综合型最小误差关系匹配超二次曲面物体识别  

Relational Match with Hybrid Minimum Error for Object Recognition of Superquadrics

作　　者：黄芳[1] 樊晓平[1] 徐发牛[1] Zhihua Qu 

机构地区：[1]中南大学信息科学与工程学院,长沙410083 [2]Department of Electrical and Computer Engineering, University of Central Florida, Orlando, FL 32816, USA

出　　处：《系统仿真学报》2009年第8期2130-2134,共5页Journal of System Simulation

基　　金：国家自然科学基金(60774023);湖南省自然科学基金(05JJ40130);湖南省科技计划项目(2007FJ3069).

摘　　要：针对超二次曲面部件级关系特征抽取存在误差,以及在超二次曲面属性描述不具备唯一性条件下的部件级目标识别问题,利用最小结构误差关系匹配,结合部件特征属性约束的方法来对目标物体进行高效准确的识别;同时利用部件间相对姿态,设计了一种综合部件属性、关系属性和结构误差的最小误差评估函数,通过剪枝的解释树搜索,有效地实现了多物体场景下的目标物体识别。A novel method was proposed for the error in extracting relation characteristic of superquadric parts and the part-level object recognition on condition that the attribute describing of superquadrics is not provided with uniqueness. The method proposed integrated the relational match of minimum structure-error with the parts attribute constraints to implement the object recognition efficiently and accurately. Using the relative poses between the parts, the minimum error evaluation function was designed through synthesizing the parts attribute, relational attribute and structure-error. Then the recognition under the multi-object scene was carried out efficiently by searching the interpretation tree pruned.

关 键 词：物体识别 超二次曲面部件 关系匹配 最小误差评估 解释树搜索 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]