by one relationship between image feature and model, greatly reduces the computational complexity, and improves the matching precision. An improved simulated annealing (SA) algorithm with a mnemon is used for model matching. Because SA has high parallel and robust performance, it will finding of global optimum instead of getting into partial one. Experiments confirm that the combination of SA and improved Hausdorff distance can effectively realize matching between model and image, and localize the object that is changed in translation and rotation.

关 键 词：维模型 HAUSDORFF距离 图像匹配 图像模型 移动对象 灰度图像 机器视觉 实验证明 搜索精度 物体 

分 类 号：TB657[一般工业技术—制冷工程] TP391[自动化与计算机技术—计算机应用技术]