WO2016138800A1 - Optimisation d'estimations de position d'un dispositif permettant une localisation à l'intérieur - Google Patents

Optimisation d'estimations de position d'un dispositif permettant une localisation à l'intérieur Download PDF

Info

Publication number
WO2016138800A1
WO2016138800A1 PCT/CN2016/071014 CN2016071014W WO2016138800A1 WO 2016138800 A1 WO2016138800 A1 WO 2016138800A1 CN 2016071014 W CN2016071014 W CN 2016071014W WO 2016138800 A1 WO2016138800 A1 WO 2016138800A1
Authority
WO
WIPO (PCT)
Prior art keywords
rssi
point devices
reference point
location
distance
Prior art date
Application number
PCT/CN2016/071014
Other languages
English (en)
Inventor
Shueng Han Gary Chan
Suining HE
Original Assignee
The Hong Kong University Of Science And Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by The Hong Kong University Of Science And Technology filed Critical The Hong Kong University Of Science And Technology
Publication of WO2016138800A1 publication Critical patent/WO2016138800A1/fr

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W64/00—Locating users or terminals or network equipment for network management purposes, e.g. mobility management

Definitions

  • This disclosure generally relates to indoor localization, and techniques for optimizing position estimates of a device for indoor localization.
  • the method may include identifying, by a system comprising a processor, wireless access point devices within an area and determining a received signal strength indicator (RSSI) of a device to be located.
  • the method may also include determining, for each respective wireless access point device of the wireless access point devices, respective reference point devices each having a respective RSSI, with respect to the respective wireless access point device, that is similar to the RSSI of the device and respective contour data representing a respective contour that maps distances of the respective reference point devices from the respective wireless access point device. Accordingly, a location of the device may be determined based on a juncture determined from the respective contour data of the respective wireless access point devices.
  • the method may include populating, by a server comprising a processor, a location fingerprint data store for reference point devices within an area within which a device is to be located based on time samples of received signal strength indicators at an access point device for the reference point devices.
  • the method may include receiving distance information for the device based on an RSSI vector of the device at the access point device and determining an upper distance bound and a lower distance bound of a location of the device based on the distance information and the location information stored in the location fingerprint data store.
  • the method may also include determining a metric to evaluate a difference between the distance information and location information stored in the location fingerprint data store and generating relaxation data representing a semi-definite programming relaxation based on the metric using the upper distance bound and the lower distance bound of the location as constraints on the generating. Accordingly, the device may be located based on the relaxation data.
  • the system may include a memory to store instructions and a processor, coupled to the memory, to execute or facilitate execution of the instructions to identify wireless access point devices within an area and determine a received signal strength indicator (RSSI) of a device to be located.
  • the processor may also execute or facilitate execution of the instructions to determine, for each respective wireless access point device of the wireless access point devices, respective reference point devices each having a respective RSSI, with respect to the respective wireless access point device, that is similar to the RSSI of the device and respective contour data representing a respective contour that maps distances of the respective reference point devices from the respective wireless access point device. Accordingly, the system may determine a location of the device based on a juncture determined from the respective contour data of the respective wireless access point devices.
  • the system may include a memory to store instructions and a processor, coupled to the memory, that may execute or facilitate execution of the instructions populate a location fingerprint data store for reference point devices within an area within which a device is to be located based on time samples of received signal strength indicators at an access point device for the reference point devices and receive distance information for the device based on an RSSI vector of the device at the access point device.
  • the processor may also execute or facilitate execution of the instructions to determine an upper distance bound and a lower distance bound of a location of the device based on the distance information and the location information stored in the location fingerprint data store, determine a metric to evaluate a difference between the distance information and location information stored in the location fingerprint data store, and generate relaxation data representing a semi-definite programming relaxation based on the metric using the upper distance bound and the lower distance bound of the location as constraints on the generating. Accordingly, the system may locate the device based on the result of the semi-definite programming relaxation process.
  • FIG. 1 illustrates an example operating environment, in accordance with various embodiments.
  • FIG. 2A illustrates a diagram of trilateration using signal contours, in accordance with various embodiments.
  • FIG. 2B illustrates an example diagram of a framework for a trilateration using signal contours system, in accordance with various embodiments.
  • FIGs. 3A-3C illustrate mean errors with the trilateration using signal contours technique under various conditions, in accordance with various embodiments.
  • FIGs. 4-6 illustrate example simulation environments for deploying the trilateration using signal contours technique, in accordance with various embodiments.
  • FIGs. 7-23 illustrate various simulation results associated with deploying the trilateration using signal contours technique in various simulation environments, in accordance with various embodiments.
  • FIGs. 24 illustrates a process associated with the trilateration using signal contours technique, in accordance with various embodiments.
  • FIG. 25 illustrates an example diagram of a framework for fusing noisy fingerprints with distance bounds information, in accordance with various embodiments.
  • FIGs. 26A-26B illustrate example simulation environments for deploying the fusing noisy fingerprints with distance bounds technique, in accordance with various embodiments.
  • FIGs. 27-41 illustrate various simulation results associated with deploying the fusing noisy fingerprints with distance bounds technique in various simulation environments, in accordance with various embodiments
  • FIGs. 42 illustrate a process associated with the fusing noisy fingerprints with distance bounds technique, in accordance with various embodiments.
  • FIG. 43 illustrates a block diagram of a computing system operable to execute the disclosed systems and methods, in accordance with various embodiments.
  • GPS Global Positioning System
  • RFID radio frequency identification
  • Wi-Fi Wi-Fi positioning technology
  • Crowded places such as street, office buildings, shopping malls, hotels, and airports, usually have a large number of wireless access points (APs) , which form a wide coverage of Wi-Fi network. Therefore, it is feasible and practicable to adopt the Wi-Fi network and mobile phone to implement personnel positioning under indoor environments.
  • APs wireless access points
  • a common and widespread localization technique used for positioning with APs is based on measuring the intensity of the received signal (received signal strength indication, or RSSI) and using a fingerprint-based approach.
  • RSSI received signal strength indication
  • a fingerprint-based indoor localization approach is usually conducted in two phases.
  • a site survey is conducted to collect RSSI vectors of reference points (RPs) at known locations. The vectors of these RSSIs form the fingerprints of the site and may be stored in a database (or data store) .
  • RPs reference points
  • a target device samples an RSSI vector at its position.
  • a comparison between the RSSI vector of the target device and stored fingerprints of the RPs are conducted using some form of similarity metric. It then estimates the target device’s position by matching the target’s RSSI to a known location of a RP.Described herein are two techniques for optimizing position estimates of a target device for indoor localization.
  • a contour-based trilateration technique is described (also referred to herein as “INTRI” ) .
  • This technique combines the strengths of trilateration with fingerprinting to form a contour including reference points with the same signal level as the target device.
  • the position of the target device may then be estimated based on the juncture of all the contours based on an optimization that minimizing the distance between the estimated position and the contours.
  • a technique that fuses noisy fingerprint data with distance bounds information is also described (also referred to herein as “Wi-Dist” ) .
  • This technique relates to fusing noisy fingerprints with distance bounds information to improve indoor localization accuracy.
  • This distance information may be spatial or temporal. Accordingly, this technique provides a generic framework applicable to a wide range of sensing techniques and enables indoor localization independent of the manner in which distances are measured.
  • FIG. 1 illustrates an example schematic of an operating environment, in accordance with various embodiments.
  • the example operating environment may include one or more device (s) 102 (or device (s) to be located, or target device (s) ) , access points or APs 104a-d, reference points or RPs (shown as 110) , server (s) 106, and data store (s) 108.
  • the device 102 may be part of a variety of computing devices such as a mobile device including a mobile phone or “smartphone, ” tablet, laptop, netbook, personal digital assistant ( “PDA” ) , media device, watch, or other types of device.
  • PDA personal digital assistant
  • the APs 104a-d may be a suitable device that allows a component (e.g. device 102 or RPs 110) to connect to a network using, for example, a Wi-Fi, or other wireless protocol.
  • a component e.g. device 102 or RPs 110
  • the APs 104a-d may be a standalone device or may be part of another component.
  • the device 102 may be in signal communication with one or more APs 104a-d.
  • the device 102 may connect to one or more APs 104a-d via a Wi-Fi protocol.
  • this signal information may be used to locate the device 102.
  • APs 104a-d may be within an area 105 that receives Radio-Frequency (RF) signals from the device 102 to measure an RSSI.
  • RF Radio-Frequency
  • the RPs 110 may be a suitable device that connects to one or more APs 104a-d via a wireless protocol. RPs 110 may also be in signal communication with one or more APs 104a-d. The APs 104a-d may also receive RSSI information from one or more RPs 110. This information may form a fingerprint database (or data store) as described further herein. For example, the server 106 may compile such information (e.g. RSSI vectors and location information) of set of RPs 110 and store such information in a data store 108 during a site survey as further described herein.
  • information e.g. RSSI vectors and location information
  • the server 106 may be directly accessible by the components, or through an intermediary.
  • the server 106 may access a remote platform such as cloud computing arrangement or service.
  • the remote platform may also include a server 106 or data store 108.
  • the server 106 can also be hardware and/or software (e.g., threads, processes, computing devices) .
  • the server 106 can house threads to perform transformations by employing one or more embodiments as described herein.
  • One or more components of FIG. 1 may communicate with each other via a network, which may be a local network, wide-area network (including the Internet) , or other suitable communications network.
