Design And Implementation Of Indoor Positioning Method Based On Wi-Fi Channel Status Information

Indoor positioning technology has been widely used in indoor navigation,behavior monitoring,safety rescue,etc.Indoor positioning technology has also made considerable progress with the development and progress of science and technology and the prevalence of intelligent terminal equipment.Compared with other positioning technologies,indoor positioning based on Wi-Fi has the advantages of lower cost and better positioning accuracy.High-precision positioning based on CSI signal has become a research hotspot in this field Because its finer granularity and more stability than Received Signal Strength(RSS).Aiming at the problems of large-scale training samples,long training time and poor positioning accuracy of non-reference point for learning methods such as neural network,a CSI fingerprint positioning method based on random forest(RF)is proposed.The CSI fingerprint are used to train the RF classification model in the offline and the trained model is used to achieve rough positioning in the online,and then WKNN algorithm is used to complete accurate positioning.In this thesis,a method of adding RSS features to compose hybrid fingerprints is proposed,which improves the spatial similarity of CSI features and improves the positioning performance of the system.The disadvantages of non-reference point on the performance of the existing positioning system is analyzed,and an improved relative majority voting combination strategy is proposed according to the actual requirements of indoor positioning system——real-time voting delay selection.Then WKNN algorithm is used to select the K with the most votes from the final voting result,and complete the final position estimation,and the positioning accuracy is improved.The experimental results show that the positioning system based on RF algorithm can save more than 70% of the time compared with ConFi positioning system during either offline phase or online phase in different environments.At the reference points interval of 1 m,the average error of RF-based method is 1.2 m lower than that of CNN-based and KL-based method,and the positioning accuracy is improved by 27.3%.When the distance between reference points is 2 m,the average error is 0.55 m and 0.71 m lower than that of CNN-based and KL-based methods,and the positioning accuracy is increased by 13.1% and 16.3%,respectively.The results prove that the proposed CSI fingerprint positioning system based on Random Forest algorithm has better positioning accuracy and real-time performance in many environments.