Pedestrian localisation for indoor environments
Summary (4 min read)
INTRODUCTION
- Some of the first ubiquitous computing systems to come out of research laboratories made use of location information to provide useful clues as to the context a person or device was situated within.
- This is the author’s version of the work.
- Recent advances in micro-electro-mechanical systems (MEMS) technologies have made such units smaller and cheaper, however also more prone to error.
- Using an off-the-shelf wearable inertial system and a particle filter to tackle the traditional drift problems associated with inertial tracking.the authors.
- The authors propose that such a system could be used to enable the deployment of location-aware applications in large buildings, where the installation of a high accuracy absolute location system is either too expensive or impractical.
TRACKING USING A FOOT MOUNTED IMU
- An IMU contains three orthogonal rate-gyroscopes and accelerometers, which report angular velocity and acceleration respectively.
- The remaining acceleration can then be integrated twice to track the position of the IMU relative to a known starting point and heading [18].
- Zero velocity updates (ZVUs) can be applied during this phase, in which the known direction of acceleration due to gravity is used to correct tilt errors which have accumulated during the previous step.
- The application of such constraints replaces the cubic-in-time error growth with an error accumulation that is linear in the number of steps [8].
- This component performs inertial navigation (with ZVUs), and segments the integrated trace into step events corresponding to strides taken by the pedestrian, as shown in Figure 1.
2.5D BUILDING REPRESENTATION
- There are many obstacles that limit the possible movement of a pedestrian within a building.
- Since it is reasonable to assume that a pedestrian’s foot is constrained to lie on the floor during the stance phase of the gait cycle, a 2.5- dimensional description of the building (in which each object has a vertical position but no depth) is sufficient.
- Connected edges must coexist in the (x,y) plane, however they may be separated in the vertical direction to allow the representation of stairs.
- The use of a 2.5D format avoids the additional complexity that would be required to construct and use a fully 3- dimensional map.
PARTICLE FILTERS
- Bayesian filters probabilistically estimate the state of a dynamic system based on noisy measurements.
- To do this the belief is first propagated according to a motion model; a probability distribution which defines the possible transitions from one state to the next.
- To update the belief it is first propagated forward according to a motion model p(st|st−1, ct) to obtain the prior: Bel−(st) = ∫ p(st|st−1, ct)Bel(st−1) dst−1. (3) where ct is control information describing the transition from the previous to the current state (for pedestrian localisation step events can be used).
- A particle filter update consists of generating the set St from the previous posterior St−1.
- The authors now outline in detail the propagation, correction and resampling steps used by their localisation filter.
Particle propagation
- The propagation step generates the state st of a new particle by sampling from the motion model distribution p(st|st−1, ct), where st−1 is a previous state selected during resampling and ct = ut is the step event for the interval (t− 1,t).
- First, the authors perturb the step according to an uncertainty model, which describes uncertainty that has built up over the step due to noise perturbing the IMU measurements.
- In order for the particle to have exited this floor polygon, the step vector must intersect one of its edges in the (x, y) plane.
- In this case the authors update the current polygon polyt = GetDstPoly(C) (13).
- This process continues recursively until one of the first two cases applies.
Particle correction
- The correction step sets the weight wt of a propagated particle.
- The height change according to the map is given by δzpoly = Height−Height(−1) (15).
- This allows localisation to occur quickly when the user climbs or descends stairs.
- This will not be close to the height change reported in the step event, causing the particles to be assigned small weights relative to particles that are located on stairs.
- The combined propagation and correction algorithm is shown in Algorithm 1.
Re-sampling
- The number of particles needed to represent Bel(s) to a given level of accuracy depends on the complexity of the distribution, which can vary drastically over time.
- This is particularly true for localisation problems, where the number of particles required to track an object after convergence is typically only a small fraction of the number required to adequately describe the distribution in the early stages of localisation [5].