  • the network may be implemented on any suitable platform including wired and wireless technologies.
  • the server 106, RPs 110, and device 102 may communicate with one or more of the APs 104a-104d using one of the 802.11 standards (or Wi-Fi as referred to herein) .
  • FIG. 1 may include or be part of a computing system 1900 as further described herein.
  • the device 102, RPs 110, or the server 106 may be a computing system 1900.
  • a target device (target, device, or device to be located) first measures its distances to a number of landmarks of known locations. It then estimates its position which best matches these measured distances (e.g., given by minimizing the error between the measured distances and the distances from the position to the landmarks) .
  • Such a localization technique has been widely used outdoors, such as applications in GPS and cellular positioning, where the landmarks are satellites and base stations (e.g., cell towers) , respectively.
  • the landmarks are satellites and base stations (e.g., cell towers) , respectively.
  • it cannot be directly applied for indoor localization due to issues such as non-line-of-sight measurement and multipath fading.
  • fingerprinting overcomes these issues, its accuracy is often hampered by signal noise and the choice of similarity metric between signal vectors.
  • INTRI Indoor Trilateration
  • This technique employs the concept of trilateration in a fingerprint-based environment.
  • RSSI Signal to Interference
  • AP e.g. instead of as a single signal vector
  • INTRI does not suffer from the problem of a dispersed set of neighbors to the same extent as other techniques. Instead, it is more effective at addressing random signals (e.g. due to signal noise and measurement uncertainty) since it is not strictly based on a fingerprint similarity comparison.
  • INTRI For a signal level received from an AP by a target device, INTRI may form a contour including all of the RPs of the same signal level for that AP, taking into account the signal noise.
  • the target device may be locate based on a juncture of all the contours. Using an optimization formulation following the spirit of trilateration, the target devices’s location may be found by minimizing the distance between the position and all the contours.
  • INTRI may furthe utilzie an online algorithm based on signal correlation to efficiently calibrate heterogeneous mobile devices to achieve higher accuracy. As described herein, INTRI has been implemented in a simulation for experiments in an international airport, shopping mall and a university campus. These simulations show that INTRI outperforms recent schemes by having a lower location error (often by more than 20%) than some traditional finger-print based approaches.
  • INTRI integrates the relatively accurate trilateration technique as used in outdoor environments with indoor fingerprinting, and thus, combines the strengths of both approaches. More specifically, it does not necessarily require the positions of APs and line-of-sight (LoS) measurements, and locates the target device at the junction of its measured signal levels by reducing or removing the dispersion problem.
  • Wi-Fi RSSI fingerprinting is primarily discussed herein, INTRI is a general approach applicable to other fingerprint signals such as channel state information, visible light, RFID, or other suitable techniques. It should be noted that INTRI, may also be a complementary module to existing sensor localization systems without necessarily modifying or adding any specialized infrastructures.
  • FIG. 2A illustrates a diagram of trilateration using signal contours (INTRI) , in accordance with various embodiments.
  • an example environment may include an AP (e.g. APs 104a-d as shown in FIG. 1) .
  • AP e.g. APs 104a-d as shown in FIG. 1 .
  • there are three APs AP1, AP2, and AP3 .
  • a signal strength (e.g. RSSI) from a certain AP may be measured by the target device (e.g. device 102 as shown in FIG. 1) .
  • a contour may be formed from the spatially distributed RPs (e.g. RPs 110 as shown in FIG. 1) whose signal level for the AP is similar to (e.g.
  • the target device matches or within a tolerance) the signal level for the target device, subject to its statistical fluctuation.
  • the target device Given the target device’s RSSI vector, the corresponding contour for each of the APs (three contours shown in the figure) may be formed. Following the spirit of trilateration, the target device’s position may then be estimated by minimizing its distances to the signal contours. Accordingly, the target device may be located somewhere on the contour line. It should be noted that FIG. 2A shows a continuous contour, but the signal contour may also comprise discrete points in space.
  • Forming contours for random signals Signal measurement is inherently noisy. Constructing contours can thus consider random fluctuation in order to effectively locate the target. Fingerprints and target RSSIs received can be statistically analyzed so as to construct contours under random signals. Accordingly, as the target is likely to be located at position where contours meet, described herein are techniques to locate such a region.
  • Efficient contour-based localization algorithm Described is a contour-based localization algorithm based on linear programming formulation. Following the spirits of trilateration, the technique estimates a target location with the objective to minimize the distances to the contours obtained above.
  • Adaptive online calibration for heterogeneous devices Typically, for the same signal, different devices may have different readings. Accordingly, device readings may benefit from being mapped, or “calibrated, ” to the corresponding signal levels stored in the database so that the contours may be correctly discovered.
  • Offline calibration for all different devices is not always efficient or scalable. Accordingly, described herein is an efficient and practical online scheme to calibrate devices, which adapts the target measured RSSI according to the stored fingerprints based on signal correlation. Using this efficient approach, INTRI achieves higher scalability in heterogeneous devices and robustness in “noisy” environments.
  • FIG. 2B illustrates an example diagram of the “INTRI” system framework, in accordance with various embodiments.
  • the Wi-Fi fingerprint database as shown may be initialized by a site survey, storing, for example, (location, RSSI vector) pairs for each RP and vendor information of the devices used for data collection. After completion of the offline (or site survey phase) , the system may be ready for online estimation.
  • INTRI may check the vendor information of the target device and if it is different from the devices used for site survey, the target RSSI vector may be calibrated using the stored fingerprints. The calibrated RSSI vector may then be used to construct the signal contours. Given the signal contours, INTRI formulates a linear programming to jointly minimize the distances to the contours and estimates the target device’s position.
  • Pattern recognition techniques have been widely studied in Wi-Fi fingerprinting localization. More recently, advanced techniques on pattern matching have been investigated including a signal propagation model that derives RSSI at different locations and utilizes rigid matching between signals and distances to determine the target location. More recent works further utilize the labeled fingerprints to derive signal propagation model and achieve higher accuracy.
  • the INTRI approach described herein combines the advantages of fingerprinting and trilateration.
  • This approach employs a geometric scheme (with spirits of trilateration) based on random fingerprint signals to constrain the target region.
  • INTRI may achieve improved localization accuracy without neighbor dispersion.
  • INTRI may also be compatible to emerging fingerprint update schemes to achieve higher robustness in real deployment.
  • Some other recent works leverage the temporal or spatial RSSI patterns for localization. These works consider location-dependent patterns such as the trend of RSSI sequences along corridors, order of RSSIs from different APs, or the unique existence of some Wi-Fi APs at some area. Once the target measures such patterns, its location is then mapped to the area. These patterns achieve promising results for constrained and narrow environment with well-defined user trajectories (like corridors or offices) . In contrast, the contours in INTRI may be solely derived from fingerprints and are applicable to any indoor environment. Furthermore, INTRI does not necessarily require the positions of APs and LoS measurement.
  • one of the simulation environments (100 ⁇ 90m 2 ) was based in an airport. Considered were all the Wi-Fi transmitters and receivers that were equipped with omni-directional antennas.
  • the RSSI g l (dBm) at distance d from AP l is given by the LDPL model:
  • EZPerfect a model-based Wi-Fi localization scheme which considers deriving signal propagation model from fingerprint data. Given Wi-Fi fingerprints, EZPerfect first finds the matching relationship between signals and distances from APs, and then locates targets with a genetic algorithm solving trilateration problem.
  • KL-divergence (KL-div) : which utilizes the Kullback-Leibler (KL) divergence distance between the signal distribution at an RP and target during comparison. Then the top k RPs with the minimum KL-divergence are utilized for final estimation.
  • Weighted k-Nearest Neighbors which computes the cosine similarity between the fingerprints and the target RSSI vector, and finds the weighted average of k-Nearest Neighbors of highest cosine similarity to estimate the target location.
  • Horus Probabilistic algorithm (Horus) : which calculates the probability distribution of the RSSI values at each RP. Given a target RSSI vector, Horus computes the overall probability of the vector at each RP and finds the top several ones with the maximum likelihood as the target location.
  • FIG. 3A shows the mean localization errors versus the number of APs.
  • the performance of all the algorithms improves as the number of APs increases. As there are more APs deployed in the site which help differentiate the RPs, the error decreases. Compared with these traditional algorithms, INTRI may achieve lower errors under different numbers of deployed APs. This is because INTRI may consider differentiating contours using contour weights, which helps distinguish the RPs and prevents dispersed estimations.
  • FIG. 3B shows the mean localization errors against the signal measurement noise ( ⁇ dB in Equation (1) ) .
  • the performance of all the algorithms may degrade as the noise in Wi-Fi signals increases.
  • Signal noise may lead to dispersed nearest neighbors and distance measurement error, which degrades the accuracy of traditional pattern matching algorithms and EZPerfect.
  • the signal fluctuation of fingerprints can be different.
  • INTRI may achieve better accuracy.
  • INTRI may consider the signal variation in the contour construction. Therefore, it may achieve higher robustness and localization accuracy under large signal noise.
  • FIG. 3C shows the mean localization errors versus survey grid size, (i.e., the site survey grid width (in square shape) was varied to change the fingerprint density) .
  • accuracy may degrade as grid sizes increase, illustrating the tradeoff between survey cost and localization accuracy.
  • INTRI considers the signal contours in the formulation to constrain the target region. By minimizing distances to contours, INTRI prevents dispersion of nearest neighbors and may achieve higher accuracy.
  • SSD signal strength difference
  • SR signal ratio
  • FIG. 4 shows the corresponding survey floor plan of RPs and targets in HKIA.
  • 340 RPs and 1,100 targets were collected.
  • the locations of RPs are predefined on the indoor map.
  • a 5m grid density in fingerprinting was used.
  • signal data was sampled from four different directions (north, west, south and east) .
  • 15 samples of RSSI vectors were collected.
  • the ground truth of the target locations was also predetermined in grid form (in the testing, the surveyors found the RP or target locations from the nearest pillars, floor tiles and other noticeable indoor landmarks) . Note that the data sampling was conducted with people around, and temporal fluctuation existed within the fingerprints and target signals.
  • the time interval between samples in Wi-Fi scanning is 1 second.
