基于高斯-粒子滤波的SLAM算法提取果实特征  

SLAM algorithm based on Gauss-particle filter in fruit feature extraction

作　　者：王丹丹[1,2] 石峰 杜雪[1] 袁赣南[1] Wang Dandan;Shi Feng;Du Xue;Yuan Gannan(College of Automation,Harbin Engineering University,Harbin 151000,China;College of Electronic Information and Electrical Engineering,Anyang Institute of Technology,Anyang 455000,China)

机构地区：[1]哈尔滨工程大学自动化学院,哈尔滨151000 [2]安阳工学院电子信息与电气工程学院,河南安阳455000

出　　处：《河南师范大学学报（自然科学版）》2020年第2期59-65,共7页Journal of Henan Normal University(Natural Science Edition)

基　　金：国家自然科学基金(51709062);河南省科技攻关项目(182102110295,172102310671,172102210158);河南科技智库调研课题项目(HNKJZK-2019-30B);安阳市科技攻关项目(121);安阳工学院博士科研启动项目(BSJ2017006);安阳工学院教育教学改革研究项目(AGJ2019053).

摘　　要：针对传统农作物采摘方式落后、采摘效率低、果实特征识别精度低等问题,提出了一种基于SIFT的果实特征匹配算法.对导航机器人采集的果实图像进行去噪与特征提取,然后对不同传感器采集到的含有一定角度偏差的图像进行匹配,得到较精准的特征位置:提出了一种高斯-粒子滤波(Gauss-Particle Filter,Gauss-PF)的SLAM(Simultaneous Localization and Mapping)算法.仿真实验表明,通过增大噪声协方差及特征位置初值误差验证算法的精度,PF和Gauss-PF算法的误差均随时间逐渐降低,且在x,y方向,后者误差均小于1 cm.新的算法具有较强的稳定性与较高的定位精度.最后在同等条件下,基于单个果实特征位置(0,0)的特征进行x,y方向2次观测,并采用Gauss-PF和PF算法对观测值进行量测估计,实验表明新算法均能在(0,0)的较小邻域[-1,1]cm误差范围内对其进行估计,高于PF算法的精度[-2,2]cm.Aiming at the problems of backward traditional crop picking methods,low picking efficiency and low recognition accuracy of fruit features,a fruit feature matching algorithm based on SIFT was proposed.The fruit images collected by navigation robots were denoised and extracted,and then the images collected by different sensors with certain angle deviations were matched to obtain more accurate feature positions.A simultaneous localization and mapping algorithm based on Gauss-Particle Filter was proposed.By increasing the noise covariance and the initial value error of the feature position,the accuracy of the algorithm is verified.The simulation results showed that the errors of PF and Gauss-PF algorithms decrease gradually with time,and the errors of the latter are stable within the error range of 1 cm in both x and y directions,which shows that the new algorithm has strong stability and high positioning accuracy.At last,under the same condition,two observations and estimates based on Gauss-PF and PF algorithm in x and y directions are carried out about the feature of a single fruit feature location(0,0).The experiment showed that the new algorithm can estimate the observation value within the error range of the feature location and about its small neighborhood[-1,1]cm,which accuracy is 50% higher than the error range[-2,2]cm of PF algorithm.

关 键 词：特征识别 特征提取 高斯-粒子滤波 量测估计 

分 类 号：TP362[自动化与计算机技术—计算机系统结构]