- Several schemes have been proposed for dynamically adapting the number of particles used to represent the distribution, such as likelihood-based adaptation [7], Kullback-Leibler divergence (KLD) sampling [5] and variants [17].
- The authors use KLD-sampling in their framework since likelihood-based adaptation is not well suited for problems where Bel(s) can be a multimodal distribution, as is often the case during indoor localisation due to symmetry in the environment [5].
- Since the propagation step in their framework uses control information (in the form of step events), the propagated belief is a reasonable estimate of the posterior.
INITIALISATION AND LOCALISATION
- To test their framework a hip-mounted ultra mobile PC (UMPC) was used to log data obtained from a foot-mounted Xsens Mtx IMU.
- The logs were then postprocessed on a desktop machine.
- Note that the trace has been manually aligned with the map to have the correct initial location and orientation, both of which are unknown to the localisation algorithm.
- Figure 4(b) shows the initial collection of particles, drawn from the prior Bel(s0).
- Figure 4(c) shows the particles after the pedestrian has taken five steps.
Environmental symmetry
- Rotational and translational symmetry cause multimodal distributions to arise during the localisation process.
- Such distributions consist of a set of distinct particle clusters, each of which represents a possible location.
- The distribution in Figure 4(d) consists of multiple clusters which have arisen due to both translational and rotational symmetry.
- Figure 5 shows the multimodal distribution on the first floor in more detail.
- Translational symmetry is a particular problem for buildings in which each floor has a similar layout.
Scalability
- The time required to incorporate a new step event into the belief is O(nlog2(n)), where n is the number of particles sampled from the prior.
- Since the number of particles required to represent the uniform prior Bel(s0) is proportional to the floor area (A) of the building, it is clear that for a large enough building it will not be possible to perform localisation in real time.
- If the initial heading were known to within 10◦ then the number of particles required to represent the prior would be reduced by a factor of 36 for a given floor area (although remaining O(A) complexity).
- Magnetometers work well in the absence of local magnetic disturbances.
- There have been many attempts to build location systems using WiFi access points [20, 14, 2].
WIFI FOR APPROXIMATE LOCALISATION
- WiFi-based location systems typically determine the position of the user via received signal strength indication (RSSI) measurements of multiple access points.
- Each query returns a list of visible WiFi access points and corresponding RSSI measurements.
- Here, however, the authors wish merely to constrain their prior, and thus have much less stringent requirements.
- The authors key requirement is that the user is actually located within the region determined by their WiFi system.
- The authors algorithm has two phases; an offline phase during which a coarse radio map is constructed, and an online phase which returns a region of space within which the user is located.
Offline phase
- Before starting their offline phase, the authors divide the 2.5D map into irregular cells, with each cell uniquely mapped to one room.
- Each cell is then subdivided until all cells have no edge length greater than 8 m.
- At least 20 queries are made within each cell, with the user standing in a variety of different orientations and positions and the doors both open and closed.
- For each access point apj the authors build the set of cells.
- The authors then define the visible region of the access point to be the union of these cells, given by Rj = ⋃ ci∈Aj ci (18) Figure 6 shows the visible region obtained by this approach for a single access point in their building.
Online phase
- At the start of a localisation the authors query the UMPC’s WiFi hardware to obtain a list of visible access points and their corresponding RSSI measurements.
- The authors assume that any access point with an RSSI measurement greater than−75 dBm would have been visible in at least one of the queries made during the offline phase within the user’s cell.
- The prior obtained using the WiFi algorithm is shown in Figure 7.
- Using KLDsampling to generate the prior results in an average of 136000 particles, less than 1/30th of the number required to represent the unconstrained prior over the whole building.
TRACKING EVALUATION
- When the particles have converged to form a single cluster, the problem solved by the filter is one of tracking rather than localisation.
- The number of particles used to represent the user’s position during tracking in their building is on average 170 and at most 500.
- The position of the Bat was queried during each stance phase detected by the inertial navigation component.
- Between each walkthrough the user walked a significant distance through other areas of the building, including different floors.