  • Wi-Fi APs were officially pre-deployed. Thus, their number, locations and transmission power were already settled before conducting a site survey.
  • the mobile APs tethered by smartphones we filtered out, and the signals of virtual APs (VAPs) were combined.
  • VAPs virtual APs
  • Overall 360 APs were detected (each RP measures 47 APs on average) .
  • Part of these APs may be installed outside the survey site since their coverage in the site is relatively small and signals are globally weak.
  • the target samples are collected one month later than RP collection. For schemes like KL-divergence and Horus which are device dependent, the calibration scheme as discussed herein was utilized.
  • FIG. 5 shows the RPs and targets on campus (100 ⁇ 50m 2 ) .
  • 250 RPs and 475 targets were collected.
  • FIG. 6 shows the RPs and targets in the shopping mall (150 ⁇ 100m 2 ) .
  • 680 RPs and 680 targets were collected.
  • the blue dots represent RPs and the red diamonds are targets.
  • Similar to airport, the officially deployed APs were given and manually changing their settings (installation locations and TX power) was not permitted.
  • overall 320 APs were detected (each RP measures 24 APs on average) .
  • overall 190 APs were detected (average 28 APs at each RP) .
  • FIG. 7 shows the histograms of AP numbers detected by targets at different sites.
  • each target detected 16 APs on HKUST and 22 APs in HKCP.
  • HKIA each target detected 16 APs on average.
  • targets in HKIA and HKUST have a similar detected AP number, the survey site in HKUST was smaller and hence it had a denser AP deployment.
  • the signal coverage of APs in the campus corridor and HKCP was constrained by the wall partitions, which helps differentiate the RPs. Therefore, better localization performance was expected on HKUST and HKCP than in HKIA. Based on these detected APs, the effect of different received AP numbers was evaluated.
  • FIG. 8 shows the corresponding signal noise in each survey site. Shown is the cumulative probability of the ⁇ l according to Equation (12) . As shown, there was a smaller signal noise in HKUST and HKCP than in HKIA. This is because the airport boarding area was large open space with many airline passengers, which leads to higher signal uncertainty. Based on the difference in signal noise, the parameter z 0 in Equation (14) for online localization was adjusted.
  • FIG. 9 shows the mean localization errors of INTRI in three sites given different z 0 .
  • the error first decreases and then increases. This is because the localization error depends on two factors: signal noise and RP fingerprint differentiation.
  • signal noise and RP fingerprint differentiation.
  • z 0 is small, the tight contours cannot accommodate the large measurement uncertainty in target. Thus, the error is high.
  • the contours can bound the target, and hence the error decreases.
  • the error increases because, as contours become wide, the differentiation between RPs becomes weak. Then more distant RPs are included in localization. The result shows that without sufficient RP fingerprint differentiation, wide contours would not be beneficial.
  • z 0 is slightly larger in HKIA due to higher signal variance in the airport (as shown in FIG. 8) .
  • z 0 4 in HKIA
  • z 0 3.5 in HKCP
  • z 0 2.5 in the HKUST campus.
  • FIG. 10 shows the running time CDF of INTRI on the targets at different sites. It shows that at the three sites, INTRI shows different computation time due to the difference in RP numbers and detected AP numbers at a target side and an RP side.
  • the running time of INTRI at HKCP was shorter than that of HKIA and HKUST. It is mainly because many of the detected APs in HKCP were installed in stores and have limited coverage (small contours) due to wall partitions, compared with those in HKIA and HKUST. A target can be quickly mapped to smaller areas and therefore the overall computation becomes much shorter.
  • FIG. 11 shows the linear signal model of four different target estimations in HKIA using target data (x-axis) RP signals (y-axis) . Based on the a and b, obtain were the calibrated signal strength, for each given target. Shown is that the linear calibration scales up at the target, which matches our observation in FIG. 22.
  • FIG. 12 compares the cumulative errors of different device calibration schemes in the airport. Uncalibrated signals in INTRI were used as the baseline case. The calibrated INTRI may improve from the uncalibrated scheme, and also outperforms SSD and SR.This is because the proposed scheme jointly considers the relative trend and the RSSI adjustment model when calibrating devices. Unlike SSD and SR, the AP signal trend may mitigate the errors in signal values when INTRI calibrates target RSSI using fingerprints, which is may be more robust under large signal noise, for example at an airport.
  • FIG. 13 shows the localization error with and without differentiating the contours using HKIA data.
  • two scenarios using 5m (default) and 10m survey grid size were two scenarios using 5m (default) and 10m survey grid size.
  • contour weights may discriminate the fingerprints by penalizing RPs with weak signals. Then the influence from the noise in the measurement was mitigated.
  • FIG. 13 plotted were the mean localization errors of these algorithms against the number of APs used.
  • Randomly selected parts of the APs were detected at a target side to simulate the miss of the RSSI due to crowds of people or site construction change, which generally exists at an airport. As the number of APs used decreases, the performance of all five algorithms may degrade. Accordingly, INTRI is less susceptible to a AP number change than the other four algorithms. This is mainly because the junction of contours can constrain the target in a small region and reduce the dispersed set of nearest neighbors.
  • FIG. 14 shows the mean localization errors against the survey grid size.
  • the minimum grid size is five meters, columns or rows of RPs are removed to form the grid sizes with multiples of five.
  • EZPerfect achieves slightly higher accuracy than WKNN, KL-divergence and Horus under larger grid size. This is because for EZPerfect additional distances from multiple APs constrain the target location and prevent large error.
  • INTRI has much higher localization accuracy at different grid sizes. This is because INTRI considers signal uncertainty in contours and constrains the target estimation through joint optimization. Without relying on accurate distance measurement, INTRI may still constrain the target estimation by signal contours under large grid size.
  • FIG. 15 the cumulative errors of INTRI with the four other algorithms in the HKIA. Due to large measurement noise in the airport, WKNN’s accuracy is weakened by the dispersed nearest neighbors. Horus assumes a certain distribution of signal level at each RP and therefore cannot represent real signal distribution under limited sampling. KL-divergence also requires large data sampling and dense fingerprints in signal distribution comparison. Therefore, it cannot adapt to the noisy airport environment. The large signal noise also degrades the distance accuracy of traditional trilateration in EZPerfect. In contrast to the above methods, INTRI jointly considers the signal noise and distances to contours, and therefore reduces the misestimation of the target.
  • FIGs. 16 and 17 show the cumulative errors in HKUST and HKCP respectively.
  • the fingerprints and target signals in HKUST and HKCP carry less signal measurement noise.
  • EZPerfect becomes slightly better than WKNN as the distance measurement becomes less noisy.
  • INTRI achieves higher accuracy than other algorithms since it considers the signal variation in constructing contours and utilizes them to reduce the dispersed nearest neighbors. Accordingly, INTRI may be general enough to work under different environments with higher accuracy.
  • Table 1 Major symbols used in INTRI formulation.
  • Section 2.3.1 first presented is how to find the signal contour for each AP, given target random RSSIs.
  • a contour (of an AP) consists of discrete RPs whose signal level is the same as the target’s measured signal level, subject to statistical fluctuation. To achieve higher localization accuracy, the RPs visited by many contours of strong signals are preferred. Therefore, in Section 2.3.2 proposed is a weighting scheme which is able to differentiate the importance of the RPs. The significant RPs are kept while the less significant ones (an unlikely target location) are filtered. After the above steps, given that the target is likely to be in the “dense” region of selected RPs, finally presented in Section 2. 3. 3 is how to located such a region based on maximally connected components.
  • N and L be the total number of RPs and distinct APs detected in the whole survey site, respectively. Further let be the random variable of the RSSI collected at RP n for AP l in the offline fingerprint collection, where 1 ⁇ n ⁇ N and 1 ⁇ l ⁇ L. Multiple samples are collected at different time ⁇ indexed 1, 2, ..., for RP n and AP l, which are denoted as
  • v ( ⁇ ) be a noise process independent from Let be a parameter determining the autocorrelation of samples. Then the signal time series can be represented as a first order autoregressive model,
  • Such autocorrelation within sequential RSSIs is related to the caching in Wi-Fi sampling.
  • Such data caching can be identified using Timing Synchronization Function (TSF) of the RSSI. Accordingly, one can conclude that an RSSI is a cached one if its TSF is identical to another in an earlier scan result.
  • TSF Timing Synchronization Function
  • FIG. 19 shows the histograms of all before and after cache filtering. Accordingly, one can see that the autocorrelation between RSSI samples decreases if we filter the cache based on TSF.
  • N l is the set of RPs detecting AP l in the site and
  • a signal contour for AP l denoted as S l , consists of a set of RPs where the target is likely within.
  • S l represents the RPs whose RSSI for AP l is likely within a certain range from g l . Therefore, in finding the contour S l , one can eliminate the RPs whose RSSI is more likely far away from the target’s, i.e., if
  • the signal contour for AP l may be found as follows. Given a target RSSI g l , utilize Equations (13) and (14) , and form the signal contour S l consisting of RP n ⁇ N l such that
  • a target measures a signal vector consisting of RSSIs from six APs, A, B, C, D, E and F.
  • the contours of D, E and F black circles
  • the RPs in black circles may share the same number of contours as those close to target. If all these RPs are considered equally without sufficient filtering or differentiation, large location errors still exist.
  • FIG. 20 Take FIG. 20 again as an illustration.
  • the signals at contours of A, B and C are stronger since they are closer to the corresponding APs. If more weight is assigned to the contours with strong signals in a final location decision, one can distinguish the important RPs more accurately.
  • a weighting scheme which differentiates RPs the most and finds the RPs with higher confidence.
  • the physical intuition of the weighting scheme is based on the log-distance path loss (LDPL) model. This technique is adopted in the weighting function for signal contour differentiation, which may achieve high localization accuracy.
  • LDPL log-distance path loss
  • the contour weight at RP n from all detected APs is defined as
  • Equation (20) for each g l at contour S l , considered is the potential that an AP is physically nearby.
  • the larger D n the more contours of strong signals hit the RP.
  • Such APs are more likely to be close to these RPs around the target, and such closeness information may be utilized to constrain the target region.
  • the RP set R consisting of the indices of RPs which have the highest contour weights as the potential area for final estimation may be found, i.e.,
  • the RPs with D n higher than ⁇ max ⁇ D n ⁇ may be dynamically selected.
  • RSSIs from an AP that is located in a region surrounded by obstacles may lead to larger ⁇ l than those from other APs with freer signal propagation.
  • an ideal line-of-sight measurement is not assumed.
  • the external parameters ( and ⁇ l ) in Equations (17) and (18) can be learned through gradient decent analysis over the fingerprint signals.
