accuracy is greatly improved by integrating the collaborative constraint on fingerprint mapping.
(a) DR/Wi-Fi integration (b) CFPM
Figure 4.44: Positioning error CDF
Results for the user following T2 are shown to be quite accurate in all scenarios. However this is mainly because that particles are not allowed to cross walls. With not very many doors to wander through, the paths of the particles are constrained by the corridor walls. Therefore, any particles that are biased by the gyro drift will be killed off, thus enabling more accurate positioning accuracy. As the trajectory becomes more complicated in T1 and more doors are seen along the path of T3, the positioning accuracy decreases evidently. Ranging constraints help to exclude the fingerprint outliers that may be caused by signal fluctuation. The two-user collaboration improves positioning accuracy by 40% compared to DR/Wi-Fi integrated positioning, three-user collaboration improves accuracy by 50%. Further trials were also carried out by increasing the ranging measurement noise standard deviation to 3m, yet the positioning error remained at the same level.
4.7
Discussions and summary
This chapter discusses and analyse the possibility of integrating relative ranging measurements between anchors and rovers in a local positioning network to improve indoor positioning results using low-cost devices.
Different network conditions, i.e. measurement quality, number of users included and network geometry, are compared by their CRLB and DOP values. The ranging measurements in this work is obtained from UWB systems. Hence the UWB ranging performance when the mobile unit is static and moving in both outdoor and indoor environments are analysed.
A Gaussian process tool is applied to predict the measurement accuracy from received signal strength patterns by producing an RQI indicator. The prediction method achieves prediction accuracy to more than 80%.
Both theoretical and simulation analysis show that the positioning ac- curacy is related to the network size and geometry when the measurement accuracy is known. As DOP is able to reflect the network size and geo- metry, it therefore also indicates the effect of the measurement error on the positioning error. To include the effect of the measurement error, MDOP is applied to indicate network conditions which weights the DOP by the predicted measurement accuracy. Based on the MDOP, the positioning system can then set the threshold which kills off particles and predict the performance of the current network.
The initial implementations of collaborative ranging positioning are demonstrated in this chapter through simulating simple trajectories as well as collecting real IMU and Wi-Fi data in indoor environments. However, because UWB ranging is easily disrupted in this building, all ranging measurements are simulated to ensure the continuity of ranging data.
The CPF algorithm which integrates ranging with inertial measurements for PDR, demonstrates that the ranging measurement obtained between two moving rovers is able to constrain measurement errors by eliminating particles which fall outside the relative constraint. CFPM integrates the ranging constraint with fingerprint mapping and inertial measurements. The simulation that implemented CFPM demonstrates that the ranging measurement could constrain measurements by eliminating outlier fin- gerprints before particles are weighted. Wi-Fi signals are unstable and the selected fingerprint locations are not always close to the true position. Sudden signal changes in the environment, either when setting up the database or during the positioning phase, could both lead to fingerprinting outliers. Ranging constraints would eliminate those that do not obey the measured geometry. In this case, particles would not need to be weighted to those outlier fingerprints anymore. This improves positioning accuracy.
Due to the fingerprint outlier elimination from ranging constraints, the quality requirement for the Wi-Fi fingerprint database is reduced and allows for faster database training methods. The map information and RSS measurement constrains the heading bias while the ranging information corrects the RSS positions. As a result of the constraint on each measure- ment, the proposed multi-sensor multi-user positioning algorithm provides improved positioning accuracy and stability for mobile users with access to
4.7. Discussions and summary
inertial and Wi-Fi measurements.
However, real life situations are far more complicated. More users could be available in the designated area; users could be walking in random directions. In the two rover ranging simulation, there are periods were ranging did not improve result significantly. Failure could also occur when both inertial measurement and RSS information are dragging particles into the wrong room on the other side of the wall, causing new particles to be eventually resampled in the wrong room.
However, as also introduced in this chapter, the actual effect of the ranging measurement integration is heavily influenced by ranging accuracy, network geometry and network size. The following chapter will discuss this in detail and also look into a collaborative fingerprint training method.
Chapter 5
Adaptive collaborative indoor position-
ing
5.1
Introduction
Collaborative positioning has been widely applied in intelligent trans- port systems so that vehicles are aware of the situation of other vehicles and infrastructure in the surrounding area. It integrates positioning related measurements from multiple sensors and users to reduce positioning errors and enhance robustness. Both sensors and users consist of two types of systems, those whose positions are known, i.e. anchors, and those whose positions are unknown, i.e. rovers. The collaborative positioning algorithm proposed in this thesis is applied for pedestrian navigation which integrates multiple sensors and users adaptively from two aspects: the integration of multi-sensors into a single positioning system and the integration of multi-users (or multi-systems) to form a collaborative network based on measurement quality and geometry. The integration of multi-systems en- ables information to be shared among the users in the network and improve the positioning accuracy of each user by constraining the measurement error of each system through relative ranging between users.
The implementation of collaborative positioning is convenient in indoor positioning scenarios as many users and sensors can be found in such environments. However, while many units can be found, it is important to identify the rovers and anchors before integration and only pick out the units that will contribute the most to enhance positioning for effective performance improvement. For each rover, its own network of rovers and anchors are selected based on the three aspects discussed in Chapter 4, the ranging measurement accuracy between the units, the collaborative network geometry and size. Based on these aspects, two adaptive col- laborative positioning algorithms are developed and introduced in this chapter. The application of collaborative positioning also introduces to a collaborative Wi-Fi fingerprint training and positioning method that is
discussed here as well.