- Figure 8 shows the path of positions obtained from the particle filter after convergence to a single cluster of particles.
Did you find this useful? Give us your feedback
Citations
1,114 citations
Cites background from "Pedestrian localisation for indoor ..."
...Some of the above challenges have been addressed in [29, 30]....
[...]
881 citations
756 citations
Cites background from "Pedestrian localisation for indoor ..."
...However due to the presence of noise in accelerometer readings, error accumulates rapidly and can reach up to 100 meters after one minute of operation [33]....
[...]
749 citations
Cites methods from "Pedestrian localisation for indoor ..."
...Such SHSINS systems naturally suffer from the same drift accrual found in pure INS systems and they are used merely because SHS outputs are simpler to handle when working with the higherlevel filters necessary to incorporate other constraints such as maps [53]....
[...]
...For the three-floor 8725 m(2) building used by Woodman, over 4,000,000 particles were needed to adequately represent such a prior [53]....
[...]
...R. Harle is with the Department of Computer Science, University of Cambridge, UK (e-mail: robert.harle@cl.cam.ac.uk)....
[...]
...Woodman and Harle eased the data collection problem by using their SHSPF system to track a user augmented with a shoe sensor moving through a building....
[...]
...SHS-PF systems were demonstrated independently in 2008 by Krach and Robinson ([56]); Widyawan, Klepal and Beauregard ([16]); and Woodman and Harle ([53])....
[...]
705 citations
Cites methods from "Pedestrian localisation for indoor ..."
..., [70] and [71]) were based on the utilization of a mapbased or fingerprinting RSS localization method....
[...]
References
11,409 citations
"Pedestrian localisation for indoor ..." refers background in this paper
...The main disadvantages are that they require O(N) memory and to update the state requires O(N(2)) computation (since each of the N weights in the new state is calculated by a summation over all N states [54])....
[...]
...Grid-based filters are optimal Bayesian filters for systems in which the state space is both discrete and finite [54]....
[...]
8,667 citations
"Pedestrian localisation for indoor ..." refers background or methods in this paper
...There have been many attempts to build location systems using WiFi access points [20, 14, 2]....
[...]
...For example the RADAR system locates the user to within 5 m for 75% of the time [2]....
[...]
4,315 citations
"Pedestrian localisation for indoor ..." refers background or methods in this paper
...As a result indoor positioning systems have been developed based on a variety of other technologies including infra-red [5], ultrasound [6, 7, 8] and radio [9, 10, 11, 12]....
[...]
...TechnicalReport696,UniversityofCambridge,2007....
[...]
...PedestrianLocalisationforIndoorEnvironments OliverWoodman RobertHarle ComputerLaboratory ComputerLaboratory UniversityofCambridge UniversityofCambridge ojw28@cam.ac.uk rkh23@cam.ac.uk ABSTRACT Location information is an important source of context for ubiquitous computing systems....
[...]
...For example the Active Badge location system was used to track employees at Olivetti research in Cambridge [5]....
[...]
...2(a), each of which regularly emitted a unique infra-red code [5]....
[...]
4,123 citations
"Pedestrian localisation for indoor ..." refers background in this paper
...In contrast Cricket is a unilateral4 ultrasonic positioning system in which mobile tags calculate their own positions by timing ultrasonic pulses that are transmitted by beacons in the environment [8, 29]....
[...]
...As a result indoor positioning systems have been developed based on a variety of other technologies including infra-red [5], ultrasound [6, 7, 8] and radio [9, 10, 11, 12]....
[...]
...Such range measurements can be obtained by timing ultrasonic pulses, which can either be emitted by beacons in the environment and detected by a receiver attached to the IMU (unilateral positioning) [8, 29], or emitted from the IMU and detected by receivers in the environment (multilateral positioning) [28, 6]....
[...]
3,237 citations
"Pedestrian localisation for indoor ..." refers background in this paper
...We have seen indoor location systems based on infra-red, ultrasound, narrowband radio, WiFi signal strength, UWB, vision, and many others [11]....