  • Equation (15) the signal contours at each RP within the signal range in Equation (15) may be found and the contour weights at that RP using Equations (18) and (19) may be calculated. The most important RPs with strong signals will be selected to form R.
  • the selected RPs in R may still have “strayed RPs” due to measurement uncertainty.
  • FIG. 21 the spatial distribution of R (red rectangles) is illustrated. We can find RPs which are physically close to each other. These RPs form a region where the target is likely located, and may be used in they localization formulation. Due to signal temporal fluctuation, some RPs (the two to the right) exist and are distant away from the target location. If included in localization, these RPs may lead to location error and unnecessary computation. Considering their spatial connectivity, one may be observe in FIG. 21 that the RPs near the target (red diamond) form a connected component with the largest cardinality.
  • an algorithm may be used to find the maximally connected components, and to find the region with dense contours by filtering out RPs not in the region.
  • N R be the cardinality of R.
  • the threshold of adjacency may be set as times of the square grid width in site survey.
  • r n [x n , y n ] be the coordinate of RP n ⁇ R. Its distance from each RP m (n ⁇ m) in S l is calculated, i.e.,
  • Equation (23) By minimizing Equation (23) , one may minimize the distance difference and hence extend the idea of trilateration.
  • the contours are derived from fingerprints and target RSSIs, no explicit knowledge of AP locations or LoS measurement is required, and therefore, the advantages of both fingerprinting and trilateration is combined in our formulation.
  • ⁇ n the weight assigned to r n in locating the target.
  • ⁇ n the weight assigned to r n in locating the target.
  • Equation (22) and (24) one may extend the idea of trilateration into finding the weights which minimize the target’s weighted sum of distances to all the contours, i.e.,
  • a target is far more likely to be between two neighboring RPs (or within the square grid formed by four RPs) .
  • a constraint over the weight ⁇ n at each RP i.e.,
  • W is a dynamic parameter determined by the maximum contour weight
  • Equations (26) and (27) one can jointly consider the physical distances (denoted as ⁇ n ⁇ ) and the contour weights (denoted as ⁇ D n ⁇ ) in our formulation.
  • map constraints in our basic formulation. Denote the set of map constraints as E. For each edge e ⁇ E, the accessible area within the map constraints is considered as
  • a e , b e and c e are the line parameters obtained from the site map in our system initialization.
  • the formula of map edges can be easily found using the nearest map constraints. Using the above, the localization problem can therefore be formulated as a linear programming (LP) :
  • INTRI returns the set of weights assigned to RPs in R which minimizes the distances to contours, i.e., the RPs which are closer to all contours are assigned higher weights, and vice versa.
  • Finding maximally connected components Given N R selected RPs, finding the maximally connected component takes (Section 2.3.3) .
  • LP-based localization To prepare the objective function of LP, calculating the distances between R and contours takes (Section 4.1) . In Formulation (30) , there are decision variables in ⁇ n ⁇ . Thus, the LP in location estimation takes weak polynomial time, i.e., (Section 2.4.2) .
  • N R ⁇ N. Further computation reduction can be done via AP filtering and RP cluster mapping. The number of APs and RPs may then be significantly reduced. In this way, INTRI can be integrated on existing on-board LBS systems and can further supports mobile targets.
  • FIG. 23 shows the similarity of signal trend between signals of the two smartphones despite the shift. Leveraging such similarity, presented is an efficient algorithm to adapt to different mobile devices.
  • the target signals g were mapped to the signal space in fingerprint database.
  • the correlation between the target RSSI vector and that of each RP may be calculated.
  • the RPs with similar signal vectors can be leveraged for online signal calibration.
  • the vector comparison is based on the correlation between g and fingerprint f n , i.e.,
  • FIG. 24 shows a process 900 associated with the trilateration using signal contours technique, in accordance with various embodiments.
  • a computing system e.g. computing system 1900 as shown in FIG. 43
  • may identify wireless AP devices e.g. AP 104a-d as shown in FIG. 1 within an area.
  • AP devices may be identified using any suitable identifier such as a device ID.
  • the computing system may determine an RSSI of a device to be located (e.g. device 102 as shown in FIG. 1) .
  • the computing system may determine, for each respective wireless access point device of the wireless access point devices, respective reference point devices (e.g. RPs 110 as shown in FIG.
  • respective reference point devices e.g. RPs 110 as shown in FIG.
  • Determining the respective reference point devices may include retrieving the respective RSSI of each of the respective reference point devices from a location fingerprint data store. In addition, determining the respective reference point devices may include determining that the respective RSSI of each of the respective reference point devices is within a statistical tolerance of the RSSI of the device. Moreover, a weighting technique may also be deployed. For example, determining the respective reference point devices may include filtering the respective reference point devices based on a respective weighting associated with each of the respective reference point devices. In addition, the respective weighting associated with each of the respective reference point devices may be based on the respective RSSI of each of the respective reference point devices.
  • the computing system may determine, for each respective wireless access point device of the wireless access point devices, respective contour data representing a respective contour that maps distances of the respective reference point devices from the respective wireless access point device.
  • the computing device may determine a location of the device based on a juncture determined from the respective contour data of the respective wireless access point devices as described in the above sections.
  • the computing device may calibrate the RSSI of the device based on identifying the type of device.
  • INTRI may be employed as a more effective technique at addressing random signals (due to signal noise and measurement uncertainty) for indoor localization than conventional methods.
  • Wi-Dist farnesoid-Dist
  • Wi-Fi wireless fingerprints with uncertain mutual distances
  • Wi-Dist may achieve a lower rate of errors through the use of a convex-optimization formulation which jointly considers distance bounds and, for example, the first two moments of measured fingerprint signals.
  • error in location estimation is inevitable. This is due to random signal fluctuation in both offline and online measurements.
  • the distance information used in the technique may be spatial, where the target estimates the distances to some of the nodes or beaconing devices in its neighborhood using, for example, Bluetooth TM , Wi-Fi direct, ultrasound, etc.
  • the distance information used in the technique may also be temporal, where the target estimates its displacement over consecutive time intervals (e.g., by step counter or inertial navigation system (INS) provided in one’s mobile phone) .
  • INS inertial navigation system
  • Wi-Dist jointly considers distance bounds and noisy fingerprints to reduce indoor localization errors.
  • Wi-Dist may require only the first two moments (mean and variance) of the fingerprint RSSI signals and may optimize locations based on a Semi-Definite Programming (SDP) .
  • SDP Semi-Definite Programming
  • Wi-Dist In contrast to the rigidity constraints or Bayesian approaches in previous works, Wi-Dist does not necessarily require overly accurate distance measurements or knowledge of distance probability distributions. It may account for target measurement noise by taking, for example, only the upper and the lower bounds of distance as input. As opposed to some works where distance must be symmetric, Wi-Dist is applicable in stances where distance may be asymmetric (in which case the upper bounds can be obtained by using the larger value of the two directions, and vice versa) . Furthermore, Wi-Dist also considers fingerprint RSSI signals to be random with a mean and variance. Such consideration of random signal noise and measurement uncertainties contributes to its improved accuracy.
  • Wi-Dist may take as input, for example, only the upper and lower bounds of distances, it is not based on rigidity constraints or Bayesian approach, and hence does not necessarily require accurate distance measurement or knowledge of probability distributions. Using the bounds as constraints, Wi-Dist may estimate the location by maximizing the overall similarity with a fingerprint map including random signals. Accordingly, it is applicable to scenarios where distances may be asymmetric, in which case the upper (or lower) bounds can be obtained by using the larger (or smaller) value of the two directions.
  • Wi-Dist is a generic framework applicable to a wide range of sensing techniques, enabling indoor localization with adaptive spatial and temporal mobile sensing independent of how distance is measured. For example, it may employ peer-to-peer spatial distances in the crowded region. For an unpopulated area, it may then switch to dead reckoning (INS) for distance measurement.
  • INS dead reckoning
  • Wi-Fi fingerprints due to its ease of deployment without extra infrastructure beyond the existing Wi-Fi one
  • Wi-Dist is general enough to be extended to other wireless fingerprint signals such as RFID or channel state information (CSI) .
  • CSI channel state information
  • Wi-Dist has been implemented in a simulation for experiments in an international airport and a university campus. These simulations indicate that Wi-Dist may achieve better accuracy than other works (e.g. often by more than 40%) .
  • FIG. 25 shows an example diagram of a framework for fusing noisy fingerprints with distance bounds information.
  • the fingerprint database e.g. data store 108 as shown in FIG. 1
  • the fingerprint database may be initialized by a site survey by storing location and Wi-Fi RSSI vector pairs of RPs (e.g. RPs 110 as shown in FIG. 1) .
  • a target device e.g. device 102 as shown in FIG. 1
  • a localization server e.g. server 106
  • the server may construct an SDP convex-optimization problem which leads to a more accurate localization of the target.
  • Wi-Fi fingerprinting techniques have been widely studied in recent years.
  • the work by Horus (see e.g., M. Youssef and A. Agrawala, “The Horus location determination system, ” Wireless Networks, vol. 14, no. 3, pp. 357–374, Jun. 2008) estimates the target location using a probabilistic model which reflects the signal distribution in the site.
  • Expectation-maximization, compressive sensing and signal geometric patterns have been implemented for fingerprint-based indoor localization.
  • the techniques above only address Wi-Fi fingerprint issues.
  • Wi-Dist fuses distance information with fingerprints to achieve better accuracy.
  • Wi-Dist formulates the localization problem as a single joint convex-optimization problem. This reduces the influence of measurement noise and achieves higher accuracy.
  • Wi-Fi SLAM see e.g., S. He and S.H. Chan, “Sectjunction: Wi-Fi indoor localization based on junction of signal sectors, ” in Proc. IEEE ICC, June 2014, pp. 2605–2610) implements robot odometer to determine the distance accurately.
  • Wi-Dist considers a step counter whose measurements may be “noisy. ”
  • Wi-Fi direct and high-pitch sound have also considered the use of Wi-Fi direct and high-pitch sound to measure distances between devices. Some of these works consider using a rigid graph constructed through rotation and translation, while others consider using Bayesian approach to infer the device location. In contrast, Wi-Dist jointly considers distance bounds and fingerprint noise, and formulates an optimization problem to estimate the target location. Therefore, Wi-Dist may be a more versatile and realistic framework accommodating measurement noises.
  • Wi-Dist is a seamless generic framework applicable to different sensor systems with temporal or spatial distance measurement and it may be extended to different application scenarios.
