Research Of Fusion Indoor Positoning Algorithm Based On Self-optimizing Particle Filter

With the rising of mobile Internet, location services have become an essential service in people’s life, especially in the indoor complex environment without GPS or with less GPS. So it is no doubt a burning issue to explore the location tracking with high precise. WiFi is one of the most widely used indoor positioning technologies. However, due to signal fluctuations caused by multipath effects and variable indoor environment, the accuracy of indoor positioning based on WiFi is affected. In addition, it will change the WiFi signal strength when people walk in the room and it can not do real-time WiFi fingerprint database matching. Because of that, indoor positioning based on WiFi doesn’t perform well for moving people. In order to address the above issues, this paper presents an algorithm based on motion prediction to correct WiFi fingerprints. And improved pedestrian dead reckoning is proposed to improve the accuracy of short-range indoor tracking. But the error PDR is cumulative with time, which causes that PDR is not suitable for positioning of moving people for a long time. So this paper puts forward a fusion indoor positioning algorithm based on self-optimizing particle filter.The work is briefly summarized as follows:Firstly, WiFi fingerprint algorithm based on motion prediction is put forward. Due to the signal fading and multipath effects caused by a harsh environment and the difference of different terminals, RSS rank fingerprint algorithm is proposed to shield basic changes in the environment and terminal differences. RSS rank vector is compared with that in the database to determine the close region for the calculation reduction of the Bayesian algorithm. And positioning is further carried on by Bayesian algorithm within those similar regions. However, the signal fluctuations caused by changing environment still affect the accuracy of the positioning algorithm. So a motion prediction method is submitted to infer the rationality of Bayes estimation results, and make certain amendments. It shows that WiFi fingerprint algorithm based on motion prediction performs better than Bayesian algorithm alone in the experiment.Secondly, this paper proposes step detection algorithm based on self-learning threshold and improved integration step length estimation algorithm to improve pedestrian dead reckoning. The issue of pedestrian gate which could be impacted by itself and environment is studied. Based on that, it analyzes acceleration signal pattern and self-learned to improve the real-time threshold. Meanwhile, it compares the different steps including running, walking and stationary and integrated step length algorithm to estimate step length. As the experiments show, the improved pedestrian dead reckoning performs better in a short-time tracking process.Finally, a self-optimizing particle filter is posed to fuse the improved WiFi location fingerprint algorithm and improved PDR for better tracking. First of all, it has the issue of an unknown initial absolute position of indoor positioning. Based on that, indoor and outdoor seamless handover algorithm is raised to find a proper handover point with the GPS feature of selected satellites if people come from outdoor to indoor. Otherwise, it is fixed by WiFi indoor. Afterwords, it is analyzed that the lack of particle diversity is not better fusion location. Due to that, a self-optimizing particle filter algorithm is proposed, where the feature of low-weight particles is embedded into the retained high weight particles after resampling. Finally, the fusion indoor tracking is carried on. It demonstrates that this algorithm can achieve high precision positioning of about 1.78 m without any additional equipments, which increases indoor positioning accuracy and robustness.