Research And Realization Of Classification Algorithms On WLAN Indoor Positioning System

Inertial measurement and wireless positioning are two main solutions of indoor positioning. Dead reckoning is one of inertial measurement techniques, which reckon next position based on last position and information of speed and direction during this period. Accelerometers and orientation sensors equipped on smart mobile devices makes indoor positioning with dead reckoning possible. Wireless positioning estimates user’s positions with wireless signals. Trilateration and fingerprinting are two main wireless positioning techniques. Trilateration estimates user’s positions by calculating the distance between user and base stations which can be estimated by received signal strength or time of arrival. Fingerprinting make use of pattern recognition which takes received signal strength as recognizable unique pattern. Indoor positioning can be rebuilt as a classification problem and solved by training a classifier. AdaBoost algorithm which is one realization of fingerprinting is applied and verified to be better than other realizations, such as k-nearest neighbors or decision tree. This paper presents a hybrid algorithm based on fingerprinting and dead reckoning, on the one hand, the hybrid algorithm minimized the granularity of fingerprinting from cell level to step level and minimized the probability of error locations, and on the other hand, this algorithm eliminated the impact of the cumulative error of dead reckoning. The hybrid algorithm achieved better accuracy and reduced the variance of location error. This paper proposed an online fingerprint updating method which updates the fingerprint database using the estimated inputs and results which are evaluated as effective estimations. This method will help reduce labor costs effectively.