  • Wi-Dist was developed based on Wi-Fi fingerprints and experiments were conducted to study its performance. In this section, discussed first are the experimental settings and performance metrics in Section 3.2.1.1. Then illustrative experimental results are presented for dead reckoning and peer-assisted localization in Sections 3.2.1.2 and 3.2.1.3 respectively.
  • Wi-Dist was evaluated in the Hong Kong International Airport (HKIA) boarding area and HKUST campus atrium. In the airport, overall 1,400 RPs in 8,000 m 2 area were collected. On the campus 394 RPs in 5,000 m 2 area were collected. FIG. 26A and FIG. 26B show their corresponding floor plans.
  • HKIA Hong Kong International Airport
  • FIG. 26B show their corresponding floor plans.
  • Wi-Dist was compared with the following typical localization schemes in the experiments:
  • Sequential Monte Carlo SMC localization
  • Particle filter Sequential Monte Carlo method which fuses INS data and Wi-Fi fingerprinting
  • Graph-based and fingerprint localization scheme (GB + FL) , which uses graph construction and Wi-Fi fingerprinting for peer-assisted localization.
  • the server constructs the rigid graph consisting of all targets. Then the system searches against the Wi-Fi signal map and finds a set of fingerprints to minimize the objective function through rotation and translation.
  • Temporal walking distance of the target is estimated from the INS sensor which counts the steps of a walking target. Each step detection is based on the periodic changes in the vertical direction of the accelerometer readings. Based on the number of steps, the distance travelled, or motion offset, can be estimated by multiplying the average stride length of the target (which is related to walking frequency) .
  • the device As the target walks, the device also collects the Wi-Fi RSSI vectors. Using the notation in Equation (11) the index m here represents the time stamp. Each target location now corresponds to a temporal measurement of a single target. The most recent M temporal targets and M-1 distances between them form a sliding window in time domain, and the estimation of the M-th target is returned as the current position.
  • a Wi-Fi temporal target P m at time m is defined as
  • the distance bound for initial or the first target in the sliding window is defined to be null. Based on the empirical test, ⁇ is set to 2 in Equation (2) .
  • the size of sliding window M is 7. 200 particles are used in the particle filter of SMC algorithm.
  • FIG. 27 plots the localization accuracy with respect to time for Wi-Dist and SMC.
  • the estimation error fluctuates as the user walks in the airport. Changes in wall partitions, crowded people, user walking direction and smartphone holding gesture introduce measurement noise in Wi-Fi and INS signals.
  • SMC sequentially considers the fingerprints and INS measurements. SMC does not jointly consider the Wi-Fi fingerprints and the distances from the multiple time periods. Therefore, large error in location estimation may occur.
  • Wi-Dist may constrain its estimations through the distance bounds in a joint optimization formulation. Therefore, Wi-Dist may achieve lower localization errors and smaller estimation fluctuation.
  • FIG. 28 shows the mean displacement measurement and corresponding standard deviation at each true walking distance.
  • the major error of displacement comes from misestimation in step counts and step length. Meanwhile the device initialization and walking curvature also leads to additional displacement errors. Based on such empirical analysis, one may obtain the displacement variance for each measured distance, which constitutes the distance bounds (Equation (2) ) in Wi-Dist temporal measurement.
  • FIG. 29 shows the mean localization errors against the number of Wi-Fi temporal target measurements.
  • the accuracy improves as more temporal samples are used. This is because joint consideration of more periods further constrains the location estimations.
  • the accuracy gradually converges, indicating that distance bounds already provide sufficient constraints.
  • peer-assisted (PA) localization may be used.
  • Peer ranging can be based on either RSS-distance mapping or sound ranging.
  • implemented and tested were sounds ranging under quiet and noisy campus environment.
  • the mean peer ranging errors under these two conditions are 0.8m and 2m respectively. Since distance constraint between two peers is asymmetric due to measurement uncertainty, the larger value in the distance measurements as the upper bound and the smaller one as the lower bound may be used.
  • 5 targets are involved in sound-based distance measurement. The cases in which the walls may partition peers during localization were not excluded.
  • FIG. 30 shows the mean localization errors against the proportion of APs removed at targets. Some received APs of each target were randomly removed to evaluate the influence of AP reduction due to wall partitioning or crowds of people. As shown, Wi-Dist and GB+FL marginally rely on the number of received APs. This is because the multiple users’ Wi-Fi samples reduce the effect of sparse AP deployment. To the contrary, FL relies on the APs to differentiate the RPs and therefore its estimation error increases as more APs are pruned.
  • FIG. 31 shows that the location estimation errors against the number of Wi-Fi samples at each target in PA localization. All the algorithms improve with more Wi-Fi samples. This is because as the number of Wi-Fi samples increases, noise from the random sampling can be reduced. Compared with GB+FL and FL, Wi-Dist achieves higher localization accuracy because it further jointly considers the measurement noise in the optimization formulation and reduces the uncertainty. However, increasing the number of Wi-Fi samples means that one may need to wait for more samples before final estimation. Accordingly, a balance may be made between accuracy and latency depending on application scenarios.
  • FIG. 32 shows the location errors against the survey grid size.
  • the minimum grid size is five meters, lines or rows of RPs are removed to form grid size with multiples of five.
  • the accuracy of the three algorithms decreases. Though less labor-intensive, larger survey grid size may more easily lead to dispersed nearest neighbors under large signal noise. Therefore, traditional algorithms like FL may not accurately differentiate these RPs.
  • Wi-Dist and GB+FL achieve more accurate location estimations with the constraints of peer-to-peer distances.
  • the rigid graphs of targets in GB+FL still suffers from pairwise distance measurement noise. By fusing signal uncertainty and distance bounds, Wi-Dist achieves higher estimation accuracy under different grid sizes.
  • FIG. 33 and FIG. 34 show the overall performance of Wi-Dist at different scenarios (INS and PA at baseline parameters) in HKIA.
  • Large indoor open space often leads to high uncertainty in Wi-Fi signals and disperse nearest neighbors in signal space.
  • the temporal and spatial distance measurement also contains large noise under the crowded scenarios.
  • Wi-Dist significantly reduces the estimation errors in HKIA.
  • Wi-Dist mitigates the effect of disperse nearest neighbors.
  • Wi-Dist was simulated for the scenarios as mentioned in Section 3.2. In this section, discussed first is the simulation setup (Section 3.2.2.1) , followed by the results for dead reckoning and peer-assisted localization (Sections 3.2.2.2 and 3.2.2.3) .
  • Wi-Fi signal strength was simulated.
  • the RSSI ⁇ (dBm) from Wi-Fi AP at a distance D can be simulated as
  • the transmission power ⁇ TX 25 dBm
  • reference distance D 0 1 m
  • Wi-Fi signal noise ⁇ db 6 dB
  • 10 APs are uniformly distributed in the survey area
  • a target takes a Wi-Fi sample every 3 seconds.
  • FIG. 37 shows the localization accuracy against the step count errors.
  • SMC locates the user based on the particle filter, which sequentially considers the Wi-Fi and INS measurements. Therefore, when step count accuracy degrades, the displacement error increases and the particles become spatially sparse, making it difficult for SMC to converge to correct locations.
  • Wi-Dist localizes the target more accurately because the joint consideration of Wi-Fi fingerprints and distance bounds of multiple periods reduces the influence of measurement uncertainty.
  • FIG. 38 shows the localization accuracy against the walking displacement errors. We can see that as the displacement error increases, the overall location accuracy decreases. Localization error in SMC increases because the particles converge slowly given large distance errors and noisy Wi-Fi measurement. Different from SMC, Wi-Dist achieves more accurate results because it utilizes the distance bounds instead of actual distance measurement. By constraining the target estimation within the intersection of these bounds, Wi-Dist may be more robust to distance uncertainty.
  • FIG. 39 plots the localization errors versus the signal noise in Wi-Fi measurement (Equation (4) ) .
  • the performance of both SMC and Wi-Dist degrades when the random signal noise increases. This is because larger signal noise makes it more difficult to differentiate the fingerprints.
  • Wi-Dist considers signal uncertainty through the expected signal difference. By minimizing the signal difference within constraints of distance bounds, Wi-Dist reduces the effect of disperse nearest neighbors and obtains better estimation results.
  • FIG. 40 shows the localization errors versus the number of users. As shown, more peer assistance provides more distance constraints over the involved users and improves the localization accuracy. Different from GB+FL, Wi-Dist shows less dependency on user connectivity. This is because Wi-Dist considers the measurement uncertainty in the optimization and jointly constrains all the users. Therefore, it does not have to involve many users to achieve high localization accuracy.
  • FIG. 41 shows the localization accuracy against the peer distance errors.
  • Gaussian noise is added to the inter-device distance measurement as an assumption.
  • the peer distance error is small, both algorithms achieve high accuracy given only Wi-Fi measurement noise in fingerprints.
  • both algorithms degrade in localization accuracy.
  • GB+FL constructs a rigid graph to constrain relative positions of different users. However, the graph shape deforms under large distance measurement errors.
  • Wi-Dist in contrast, shows more robustness by using joint optimization based on fingerprints and distance bounds. Without assuming a rigid graph, Wi-Dist can achieve more robust localization estimation.
  • Table 1 Major symbols in the problem formulation.
  • a site survey is conducted with a total of Q reference points (RPs) .
  • RPs Q reference points
  • time samples of Wi-Fi RSSI readings are collected. Due to the random nature of radio signals, multiple samples are collected in order to reduce the uncertainty in the signal measurements.
  • X be an M ⁇ 2 matrix of all these points, i.e.,
  • the target m For each of the targets (spatial or temporal) , let be the RSSI value at target location for Wi-Fi AP l, Similar to the RP RSSI vector, we define the target m’s sampled RSSI vector as
  • ⁇ mn the distance (spatial or temporal) between targets and for any m, n ⁇ V, i.e.,
  • W m stores the distance bounds, i.e.,
  • the RP positions R are used to estimate the location of the target. Let ⁇ mq be the weight assigned to RP q to locate m, so that
  • W be an M ⁇ Q matrix of ⁇ mq , r q ⁇ R, i.e.,
  • a metric to evaluate the difference between the target Wi-Fi samples and the stored fingerprints under measurement noise may be introduced.
  • J mq the shared APs between Wi-Fi measurement point m and RP q (0 ⁇
  • the expected signal difference between RP q and the target m’s RSSI in AP l is derived as:
  • Equation (18) one may present the following the objective function for Wi-Dist. To jointly measure the overall difference of all targets with the stored signal map, one may find the weights which minimize all the targets’ weighted sum of expected signal difference as:
  • a matrix W may be created so as to satisfy
  • Formulation (20) can be generalized into the following problem:
  • the proof may be briefly described as follows. Suppose a polynomial-time algorithm that takes as input the distances between different targets as well as fingerprint measurement points to recover their original positions. Therefore, one may minimize the overall sum of target signal differences with the fingerprint map to obtain the candidate locations, and then find among the candidate locations, the locations that satisfy the corresponding distance bounds.
