Research On WiFi Signal Based Indoor Positioning In Underground Parking Lot

With the rapid development of China’s economy,China’s car ownership is steadily increasing.Open-air parking spaces can no longer meet the demand for parking,so the number of underground parking lots is gradually increasing.However,there are pillars,walls and other structures in the underground parking lot.The narrow space,limited sight and complex structure cause problems such as difficulty in finding cars and wasting parking spaces.With the help of positioning technology,people can find free parking spaces through path planning.The satellite positioning system is currently one of the most widely used positioning technologies.In indoor environments,satellite signals are easily obstructed by buildings,so GPS positioning errors are relatively large.At this time,other technologies are needed to achieve indoor positioning.Since most mobile devices can receive WiFi signals and WiFi access points are widely deployed indoors,indoor positioning technology based on WiFi signals has become a hot spot in indoor positioning research.WiFi positioning technology based on location fingerprints is currently the most widely used WiFi indoor positioning technology.Positioning based on location fingerprints can be divided into two stages:offline collection and online positioning.The goal of the offline collection stage is to build a WiFi fingerprint database.Aiming at the time-consuming and laborious problem of WiFi fingerprint database construction,this article uses a semi-supervised learning model to reduce the scale of the data.The main task of the online positioning stage is to match the similarity of WiFi location fingerprints.At present,most positioning algorithms do not consider the difference in the physical characteristics of the WiFi signal in the 2.4GHz and 5GHz frequency bands.The dual-band signals are mixed indiscriminately.In this article,WiFi signals in different frequency bands are distinguished,and a dual-band WiFi positioning algorithm based on threshold selection is proposed.The main work of this paper is as follows:(1)A database with a time span of 3 months and 23,800 WiFi fingerprints was constructed in the underground parking lot of North China Electric Power University.It contains dual-band WiFi signals,data from different orientations of data collectors and data collected in different time periods.In addition,the coordinates of access points,temperature,humidity,and the number of vehicles were also recorded.(2)Aiming at different characteristics of dual-band WiFi,a dual-band WiFi positioning algorithm based on threshold selection is proposed.While retaining the advantages of the 5GHz band signal,it removes its negative impact on the positioning effect.Experimental results show that the threshold selection algorithm has improved positioning accuracy and positioning effect stability under the KNN algorithm compared with mixing dual-band signals indiscriminately.(3)The construction of WiFi fingerprint database is time-consuming and laborious,but the cost of acquiring WiFi fingerprint data without coordinates is relatively low.The semi-supervised learning method can use unlabeled samples to improve the training effect.This paper applies a semi-supervised learning model in the top conference papers of artificial intelligence to the online positioning stage of WiFi indoor positioning.The experimental results show that the model is comparable to the commonly used indoor positioning methods in terms of positioning effect.The advantage is that it can significantly reduce the construction cost of WiFi fingerprint database.