单目视觉技术在室内定位中的应用研究

Application of monocular vision technology in indoor positioning

针对有源室内定位技术存在的基建投资较大、信号稳定性要求高、定位效果不理想等问题,文中提出一种成本较低、定位精度不受信号稳定性影响的无源室内定位方法。该方法应用单目视觉技术分析实时照片从而确定用户的室内定位。首先使用AlexNet网络的迁移模型识别照片得到初步定位,进而结合尺度不变特征变换算法和相机位姿估计算法解算出精确定位。经测试,92%的测试点定位误差在1.5 m以下。结果表明,文中提出的定位算法能实现较高的定位精度,为无源室内定位的相关研究提供参考。

To deal with the problems of the active indoor positioning technology,including large infrastructure investment,high requirement on the stability of the signal transmitter and unsatisfactory positioning effect,this paper proposes a passive indoor positioning algorithm with low cost and no influence of signal strength on the positioning accuracy.The monocular vision technology is adopted to analyze the photos taken in real time to determine the user's position in the road network.The AlexNet is applied by transfer learning to recognize the indoor scene and identify the photos for preliminary positioning,and then scale invariant feature transform(SIFT)and robust improvement solution to perspective-n-points problem(RPnP)are utilized to calculate and determine the accurate position.The experimental result shows that 92%of targeted positioning errors are within 1.5 cm.This positioning algorithm proposed achieves a high accuracy,which can provide a reference for similar studies.