| As an emerging strategic industry,LBS(location-based service)is more and more prosperous and people have put higher demands on indoor positioning technology.There are common indoor positioning technologies:base station positioning,Wi-Fi positioning,UWB positioning,Bluetooth positioning,RFID positioning,visual positioning technology and so on.Wireless positioning and computer vision positioning have their own advantages and disadvantages,such as high precision but high computational complexity of computer vision positioning;simple to use but poor positioning accuracy for wireless positioning.Therefore,a single indoor positioning technology cannot meet people's high requirements.Indoor positioning based on the fusion of image and wireless communication signals can exert their respective advantages.This thesis has completed the following work:1.An indoor wireless signal attenuation model based on image signal assistance was proposed.Aiming at logarithmic distance path loss model doesn't take the human interference into consideration,this thesis quantitatively analyzes the influence of human interference for wireless signal and an improved model based on the number of people and the log-distance path loss model is proposed to provide accurate distance information for positioning.Experimental results show that the distance measured by this model is closer to the actual measurement,which proves the validity of the model.Meanwhile,positioning performance is improved in indoor scene with different personnel frequency,such as lobby,offices and corridors.The root mean square error(RMSE)of measurement and positioning results is reduced from 2.37 meters to 1.14 meters,which is better than that based on crowdsourcing(CS)and the positioning algorithm based on the FM,which improves the accuracy of indoor positioning.2.A method based on deep CNN(Convolution Neural Network)model of personnel location and personnel statistic was proposed.The wireless signal attenuation model proposed needs the number of people in the signal propagation path.This thesis uses CNN model to achieve it.This thesis compares the detection performance of the following technologies,including the FCLN(Fully Convoluted Localization Network)method used in this paper,HOG(Histogram Oriented Gradient)features widely used in human detection,and manual counting.The results show that FCLN model is ideal for human detection and counting error rate under different indoor scenarios is better than HOG and more accurately estimates the path loss of wireless signals under ambulatory movement.3.A feature extraction algorithm of FAST-SURF is proposed.Visual positioning can play an important role in the blind coverage of signal in wireless positioning.However,the image features extraction and matching are computationally intensive and affect the real-time performance of positioning.In this thesis,FAST-SURF algorithm is proposed based on SURF algorithm.Under the database established in the experimental environment,SURF and FAST-SURF algorithms were respectively used to extract and match the features of the same images.Experimental results show that the correctness of the feature matching can reach 85%while the SURF algorithm only 78%,and the running time decreases to 33.28ms,which is better than SURF algorithm based on 53.74ms.The performance for computation and real-time has been improved.4.For the problem of poor positioning accuracy for a single positioning method,this thesis studies the multi-source information fusion strategy and uses marginal particle filtering to fuse visual and wireless positioning information.Experiments show that the fusion algorithm further improves the accuracy of indoor positioning,effectively optimizes the smoothness of the positioning estimation,and has good convergence.Through the experimental measurement,wireless indoor positioning with the assistance of image signal is superior to other related algorithms in real-time and positioning accuracy,which has important theoretical and practical value. |
PDF Full Text Request