基于视觉物体识别的抗差岭估计定位算法  

Visual Objects Detection Based Robust Ridge Regression Indoor Localization Method

作　　者：徐昊玮[1] 廉保旺[1] 邹晓军[1] 岳哲[1] 吴鹏[1] XU Haowei, LIAN Baowang, ZOU Xiaojun, YUE Zhe, WU Peng(School of Electronics and Information, Northwestern Polytechnical University, Xi＇an 710072, China)

机构地区：[1]西北工业大学电子信息学院,西安710072

出　　处：《电子与信息学报》2018年第10期2453-2460,共8页Journal of Electronics & Information Technology

基　　金：国家自然科学基金(61473308;61771393)~~

摘　　要：基于视觉物体识别的室内定位算法是一种新型的室内定位解决方案,算法通过物体检测、位置匹配、定位方程解算等步骤确定用户位置。然而,受到单目相机视域较小和物体检测算法精度较低的影响,根据检测物体测距信息而构成的定位方程存在严重的病态性,极大降低了算法的定位成功率和定位精度。因此,该文提出一种抗差岭估计定位解算算法,通过引入岭参数将无偏估计变为有偏估计,实现均方误差最小约束条件下的最优位置估计,并利用迭代选权降低了质量较差的观测量对定位精度的影响。实验结果表明,与OLS (Ordinary Least Square), LM (Levenberg-Marquardt)和RR (Ridge Regression)算法相比,该文提出的抗差岭估计定位解算算法能够有效提高基于视觉物体识别的室内定位方法的成功率和精度。The indoor vision positioning algorithm based on object detection is a novel indoor positioning solution, which determines the position of the user through the process of objects detection, position matching, location equation calculation, etc. However, limited by the field-of-view of monocular camera and objects detection accuracy, the localization equation, which is constructed according to the detected objects range information, is seriously ill conditioned. Therefore, this paper proposes a novel localization method based on an improved robust ridge regression estimation, which reduces the influence of the lower accurate observations by iterative weight selection. The experimental results show that compared with Ordinary Least Square （OLS）, Levenberg-Marquardt （LM） and Ridge Regression （RR） algorithms, the proposed improved robust ridge regression estimation algorithm can effectively improve the positioning success rate and positioning accuracy of the object detection based indoor navigation method.

关 键 词：室内定位 岭估计 深度学习 

分 类 号：V249.328[航空宇航科学与技术—飞行器设计]