WO2016042296A2 - Determining the position of a mobile device in a geographical area - Google Patents
Determining the position of a mobile device in a geographical area Download PDFInfo
- Publication number
- WO2016042296A2 WO2016042296A2 PCT/GB2015/052615 GB2015052615W WO2016042296A2 WO 2016042296 A2 WO2016042296 A2 WO 2016042296A2 GB 2015052615 W GB2015052615 W GB 2015052615W WO 2016042296 A2 WO2016042296 A2 WO 2016042296A2
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- orientation
- acceleration
- signal
- signals
- motion
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/20—Instruments for performing navigational calculations
- G01C21/206—Instruments for performing navigational calculations specially adapted for indoor navigation
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/10—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 by using measurements of speed or acceleration
- G01C21/12—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 by using measurements of speed or acceleration executed aboard the object being navigated; Dead reckoning
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/10—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 by using measurements of speed or acceleration
- G01C21/12—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 by using measurements of speed or acceleration executed aboard the object being navigated; Dead reckoning
- G01C21/16—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 by using measurements of speed or acceleration executed aboard the object being navigated; Dead reckoning by integrating acceleration or speed, i.e. inertial navigation
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C22/00—Measuring distance traversed on the ground by vehicles, persons, animals or other moving solid bodies, e.g. using odometers, using pedometers
- G01C22/006—Pedometers
Definitions
- the present invention relates to methods and devices related to determining the position of a mobile device in a geographical area.
- the invention is particularly applicable where the geographical area is an indoor area. More particularly, but not exclusively it relates to improved methods of processing data from orientation and acceleration sensors.
- Indoor tracking and navigation is a fundamental need for pervasive and context-aware smartphone applications.
- Applications of indoor positioning include smart retail, navigation through large public spaces like transport hubs, and assisted living.
- the ultimate objective of indoor positioning systems is to provide continuous, reliable and accurate positioning on smartphone class devices.
- Inertial tracking, or “dead reckoning”, with or without the aid of existing maps, has received a lot of attention recently, but existing solutions lack generality and robustness. They are typically targeted at achieving high accuracy in specific contexts, and are fine tuned for specific phone placements (i.e. positions on the user's body, etc.), users, devices or environments. When tested in different conditions from the ones designed for, their performance degrades significantly.
- the majority of existing solutions assume that the user holds the device in a specific way (e.g. texting mode) [1] (see list of references below). More advanced solutions first use a classifier to infer the user walking pattern and phone placement (e.g. hand swinging, texting, etc.) [2], [3], and then exploit this information to optimize the inertial tracking system [4].
- a classifier to infer the user walking pattern and phone placement (e.g. hand swinging, texting, etc.) [2], [3], and then exploit this information to optimize the inertial tracking system [4].
- the state of the art approaches can only handle a limited number of phone placement options. When the user deviates from this predefined set, the tracking system is at a loss for how to handle the new case.
- the inertial tracking system typically assumes knowledge of a number of parameters that account for user, device and environment variability.
- Examples of user variability parameters include those that relate the user height, step frequency and acceleration variance, to their step length [5]-[8]. Further parameters are needed to model the noise of inertial sensors, and environment-specific magnetic distortions [9].
- PDR pedestrian dead reckoning
- the invention employs two main principles:
- R-PDR a robust pedestrian dead reckoning algorithm that is based on general physics principles underpinning human motion, and avoids using context-specific heuristics, such as if phone is in a pant (or "trouser”) pocket, use technique A; else if in text mode, use technique B.
- PDR pedestrian dead reckoning
- Lifelong learning While PDR parameters often require context-specific tuning, in the present invention this information is learnt, rather than being manually tuned or assumed as an initial input. To this end, it uses an interplay between the pedestrian dead reckoning (PDR) and map-matching (MM) layers.
- PDR pedestrian dead reckoning
- MM map-matching
- Existing work typically feeds PDR output into a map matching technique (e.g. particle filter/HMM/CRF), which exploits map information to correct the inertial trajectory.
- a part of the invention which named herein Lifelong Learning Tracker (LL-Tracker) introduces a reverse link between the two layers. It augments the functionality of the map-matching layer to learn context-specific parameters, which are then fed back to the underlying PDR layer.
- LL-Tracker Lifelong Learning Tracker
- the contributions and aims of the present invention include:
- a lifelong learning tracker (LL-Tracker) that eases the effort of parameter tuning by introducing a feedback loop between the map matching and pedestrian dead reckoning layers.
- An aim of the present invention is to provide an indoor tracking system that is robust to noise introduced by inertial measurement unit (IMU) sensors and to context variability.
- IMU inertial measurement unit
- PDR algorithms typically include processes for motion mode recognition, step detection, step length estimation and step heading estimation.
