仿人足球机器人视觉系统精确定位的特征提取  

Feature Extraction for Accurate Localization of Humanoid Soccer Robot Vision System

作　　者：许君杰 郭丽峰[2] 刘莉[1] 张继文[1] XU Jun-jie;GUO Li-feng;LIU Li;ZHANG Ji-wen(Department of Mechanical Engineering,Tsinghua University,Beijing 100084,China;School of Mechanical Engineering,Beijing Institute of Petrochemical Technology,Beijing 102617,China)

机构地区：[1]清华大学机械工程系,北京100084 [2]北京石油化工学院机械工程学院,北京102617

出　　处：《机械设计与制造》2023年第6期258-262,共5页Machinery Design & Manufacture

基　　金：摩擦学国家重点实验室资助项目(SKLT2019C08)。

摘　　要：针对Robocup等比赛中机器人的定位中的视觉特征选取问题,线特征以其抗噪性好,蕴含丰富环境信息的优势获得了人们的青睐。该文针对仿人足球机器人球场比赛定位中的边线特征提取,将整体方案分解为颜色空间映射、可能片段查找、增量式算法拟合三个部分。通过预先学习标定生成的颜色查找表,将原始图像映射成多值图像;之后针对不同颜色对应的不同状态,使用有限状态机方法识别可能属于边线特征的有效片段;使用增量式算法提取出图像中的直线特征,最后增加参数调节来进一步优化特征提取结果,提高直线特征的精度。这一方法不仅可以提取直线特征,同时对于不规则的曲线特征检测,也能够使用近似的线段集合进行拟合。相比于传统方法如Hough变换等,该算法具有算法简单,提取速度快,特征精度高的优点。实验表明,算法的全局定位精度在10cm以内,特征提取时间不大于5ms,符合足球机器人实际比赛需求。Line features excel in noise immunity and abundant environmental information and have been chosen as the observed features for localization in Robocup and other competitions.The method to extract field lines from Robocup court was divided into 3 steps:mapping pixels into color space,searching possible segments,and line fitting with incremental algorithm.Through the color look-up table pre-generated by pre-calibration and pre-learning,the original image was mapped into a multi-valued image.Since different colors correspond to different states,the finite state machine method was used to identify the effective fragments that might belong to the field lines.Then the incremental algorithm was used to extract the straight-line features in the image.Finally,new parameters were added and adjusted to optimize the extraction results and improve the accuracy of the straight-line features.This method not only can extract straight line features,but also can use approximate line segments to fit the curve.Compared with traditional methods such as Hough transform,the method has the advantages of simple algorithm,fast speed and high accuracy.Experiments show that the global localization accuracy of the algorithm is within 10cm,and the cost time of feature extraction is not more than 5ms,which meets the actual competition requirements of humanoid soccer robots.

关 键 词：仿人机器人 计算机视觉 线特征 有限状态机 增量式算法 

分 类 号：TH16[机械工程—机械制造及自动化] TP242.6[自动化与计算机技术—检测技术与自动化装置]