Aiding Indoor Pedestrian Navigation with Building Heading
January 12, 2011 By: Khairi Abdulrahim, Chris Hide, Terry Moore, Chris HillThis article presents a simple solution to one of the main challenges of indoor pedestrian navigation. Many researchers have shown the benefits of using low cost shoe mounted IMUs, but the major drawback of these systems is the rapid and dramatic growth in heading drift error. The method proposed constrains this heading drift by simply using the orientation of buildings (derived from aerial imagery). Using this approach a series of real world trials have demonstrated that it is possible to return to a known point with a position error of just 0.1% of the total distance travelled, whilst using only a self contained low cost IMU.
Heading Derivation and Kalman Filter
Street level mapping is very useful for street level navigation, because it provides useful street level information to users, including features such as buildings outline and roads, and uses either line map (2D representation) or aerial imagery (3D-like representation). An extra piece of information commonly found from this type of map is the orientation of the map with respect to the cardinal points. This important map information, together with classical edge detection algorithm, is manipulated to derive the orientation or ‘heading’ of building, from building images. Figure 1 shows the derived ‘heading’ of a building.
Figure 1. The derived heading using building aerial imagery.
The heading drift error of the IMU, 
can now be estimated in Kalman filter by forming the observation 
, with:
where 
is the current derived building heading measurement and 
is the step heading which is based on the change in position caused by a single step; it consists of not only the true heading plus drift, but also other unmodelled errors from the inertial measurement unit. The step heading is first tested using an empirically determined threshold and if this passes the test, it will then be used to form an observation model. 
is the measurement noise with covariance matrix 
.
Trial
A trial using the low cost IMU was undertaken at Queen’s Medical Centre Hospital, of the University of Nottingham, as shown in figure 2. This building is selected because it represents a typical building with many straight features. The normal walking trial was done for about 40 minutes with an approximate distance of 2700 meters. The trial started and ended at the same location to make sure that we could quantify the return position error. The u-Blox HSGPS receiver was also used for comparison purposes to indicate the performance of a high sensitivity receiver in this building.
Figure 2 shows the trajectory results for the trial. In figure 2, the green line shows the output of new system which uses the building heading update and the red dot marker shows the HSGPS output. Although HSGPS receiver can track more than 4 satellites in some parts of the building, there is no useful position comparison that can be made between the new system and the GPS solution; because of the jumps in HSGPS solutions. The difference between the start and end position for the proposed system is calculated to be about 2.30m; only about 0.1% error in position from a total walking distance of 2700 meters, which is a significant improvement in performance for a self-contained-low-cost shoe mounted IMU system.
Figure 2. The proposed solution (green) and HSGPS solution (red)
Discussion
The use of building heading information provides many advantages for an indoor navigation system. Firstly, it is so simple and elegant, in a sense that only single heading information is necessary and this information is only needed once for the developed algorithm to work out the heading measurement. No room level map is needed and potentially, each building can have its own dedicated heading stored in a database which is accessible to users who want to navigate in such building. Once the system has worked out its heading, there is no need for a repeated request anymore. This is very important for a future low cost system which will potentially be looking for a real time solution.
Secondly it is clear that good performance can be obtained without the need for dedicated infrastructure such as RFID, UWB, WiFi, image sensors and so forth. It means that the cost of the system is now not proportional to the size of the navigation area. This is clearly of benefit because a true low cost system can potentially be developed from this approach. It must be mentioned however that the algorithm assumes the user to be walking in either four main headings most of the time in typical indoor building. This is indeed assumed to be valid because in a typical building which has rectangular orientation; most of the corridors, walls and rooms are consistent with building’s orientation and as such, restricting users to only walk in either four of these main headings. It is envisaged that extended period of walking in other than main headings will cause suboptimal results, although it has been demonstrated that the algorithm is robust enough to cope with short periods of movement that do not follow these principal directions.
It should be noted, however, that it is not an intention of the authors to assess how accurate the derived heading from aerial imagery is, but only to show a concept of extracting information from available aerial imagery or map and using this information in Kalman filter to restrict heading error in low cost MEMS IMU system.
Reference
This article is based on UPIN-LBS 2010 following paper:
K. Abdulrahim, C. Hide, T. Moore, and C. Hill, "Aiding MEMS IMU with Building Heading for Indoor Pedestrian Navigation," in Ubiquitous Positioning, Indoor Navigation and Location Based Services (UPINLBS 2010) Helsinki, Finland, 2010.
RSS
Become a Fan
Join our Network
Follow Us
Please send any technical comments or questions to our webmaster.