一种基于结构随机森林的家庭日常工具部件功用性快速检测算法

发布时间：2018-07-09 20:33

本文选题：机器视觉 + 功用性检测　；参考：《光学学报》2017年02期

【摘要】：家庭日常工具的部件功用性主动认知是家庭服务机器人智能提升的重要方面。为满足服务机器人实时自主作业的需要,提出了一种基于结构随机森林(SRF)的工具部件功用性快速检测算法。在离线训练阶段,利用SRF训练功用性边缘检测器与功用性检测器,并通过评估功用性检测结果的Fβ值确定工具各部件功用性对应的先粗糙后逐步精细化(coarse-to-fine)阈值。在线检测阶段,首先使用功用性边缘检测器计算功用性区域边缘的初步概率图,继而加以coarse-to-fine阈值滤波得到包含工具部件功用性的外接矩形区域,最后对该区域使用功用性检测器进行检测。实验结果表明,在普通非图形处理器系统下,相较于现有的全局搜索检测方法,本文方法对各功用性部件的检测效率均明显提升,且召回率和精度都有提高。
[Abstract]:The functional active cognition of the components of family daily tools is an important aspect of the intelligent promotion of home service robots. In order to meet the need of real-time autonomous operation of service robots, a fast detection algorithm for the performance of tool components based on structured random forest (SRF) was proposed. In the off-line training stage, the functional edge detector and the functional detector are trained by the SRF-SRF, and the coarse-to-fine threshold is determined by evaluating the F 尾 value of the result of the functional detection. In the phase of on-line detection, the primary probability diagram of the edge of the functional region is calculated by using the functional edge detector, and then the external rectangular region containing the utility of the tool component is obtained by coarse-to-fine threshold filtering. Finally, a functional detector is used to detect the area. The experimental results show that compared with the existing global search detection methods, the efficiency of the proposed method for detecting functional components in general non-GPU systems is significantly improved, and the recall rate and accuracy are improved.
【作者单位】：燕山大学信息科学与工程学院;河北省计算机虚拟技术与系统集成重点实验室;
【基金】：国家自然科学基金(61305113) 河北省自然科学基金(F2016203358)
【分类号】：TP391.41;TP242

【参考文献】