Design And Implementation Of Positioning System Combining Visible Light Communication And Machine Vision

With the orderly advancement of the national integrated communication network deployment in space,air,land,and sea,it is urgently needed to achieve fast and accurate communication network node positioning in scenarios such as deep sea,deep earth,and deep space.However,in deep sea and deep earth scenarios,due to the limitations of electromagnetic wave propagation,satellite positioning technologies such as Global Positioning System(GPS)that are commonly used on land cannot be used normally.Therefore,research on reliable and efficient positioning in such scenarios has received increasing attention and has broad application prospects in fields such as underwater item salvaging and cave exploration.Two typical positioning methods in electromagnetic-restricted environments are based on Visible Light Communication(VLC)and Machine Vision(MV).Visible light positioning uses Light Emitting Diodes(LEDs)as transmitters and Photo-Diodes(PDs)or image sensors as receivers,while vision-based positioning uses image sensors to capture image information for processing and positioning.When using an image sensor as a receiver,one positioning method is to establish an image database in advance,capture images and extract features during the positioning stage,and compare them with the image features in the database.Feature extraction can be completed using image descriptors,but descriptors often have low precision,and as the number of images in the database increases,the execution time of the algorithm will increase linearly.Convolutional Neural Networks(CNNs)are also gradually being used more for data extraction.However,using the output of the network’s hidden layer for feature description faces the problem of poor interpretability,making it difficult to optimize when occasional large errors occur,and the scalability of the positioning area in later stages is insufficient.Another positioning method is to convert the positioning problem into a camera pose estimation problem,which is often solved using Perspective-n-Point(PnP)and derivative algorithms.One key problem is to match the world coordinates and pixel coordinates of the transmitting LED one by one.Existing solutions generally require the introduction of PDs as receivers or the introduction of more complex modulation and encoding at the transmitter,which increases the complexity of the system.In response to the problems existing in the above two positioning methods,this thesis proposes two new positioning schemes and builds an actual platform for testing.The main research content and innovation points are as follows:(1)In response to the shortcomings of the two feature extraction methods in image feature extraction,a positioning scheme that combines CNNs and image descriptors is proposed.This scheme uses image descriptors for coarse positioning and CNNs for fine positioning.This thesis tests the performance of different networks and descriptors and completes the algorithm based on the ORB descriptor and EfficientNet network,and builds an actual hardware platform and software demonstration platform.The proposed positioning scheme improves the accuracy and execution efficiency of positioning compared to using image descriptors alone and limits the occurrence of large errors compared to using CNNs alone,while also increasing scalability.Additionally,this scheme also has strong anti-interference performance.(2)In response to the high system complexity problem faced by the LED coordinate correspondence stage in the PnP-based positioning scheme,an improved PnP positioning scheme based on LED color recognition is proposed.This scheme uses LED color information,adopts color space transformation and image moment extraction to extract the geometric center of the multi-color LED transmitter,proposes an iterative optimization method to improve the identification accuracy of LED pixel coordinates,and calculates the final positioning result through the EPnP algorithm and coordinate transformation.The effectiveness of the optimized scheme is verified through experimental platform testing,and the algorithm has certain real-time and anti-interference performance.