  • an algorithm can be used to solve the SSP by applying it to an instance of the problem. After reaching its polynomial time bound, the algorithm will either have returned a solution or not. In the first case, one may check if the solution with pairwise distances returned is consistent with the distance bounds. Similarly, in the SSP one may check the sum of elements in polynomial time and accept if and only if the check succeeds. In the second case, the instance may be rejected. For both cases, the correct answer for SSP may be returned. Since SSP is already NP-hard, the problem is as hard as the SSP. Thus, the problem in Definition 1 is NP-hard.
  • SDP Semi-Definite Programming
  • Constraint (23) is nonconvex due to Given a symmetric matrix A, let represent that A is a positive semidefinite matrix.
  • A is a positive semidefinite matrix.
  • Constraint (25) is a nonlinear constraint, which can be further transformed into a linear matrix inequality. Then it can be solved efficiently by a convex optimization solver. The transformation is through a Schur complement:
  • H be a matrix partitioned in four blocks, consisting of four matrices B, E, C and D, i.e.,
  • Formulation (24) may then be transformed into an SDP problem:
  • the formulation above may be directly applied in peer-assisted localization.
  • m as the time stamp may be taken.
  • the computational complexity of the solution may be analyzed. Given Q RPs and L APs, the complexity of signal difference calculation is Given M temporal or spatial target measurements (usually M is small) , the computation of SDP relaxation is bounded by Using some commercial SDP solver this problem can be solved efficiently. Further computational reduction can be achieved by AP filtering and RP cluster mapping.
  • FIG. 42 shows a process 1700 associated with the fusing noisy fingerprints with distance bounds technique, in accordance with various embodiments.
  • a computing system e.g. server 106 as shown in FIG. 1
  • populating the location fingerprint data store may include storing only a mean and a variance of the time samples of the received signal strength indicators for a respective reference point device.
  • the computing device may receive distance information for the device based on an RSSI vector of the device at the access point device.
  • the distance information may include distance information received at various time intervals.
  • the distance information may include distance information for the device is based on determining a current estimated position of the device to a previously estimated position of the device.
  • the computing device may determine an upper distance bound and a lower distance bound of a location of the device based on the distance information and the location information stored in the location fingerprint data store.
  • the computing device may determine a metric to evaluate a difference between the target signals and those stored in the location fingerprint data store. For example, determining the metric may include determining a fingerprint noise value for each of the reference point devices. In addition, determining the metric may also include determining a respective weight associated with each of the reference point devices.
  • the computing device may generate relaxation data representing a semi-definite programming relaxation based on the metric using the upper distance bound and the lower distance bound of the location as constraints on the generating. Accordingly, in 1760, the device may be located based on the relaxation data.
  • processor can refer to substantially any computing processing unit or device comprising, but not limited to, single or multi core processors, parallel platforms, an integrated circuit, an application specific integrated circuit (ASIC) , a digital signal processor (DSP) , a field programmable gate array (FPGA) , a programmable logic controller (PLC) , a complex programmable logic device (CPLD) , a discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions and/or processes described herein.
  • ASIC application specific integrated circuit
  • DSP digital signal processor
  • FPGA field programmable gate array
  • PLC programmable logic controller
  • CPLD complex programmable logic device
  • FIG. 43 and the following discussion, are intended to provide a brief, general description of a suitable environment in which the various aspects of the disclosed subject matter can be implemented, e.g., various processes described herein. While the subject matter has been described above in the general context of computer-executable instructions of a computer program that runs on a computer and/or computers, those skilled in the art will recognize that the subject innovation also can be implemented in combination with other program modules. Generally, program modules include routines, programs, components, data structures, etc. that perform particular tasks and/or implement particular abstract data types.
  • inventive systems can be practiced with other computer system configurations, including single-processor or multiprocessor computer systems, mini-computing devices, mainframe computers, as well as personal computers, hand-held computing devices (e.g., PDA, phone, watch) , microprocessor-based or programmable consumer or industrial electronics, and the like.
  • the illustrated aspects can also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network; however, some if not all aspects of the subject disclosure can be practiced on stand-alone computers.
  • program modules can be located in both local and remote memory storage devices.
  • Computer 1912 includes a processing unit 1914, a system memory 1916, and a system bus 1918.
  • System bus 1918 couples system components including, but not limited to, system memory 1916 to processing unit 1914.
  • Processing unit 1914 can be any of various available processors. Dual microprocessors and other multiprocessor architectures also can be employed as processing unit 1914.
  • System bus 1918 can be any of several types of bus structure (s) including a memory bus or a memory controller, a peripheral bus or an external bus, and/or a local bus using any variety of available bus architectures including, but not limited to, Industrial Standard Architecture (ISA) , Micro-Channel Architecture (MSA) , Extended ISA (EISA) , Intelligent Drive Electronics (IDE) , VESA Local Bus (VLB) , Peripheral Component Interconnect (PCI) , Card Bus, Universal Serial Bus (USB) , Advanced Graphics Port (AGP) , Personal Computer Memory Card International Association bus (PCMCIA) , Firewire (IEEE 1194) , and Small Computer Systems Interface (SCSI) .
  • ISA Industrial Standard Architecture
  • MSA Micro-Channel Architecture
  • EISA Extended ISA
  • IDE Intelligent Drive Electronics
  • VLB VESA Local Bus
  • PCI Peripheral Component Interconnect
  • Card Bus Universal Serial Bus
  • USB Universal Serial Bus
  • AGP Advanced Graphics Port
  • System memory 1916 includes volatile memory 1920 and nonvolatile memory 1922.
  • a basic input/output system (BIOS) containing routines to transfer information between elements within computer 1912, such as during start-up, can be stored in nonvolatile memory 1922.
  • nonvolatile memory 1922 can include ROM, PROM, EPROM, EEPROM, or flash memory.
  • Volatile memory 1920 includes RAM, which acts as external cache memory.
  • RAM is available in many forms such as SRAM, dynamic RAM (DRAM) , synchronous DRAM (SDRAM) , double data rate SDRAM (DDR SDRAM) , enhanced SDRAM (ESDRAM) , Synchlink DRAM (SLDRAM) , Rambus direct RAM (RDRAM) , direct Rambus dynamic RAM (DRDRAM) , and Rambus dynamic RAM (RDRAM) .
  • DRAM dynamic RAM
  • SDRAM synchronous DRAM
  • DDR SDRAM double data rate SDRAM
  • ESDRAM enhanced SDRAM
  • SLDRAM Synchlink DRAM
  • RDRAM Rambus direct RAM
  • DRAM direct Rambus dynamic RAM
  • RDRAM Rambus dynamic RAM
  • Computer 1912 can also include removable/non-removable, volatile/non-volatile computer storage media, networked attached storage (NAS) , e.g., SAN storage, etc.
  • NAS networked attached storage
  • FIG. 43 describes software that acts as an intermediary between users and computer resources described in suitable operating environment 1900.
  • Such software includes an operating system 1928.
  • Operating system 1928 which can be stored on disk storage 1924, acts to control and allocate resources of computer 1912.
  • System applications 1930 take advantage of the management of resources by operating system 1928 through program modules 1932 and program data 1934 stored either in system memory 1916 or on disk storage 1924. It is to be appreciated that the disclosed subject matter can be implemented with various operating systems or combinations of operating systems.
  • a user, client, etc. can enter commands or information into computer 1912 through input device (s) 1936.
  • Input devices 1936 include, but are not limited to, a pointing device such as a mouse, trackball, stylus, touch pad, keyboard, microphone, joystick, game pad, satellite dish, scanner, TV tuner card, digital camera, digital video camera, web camera, and the like.
  • These and other input devices connect to processing unit 1914 through system bus 1918 via interface port (s) 1938.
  • Interface port (s) 1938 include, for example, a serial port, a parallel port, a game port, and a universal serial bus (USB) .
  • Output device (s) 1940 use some of the same type of ports as input device (s) 1936.
  • a USB port can be used to provide input to computer 1912 and to output information from computer 1912 to an output device 1940.
  • Output adapter 1942 is provided to illustrate that there are some output devices 1940 like monitors, speakers, and printers, among other output devices 1940, which use special adapters.
  • Output adapters 1942 include, by way of illustration and not limitation, video and sound cards that provide means of connection between output device 1940 and system bus 1918. It should be noted that other devices and/or systems of devices provide both input and output capabilities such as remote computer (s) 1944.
  • Remote computer (s) 1944 is logically connected to computer 1912 through a network interface 1948 and then physically connected via communication connection 1950.
  • Network interface 1948 encompasses wire and/or wireless communication networks such as local-area networks (LAN) and wide-area networks (WAN) .
  • Communication connection (s) 1950 refer (s) to hardware/software employed to connect network interface 1948 to bus 1918. While communication connection 1950 is shown for illustrative clarity inside computer 1912, it can also be external to computer 1912.
  • the hardware/software for connection to network interface 1948 can include, for example, internal and external technologies such as modems, including regular telephone grade modems, cable modems and DSL modems, ISDN adapters, and Ethernet cards.
  • ком ⁇ онент can be a processor, a process running on a processor, an object, an executable, a program, a storage device, and/or a computer.
  • an application running on a server and the server can be a component.
  • One or more components can reside within a process, and a component can be localized on one computer and/or distributed between two or more computers.
  • these components can execute from various computer readable media having various data structures stored thereon.
  • the components can communicate via local and/or remote processes such as in accordance with a signal having one or more data packets (e.g., data from one component interacting with another component in a local system, distributed system, and/or across a network, e.g., the Internet, a local area network, a wide area network, a cloud based infrastructure, a cloud based computing system, etc. with other systems via the signal) .
  • a signal having one or more data packets (e.g., data from one component interacting with another component in a local system, distributed system, and/or across a network, e.g., the Internet, a local area network, a wide area network, a cloud based infrastructure, a cloud based computing system, etc. with other systems via the signal) .
  • a signal having one or more data packets (e.g., data from one component interacting with another component in a local system, distributed system, and

