基于3维模型的单视图不规则物体定位  

Model-Based Irregular Object Localization from a Single View Image

作　　者：李成龙[1,2] 钟凡[1,2] 秦学英[1,2] 

机构地区：[1]山东大学计算机学院,济南250101 [2]山东省软件工程重点实验室,济南250101

出　　处：《计算机辅助设计与图形学学报》2015年第1期68-75,共8页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金广东省联合基金(U1035004);国家自然科学基金(61173070;61202149)

摘　　要：单幅图像中物体的定位需要估计物体在3维空间中的位置和姿态,在静态场景中等价于摄像机定标,具有广泛的应用价值.针对具有3维模型,但缺失纹理信息的不规则物体的定位问题,提出一种基于轮廓匹配的定位方法.首先使用基于图像分割的方法提取输入图像中物体轮廓线,然后可以将图像轮廓线与给定位置和姿态参数下渲染3维模型的轮廓线进行匹配,匹配误差可以表示为位置与姿态参数的函数.由于该函数不能解析表达与求解,需要通过离散采样计算导数及目标函数值.位置与姿态参数的最优值可以通过LM(Levenberg-Marquardt)方法进行求解.实验结果表明,该方法可以快速收敛,并具有很高的精确性和鲁棒性.Object localization aims to estimate the position and orientation of the object in 3D space. In staticscenes, it is equivalent to camera tracking, which has a wide range of applications. In this paper, we propose acontour-based approach to localize texture-less irregular objects with known 3D models. The target object is firstextracted from the input image via image segmentation, the contour is then matched with rendered 3D model withgiven position and orientation parameters. The matching error can be expressed as a function of the position andorientation parameters. Since the function cannot be analytically expressed and solved, we calculate the matchingscore and derivative by discrete sampling, the optimal score and derivative parameters can be solved efficientlyvia LM （Levenberg-Marquardt） solver. Experimental results show that this method can converge quickly, and canachieve very high accuracy and robustness.

关 键 词：姿态估计 轮廓线 匹配函数 Levenberg-Marquardt方法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]