Research On Fingerprint Location Method Based On Wireless Signal In Indoor

Location Based Services(LBS), which offers customers a combination of mobile communications and navigational services to meet their need of location. It uses mobile terminals and mobile networks to target mobile user’s actual location. At present, a massive amount of LBS is based on indoor location information. Highly accurate location information not only could meet users’ need of the position but also satisfy the basic premise of convenient navigation. E-commerce can offer a specific recommendation based on accurate positions of users. Emergency rescue(such as fire, collapse, etc.) can provide timely and efficient rescue by getting the accurate position of rescued target. Besides, more and more applications are taking use of the location information to assist some routing work. For example, in a hospital, doctors can manage patients and medical goods by location tracking. For social applications, we can take use of users’ indoor position to build groups in order to enrich people’s social activities. All the application mentioned above are based on the location information, so it is critical to obtain indoor location.This paper summarizes the different characteristics of outdoor location and indoor location. Moreover, it describes the current common indoor location technologies and methods, which include wireless LAN based methods, geomagnetic filed based methods and other methods based on infrared, ultrasonic and RFID. Then it illustrates that Wi-Fi based methods are wildly used as its wildly layout and suitable effective distance. Basing on the analysis research status of Wi-Fi based indoor localization at home and abroad, and the basic principle of Wi-Fi fingerprint based location, this paper proposes two key problems that still exist. The first problem is that the location accuracy is subject to the calibration work at offline phrase. The second problem is location accuracy decrease caused by the dynamic change of environment.For the first problem, the location accuracy is subject to the calibration work in offline phrase, this paper gives a detailed analysis of the reason. As in grid-based calibration method, the calibration would be under the burden of a huge workload. Thus, the grid cannot be too small, which will limit the location accuracy. To solve this problem, we study the Extreme Learning Machine(ELM), and then propose weighted ELM(W-ELM) model to solve restrictions caused by single output. In detail, we use the calibrated data to train a W-ELM model at offline phase. Then at the online stage, we input fingerprint feature to the model, and select several nearest neighbors based on the output values. Then we use proposed weighted quadrature equation to get the final result. Experiments show that W-ELM model effectively expand the result searching space. Comparing with basic ELM, NN and SVM models, W-ELM can effectively improve the location accuracy.For the second problem, the dynamic changing of environment causes location accuracy decreased, this paper researches the Online Sequential ELM(OS-ELM). Then it proposes a Modified OSELM(M-OSELM) model to deal with the cover range and timeliness of incremental data. When location accuracy decreases, we can collect a little amount of new data to modify the old model instead of retraining. This paper also proposed the weight calculation equation based on the cover range and timeless of new incremental data. Experiments show that the weight index of the cover range and timeless of new incremental data can effectively improve the location accuracy. The M-OSELM combing two weight indexes can achieve higher location accuracy than NN and SVM.