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Position Fixing By Use Of Radio Waves (AREA)

Abstract

L'invention concerne des techniques pour optimiser des estimations de position permettant une localisation à l'intérieur. Une technique consiste à combiner l'efficacité d'une trilatération et d'une prise d'empreinte digitale. Cette technique crée un contour comprenant des points de référence ayant le même niveau de signal d'un dispositif cible. La position du dispositif peut ensuite être estimée sur la base de la jonction de tous les contours et une formulation d'optimisation suivant l'esprit de la trilatération peut être employée pour réduire au minimum la distance entre la position estimée et les contours. En outre, une seconde technique décrit la fusion d'empreintes digitales bruyantes avec des informations de distance mutuelle incertaines. Cette technique peut être utilisée sous la forme d'un cadriciel générique applicable à une large plage d'empreintes digitales sans fil et peut réduire les erreurs avec une formulation d'optimisation convexe.
PCT/CN2016/071014 2015-03-03 2016-01-15 Optimisation d'estimations de position d'un dispositif permettant une localisation à l'intérieur WO2016138800A1 (fr)

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
US201562176979P 2015-03-03 2015-03-03
US62/176,979 2015-03-03
US201562283857P 2015-09-14 2015-09-14
US62/283,857 2015-09-14

Publications (1)

Publication Number Publication Date
WO2016138800A1 true WO2016138800A1 (fr) 2016-09-09

Family

ID=56849175

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/071014 WO2016138800A1 (fr) 2015-03-03 2016-01-15 Optimisation d'estimations de position d'un dispositif permettant une localisation à l'intérieur

Country Status (1)

Country Link
WO (1) WO2016138800A1 (fr)