[...]
Related Papers (5)
Frequently Asked Questions (19)
Q2. What have the authors stated for future works in "Pedestrian localisation for indoor environments" ?
Their result of less than 0. 73 m error 95 % of the time is very promising, and the authors hope to improve it in the future in at least two ways: • Automated construction of a high resolution radio map. This map could be used to further constrain future priors, and could potentially allow the system to adapt to changes in the WiFi infrastructure such as the removal or addition of access points.
Q3. How many particles are used to represent the unconstrained prior?
Using KLDsampling to generate the prior results in an average of 136000 particles, less than 1/30th of the number required to represent the unconstrained prior over the whole building.
Q4. How is the position of a Bat calculated?
To calculate the position of a Bat the system applies a multi-lateration algorithm to times of flight obtained for a pulse emitted by the Bat and received by multiple receivers installed at known locations in the ceiling.
Q5. What is the simplest way to estimate the posterior?
Since the propagation step in their framework uses control information (in the form of step events), the propagated belief is a reasonable estimate of the posterior.
Q6. What is the function used to track the position of the user?
The cluster of particles tracks the position of the user, updating at the end of each stance phase when a new step event is reported by the inertial navigation component.
Q7. How do the authors determine the position of the user?
WiFi-based location systems typically determine the position of the user via received signal strength indication (RSSI) measurements of multiple access points.
Q8. What is the way to estimate the co-ordinate locations of devices?
State-of-the-art WiFi location systems use RSSI measurements to attempt to estimate the co-ordinate locations of devices, with the best results coming from complex probabilistic approaches.
Q9. What is the way to reduce the cubic-in-time drift problem?
For foot-mounted IMUs the cubic-in-time drift problem can be reduced by detecting when the foot is in the stationary stance phase (i.e. in contact with the ground) during each gait cycle.
Q10. How many queries are made within each cell?
At least 20 queries are made within each cell, with the user standing in a variety of different orientations and positions and the doors both open and closed.
Q11. What are the notable particle filters used in the location of a pedestrian?
Most notable are The Location Stack [10, 12] and Placelab [13], which use particle filters to update the location of a user based on measurements received from a variety of different sensor systems.
Q12. What is the cost of a location system?
Since the system requires very little fixed infrastructure, the monetary cost is proportional to the number of users, rather than to the coverage area as is the case for traditional indoor location systems.
Q13. What is the tupleui of the ith step event?
The ith step event is reported as a tupleui = (l, δz, δθ) (1)in which l is the horizontal step length, δz is the change in height over the step, and δθ is the change in heading between the previous and current steps.
Q14. What is the weight of a particle assigned to a wall?
If a wall is intersected during the propagation step used to generate the state of the particle, then it is assigned a weightwt = 0 (14)If a wall is not intersected, the particle is assigned a weight based on the difference between the height change δz of the current step and the difference in height between the start and end floor polygons.
Q15. How many particles are needed to represent Bel(s) to a given level of accuracy?
The number of particles needed to represent Bel(s) to a given level of accuracy depends on the complexity of the distribution, which can vary drastically over time.
Q16. What is the stance phase of the gait cycle?
Since it is reasonable to assume that a pedestrian’s foot is constrained to lie on the floor during the stance phase of the gait cycle, a 2.5- dimensional description of the building (in which each object has a vertical position but no depth) is sufficient.
Q17. What are the two problems commonly faced during localisation tasks?
The example presented illustrates two problems commonly faced during localisation tasks: symmetry of the environment and scalability.
Q18. How can the authors build a high resolution radio map?
By querying the visible access points whilst tracking with the system presented in this paper, it should be possible to construct and update a high resolution radio map online.
Q19. What could be used to further constrain future priors?
This map could be used to further constrain future priors, and could potentially allow the system to adapt to changes in the WiFi infrastructure such as the removal or addition of access points.•