The doctoral thesis deals with localization in indoor environment by using a smartphone. The latter has in a way become an indispensable tool of modern man, due to, among other things, the ever-increasing role in the field of personal navigation systems. Indoor localization is currently the subject of numerous studies and projects since none of the existing positioning systems have been established in such a way as the global navigation satellite systems for localization and navigation in the outdoor environment. The reasons for this lie in the high cost of precision positioning systems and in the poor performance of affordable systems. A new indoor positioning system may come into wider use if a compromise is reached, i.e. when sufficient accuracy and precision of localization are provided for a relatively low price. Modern smartphones are equipped with numerous sensors (inertial sensors, camera, barometer) and communications modules (WiFi, Bluetooth, NFC), which enable the implementation of various localization algorithms. To exploit the potential of modern mobile devices as much as possible, we have developed three different localization systems, namely visual localization, inertial navigation system and radio localization. Within the visual localization a visual odometry algorithm was implemented on a smartphone, where it can run in real time in two separate threads. On inertial navigation system a step counter (pedometer) and a digital compass, which are based on the use of accelerometer, gyroscope and magnetometer, were implemented for the purpose of dead reckoning. Since visual odometry as well as the inertial navigation system are sensitive to external disturbances, both localization systems were combined with the extended Kalman filter. In this way, an accurate and robust relative localization system which enables determining the position relative to a known starting point was obtained. In order to determine global positions in indoor environment, we undertook the development of radio localization, under which we tested several different methods, namely trilateration, fingerprint-based method and particle swarm optimization, which can operate with analysis of measurement of Bluetooth signals’ strengths. For the needs of the operation of aforementioned methods, we built a non-linear, as well as fuzzy models, which describe the change in signals’ strengths as a function of the distance to the transmitters. The developed relative localization system played a key role at measurement acquisition on which the models were built, since signals’ strengths as a function of the distance to the transmitters can simply be obtained by its use. The best localization results were obtained by using the fuzzy models and the fingerprint-based method. In this case 53 % of the position errors were smaller than 0.5 m and 99 % of the position errors were smaller than 1 m. In order to get even better results of the global localization, a fusion of the global positioning system based on Bluetooth fingerprints and relative localization system, which consists of visual odometry and inertial navigation system, was implemented. By combining these two systems, the errors of the global positions were reduced, since in this case more than 80 % of the errors were smaller than 0.5 m. With excellent results many doors open for the developed localization system, as it can be used in guiding autonomous mobile systems or people inside buildings.
|