Cited By (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107017956A (zh) * 2017-04-19 2017-08-04 深圳市尧元科技有限公司 一种节点传输误差分析方法及系统
CN107677989A (zh) * 2017-10-26 2018-02-09 武汉大学 一种基于rssi最大值进行rssi去除噪声的室内位置定位方法
CN109143161A (zh) * 2018-09-30 2019-01-04 电子科技大学 基于混合指纹质量评价模型的高精度室内定位方法
WO2019062734A1 (fr) * 2017-09-28 2019-04-04 知谷（上海）网络科技有限公司 Procédé et dispositif de positionnement en intérieur basé sur des zones d'accès wi-fi
CN109633530A (zh) * 2018-11-30 2019-04-16 哈尔滨工业大学（深圳） 一种定位方法及系统、设备、存储介质
CN111050275A (zh) * 2019-11-26 2020-04-21 武汉虹信技术服务有限责任公司 一种基于rssi特征值的蓝牙定位方法
JP2020515862A (ja) * 2017-03-28 2020-05-28 オートマトン， インク．Ａｕｔｏｍａｔｏｎ， Ｉｎｃ． Ｒｆｉｄタグを位置特定するための方法および装置
CN111965600A (zh) * 2020-08-14 2020-11-20 长安大学 一种强遮挡环境下基于声音指纹的室内定位方法
CN112070201A (zh) * 2020-08-26 2020-12-11 成都睿芯行科技有限公司 一种基于Gmapping提升建图速度的方法
TWI722700B (zh) * 2019-12-06 2021-03-21 財團法人工業技術研究院 測距裝置、測距方法和校正訊號功率之方法
CN113486223A (zh) * 2021-06-07 2021-10-08 海南太美航空股份有限公司 一种航线显示方法、系统和电子设备
WO2022054022A1 (fr) * 2020-09-14 2022-03-17 Mahan Tabatabaie Localisation intérieure basée sur des activités précédentes
CN114353787A (zh) * 2021-12-06 2022-04-15 理大产学研基地(深圳)有限公司 一种多源融合定位方法
US20220132270A1 (en) * 2020-10-27 2022-04-28 International Business Machines Corporation Evaluation of device placement
CN114745658A (zh) * 2022-03-01 2022-07-12 华中科技大学 基于决策融合的距离估计方法及介质
CN115087095A (zh) * 2022-06-21 2022-09-20 五邑大学 一种基于csi的加权knn的可见光室内定位方法
WO2022250675A1 (fr) * 2021-05-27 2022-12-01 Hewlett-Packard Development Company, L.P. Déductions d'emplacements de dispositifs électroniques
WO2024088225A1 (fr) * 2022-10-25 2024-05-02 华为技术有限公司 Procédé et système de télémétrie bluetooth, et dispositif électronique

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1965593A (zh) * 2004-05-18 2007-05-16 艾雷斯贝斯有限公司 以定义搜索区域来优化定位计算为特征的无线节点定位机制
WO2013036586A1 (fr) * 2011-09-09 2013-03-14 Alcatel-Lucent Régression de noyau à divergence kl pour empreinte non gaussienne basée sur localisation
CN103428629A (zh) * 2012-05-18 2013-12-04 中国电信股份有限公司 混合定位实现方法及系统

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1965593A (zh) * 2004-05-18 2007-05-16 艾雷斯贝斯有限公司 以定义搜索区域来优化定位计算为特征的无线节点定位机制
WO2013036586A1 (fr) * 2011-09-09 2013-03-14 Alcatel-Lucent Régression de noyau à divergence kl pour empreinte non gaussienne basée sur localisation
CN103428629A (zh) * 2012-05-18 2013-12-04 中国电信股份有限公司 混合定位实现方法及系统

Cited By (29)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11215691B2 (en) 2017-03-28 2022-01-04 Automaton, Inc. Methods and apparatus for locating RFID tags
EP3602101A4 (fr) * 2017-03-28 2021-06-23 Automaton, Inc. Procédés et appareil de localisation d'étiquettes rfid
US11408965B2 (en) 2017-03-28 2022-08-09 Automaton, Inc. Methods and apparatus for locating RFID tags
JP2020515862A (ja) * 2017-03-28 2020-05-28 オートマトン， インク．Ａｕｔｏｍａｔｏｎ， Ｉｎｃ． Ｒｆｉｄタグを位置特定するための方法および装置
US12013474B2 (en) 2017-03-28 2024-06-18 Automaton, Inc. Methods and apparatus for locating RFID tags
CN107017956A (zh) * 2017-04-19 2017-08-04 深圳市尧元科技有限公司 一种节点传输误差分析方法及系统
WO2019062734A1 (fr) * 2017-09-28 2019-04-04 知谷（上海）网络科技有限公司 Procédé et dispositif de positionnement en intérieur basé sur des zones d'accès wi-fi
CN107677989A (zh) * 2017-10-26 2018-02-09 武汉大学 一种基于rssi最大值进行rssi去除噪声的室内位置定位方法
CN107677989B (zh) * 2017-10-26 2019-09-10 武汉大学 一种基于rssi最大值进行rssi去除噪声的室内位置定位方法
CN109143161B (zh) * 2018-09-30 2023-01-10 电子科技大学 基于混合指纹质量评价模型的高精度室内定位方法
CN109143161A (zh) * 2018-09-30 2019-01-04 电子科技大学 基于混合指纹质量评价模型的高精度室内定位方法
CN109633530A (zh) * 2018-11-30 2019-04-16 哈尔滨工业大学（深圳） 一种定位方法及系统、设备、存储介质
CN111050275A (zh) * 2019-11-26 2020-04-21 武汉虹信技术服务有限责任公司 一种基于rssi特征值的蓝牙定位方法
US11259266B2 (en) 2019-12-06 2022-02-22 Industrial Technology Research Institute Distance estimation device and method and signal-power calibration method
TWI722700B (zh) * 2019-12-06 2021-03-21 財團法人工業技術研究院 測距裝置、測距方法和校正訊號功率之方法
CN111965600A (zh) * 2020-08-14 2020-11-20 长安大学 一种强遮挡环境下基于声音指纹的室内定位方法
CN112070201A (zh) * 2020-08-26 2020-12-11 成都睿芯行科技有限公司 一种基于Gmapping提升建图速度的方法
WO2022054022A1 (fr) * 2020-09-14 2022-03-17 Mahan Tabatabaie Localisation intérieure basée sur des activités précédentes
US11805389B2 (en) * 2020-10-27 2023-10-31 International Business Machines Corporation Evaluation of device placement
US20220132270A1 (en) * 2020-10-27 2022-04-28 International Business Machines Corporation Evaluation of device placement
WO2022250675A1 (fr) * 2021-05-27 2022-12-01 Hewlett-Packard Development Company, L.P. Déductions d'emplacements de dispositifs électroniques
CN113486223A (zh) * 2021-06-07 2021-10-08 海南太美航空股份有限公司 一种航线显示方法、系统和电子设备
CN114353787B (zh) * 2021-12-06 2024-05-10 理大产学研基地(深圳)有限公司 一种多源融合定位方法
CN114353787A (zh) * 2021-12-06 2022-04-15 理大产学研基地(深圳)有限公司 一种多源融合定位方法
CN114745658A (zh) * 2022-03-01 2022-07-12 华中科技大学 基于决策融合的距离估计方法及介质
CN114745658B (zh) * 2022-03-01 2024-06-04 华中科技大学 基于决策融合的距离估计方法及介质
CN115087095A (zh) * 2022-06-21 2022-09-20 五邑大学 一种基于csi的加权knn的可见光室内定位方法
CN115087095B (zh) * 2022-06-21 2024-05-03 五邑大学 一种基于csi的加权knn的可见光室内定位方法
WO2024088225A1 (fr) * 2022-10-25 2024-05-02 华为技术有限公司 Procédé et système de télémétrie bluetooth, et dispositif électronique

Similar Documents

Publication Publication Date Title
WO2016138800A1 (fr) Optimisation d'estimations de position d'un dispositif permettant une localisation à l'intérieur
Wen et al. Fundamental limits of RSS fingerprinting based indoor localization
He et al. Contour-based trilateration for indoor fingerprinting localization
Niu et al. WicLoc: An indoor localization system based on WiFi fingerprints and crowdsourcing
Jiang et al. ALRD: AoA localization with RSSI differences of directional antennas for wireless sensor networks
He et al. Fusing noisy fingerprints with distance bounds for indoor localization
US9648466B2 (en) Generating a model for positioning
Chen et al. A practical approach to landmark deployment for indoor localization
US9674656B2 (en) Wireless-based localization using a zonal framework
Elbakly et al. A robust zero-calibration RF-based localization system for realistic environments
He et al. INTRI: Contour-based trilateration for indoor fingerprint-based localization
Jiang et al. Wi-Fi fingerprint based indoor localization without indoor space measurement
He et al. Maxlifd: Joint maximum likelihood localization fusing fingerprints and mutual distances
CN104038901A (zh) 一种减少指纹数据采集工作量的室内定位方法
Zhao et al. GraphIPS: Calibration-free and map-free indoor positioning using smartphone crowdsourced data
Figuera et al. Nonparametric model comparison and uncertainty evaluation for signal strength indoor location
Zhou et al. Application of backpropagation neural networks to both stages of fingerprinting based WIPS
Gong et al. A usability-enhanced smartphone indoor positioning solution using compressive sensing
Seco et al. A review of multidimensional scaling techniques for RSS-based WSN localization
Huilla et al. Indoor localization with Wi-Fi fine timing measurements through range filtering and fingerprinting methods
Marquez et al. Understanding LoRa-based localization: Foundations and challenges
Kashniyal et al. Wireless sensor networks localization using progressive Isomap
Tian et al. Wireless Localization Techniques
Xu et al. Variance-based fingerprint distance adjustment algorithm for indoor localization
Xie et al. Accelerating crowdsourcing based indoor localization using CSI

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 16758416

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 16758416

Country of ref document: EP

Kind code of ref document: A1