关于卫星机器人地面装配的目标快速识别方法

发布时间：2018-05-27 10:47

本文选题：卫星装配 + 目标识别　；参考：《系统工程与电子技术》2017年05期

【摘要】：针对在卫星机器人地面装配领域,基于尺度不变特征变换的目标识别方法存在实时性能缺陷的问题,提出结合标量量化描述和倒排文件索引的典型目标快速识别方法,以满足卫星装配过程中准确高效识别工件和装配体的需求。在初始检测、定位和描述特征点的基础上,通过中值划分和遮蔽掩模方式完成浮点型描述向量的标量量化;利用倒排文件结构的查询策略快速搜索近邻特征点;依据距离比率准则和随机采样原则筛选稳定匹配点;通过仿射变换求解出矩形框中心坐标和边界宽度识别目标。实验结果表明,所提识别算法的平均正确率均值曲线图包围面积平均达到尺度不变特征变换的90.12%,能够正确匹配特征点并框选有效目标,具备相似的区分性能优势;同时匹配阶段执行时间只有尺度不变特征变换的19.54%,总体执行时间也只有49.84%,具有实时性能方面的优势。
[Abstract]:In the field of satellite robot ground assembly, the target recognition method based on scale-invariant feature transformation has the defect of real-time performance. A fast target recognition method based on scalar quantization description and inverted file index is proposed. In order to meet the needs of accurate and efficient identification of parts and assembly during satellite assembly. On the basis of initial detection, location and description of feature points, the scalar quantization of floating point description vectors is accomplished by means of median partitioning and masking, and the nearest neighbor feature points are quickly searched by the query strategy of inverted file structure. According to the distance ratio criterion and random sampling principle, the stable matching points are selected, and the center coordinates and boundary width of rectangular frame are solved by affine transformation. The experimental results show that the average correct rate of the proposed recognition algorithm is the mean curve of the mean curve, which reaches 90.12 of the scale-invariant feature transformation, and it can correctly match the feature points and select the effective target in the frame, so it has the similar distinguishing performance advantage. At the same time, the execution time of matching phase is only 19.54 of the scale invariant feature transformation, and the overall execution time is only 49.84, which has the advantage of real-time performance.
【作者单位】：河北工业大学机械工程学院;哈尔滨工业大学机器人技术与系统国家重点实验室;
【基金】：国家高技术研究发展计划(863计划)(2015AA043101)资助课题
【分类号】：TP391.41;TP242

【相似文献】