Research On Key Technologies Of Indoor Positioning Based On Computer Vision

With the development of smart devices and the improvement of people's living standards,people's demand for location services is growing stronger.In view of the problems that GPS(Global Positioning System)cannot be effectively located in indoor environment and indoor positioning technologies,such as WiFi(WIreless-FIdelity)and UWB(Ultra Wideband),must build the base stationthis and may have non-line-of-sight effects.The visual positioning scheme using mobile phone camera has been developed in recent years.This thesis studies the key technology of indoor positioning based on computer vision.In this thesis,the framework of the visual positioning system is established firstly.Then the key technologies involved in the system,including image retrieval and pose estimation,are studied and improved.Finally,the visual positioning scheme of this thesis is experimentally verified and analyzed.The research work in this thesis focuses on the improvement of existing visual positioning algorithms that aims at providing a new research method for subsequent visual positioning research.The specific work mainly includes:1?On the basis of the visual positioning system established in this thesis,this thesis proposes an improved global image matching algorithm based on weighted block-based HSV(Hue,Saturation,Value)color quantization descriptor that is aimed at the problem of insufficient accuracy and robustness in traditional global feature descriptor matching.The improved algorithm is experimentally analyzed on the standard dataset.Experiments show that this improved algorithm can achieve more accurate,more robust and more efficient image matching retrieval.2 ? For the traditional eight-point algorithm,the miscalculation of solving fundamental matrix misalignment may exist due to points distribution concentration.This thesis proposes an improved eight-point algorithm to solve the fundamental matrix,and compares its accuracy and stability with that of the traditional eight-point method through experiments.The experimental results verify the improvement of system performance by the improved algorithm proposed in this thesis.3?Based on the PnP(Perspective-n-Point)pose estimation problem,this thesis proposes a visual positioning scheme based on 2D-3D feature point matching.The image retrieval experiment under the scheme is carried out on the standard image dataset.The performance of the scheme and the scheme base on fundamental matrix solving are compared and analyzed.The experimental results show that this scheme optimizes the image retrieval algorithm and improves the retrieval efficiency.4?According to the two visual positioning schemes established in this thesis,the experimental environment is established in this thesis.The feasibility of the visual positioning system is verified by actual scenario experiments.Finally,the positioning results and positioning errors of visual positioning systems are analyzed.The positioning performances of the two visual positioning schemes are compared and analyzed.