- Motion classification Despite a plethora of work on recognising different motion modes, to our knowledge, there is still a lack of a comprehensive classification scheme that captures the wide range of human motion and device attachment. Moreover, solutions that capture many motion modes are typically complex, since they have to include a large number of motion specific optimisation rules, especially for solving the problem of step detection.
- Device orientation estimation The performance of the PDR algorithm largely depends on the ability to accurately track the orientation of the mobile device. This task is particularly challenging in the context of unconstrained (mobile/wearable) devices embedded with low- cost IMU sensors. Generally, the gyro drift and magnetic disturbances, especially in indoor environments, are the major obstacles to accurate orientation estimation. Addressing this problem is essential, because orientation information is used by nearly all tasks of the PDR algorithm.
- Acceleration long/short term drift Another major challenge is the correction of short-term and long-term drift in the acceleration signal.
- the noise that distorts the acceleration signal is a major issue because it causes inaccuracy in the orientation estimation.
- a small tilt error of the device would distort the acceleration signals projected to the earth coordinate frame, with part of the vertical acceleration, e.g. gravity component, going to the horizontal acceleration.
- Parameter tuning The inertial tracking system generally needs careful tuning of parameters such as sensor bias and step length estimation parameters, to name a few.
- parameters such as sensor bias and step length estimation parameters, to name a few.
- FIGURE 1 is the overall architecture of the system according to an embodiment of the invention.
- FIGURE 2 is the acceleration and orientation signals of some typical motions of a device carried by a user
- FIGURE 3a is a graph comparing gravity estimated with various techniques
- FIGURE 3b shows graphs of acceleration estimated with and without a technique of an embodiment of the invention
- FIGURE 4 shows graphs the effect on components of acceleration processed with a technique of an embodiment of the invention
- FIGURE 5a shows a comparison of the detection of user step according to different techniques and different placements of a device on a user's body.
- FIGURE 5b shows a comparison of the rejection of various non-step motions of user according to different techniques
- FIGURE 6a is a set of graphs illustrating least squares heading estimation in an embodiment of the invention.
- FIGURE 6b is a comparison of various heading estimation techniques
- FIGURE 6c is a comparison of measured trajectories obtained by various heading estimation techniques
- FIGURES 7a, 7b and 7c show a series of trajectories demonstrating the efficacy of learning sensor bias
- FIGURES 8a, 8b and 8c show experimental results showing the importance and effectiveness of the lifelong learning in accordance with an embodiment of the invention on step length estimation
- FIGURES 9a and 9b are graphs comparing the accuracy of various tracking algorithms
- FIGURES 10a, 10b and 10c are graphs demonstrating the effectiveness of lifelong parameter methods
- FIGURES 1 1a, 1 1 b and 11 c are graphs demonstrating the effectiveness of lifelong bias learning
- FIGURES 12a and 12b show the performance of an embodiment of the invention in various environments
- FIGURES 13a, 13b, and 13c illustrate experiments performed in various environments
- FIGURE 14 is a diagram of a motion identification module of an embodiment of the invention.
- FIGURE 15 is a diagram of a step detection module of an embodiment of the invention
- FIGURE 16a is a diagram of a known Kalman filter arrangement for the orientation signal
- FIGURE 16b is a diagram of a Kalman filter arrangement for the orientation signal using both a magnetometer and gravity signal as update inputs
- FIGURE 16c is diagram of a Kalman filter arrangement for the orientation signal using a gravity as an update signal in accordance with an embodiment of the present invention
- FIGURE 17 is a diagram of an acceleration module of an embodiment of the invention.
- Fig. 1 shows the system architecture in an embodiment of the invention.
- a smart phone 100 having sensors such as an accelerometer, magnetometer, gyroscope, etc. is provided with software, the modules of which are shown in the rest of the Figure.
- the accelerometer, magnetometer and gyroscope are the main sensors but others could be taken into account by the system to provide position, acceleration and orientation signals, such as a barometer, light sensors and so on.
- a first layer of the software a Robust Pedestrian Dead Reckoning (R-PDR) 10, is equipped with three novel modules: Firstly, it features a novel motion classification scheme 1 , which uses only two classes in its classification, but nonetheless captures a wide spectrum of human motion and device placement combinations. Secondly, leveraging simple observations about human motion, R-PDR has an acceleration estimate module 2 with a robust mechanism for correcting short-term and long-term acceleration bias. Thirdly, the R- PDR has a device orientation estimating module 3 that is novel and robust. (The word "layer” is uses herein only to denote a larger software module that contains smaller ones of modules described in.)
- This data is then fed into the Map Matching layer 20, which takes into account map constraints to correct the problem of long-term drift. Unlike existing systems where information flows upwards only, there is a feedback loop provided.
- the map-matching layer 20 has a location estimation module which uses the trajectory along with a map (101 indicates a display of the map on a screen of the device) to provide a location, the location being used, for example to inform the user of it.
- the location signal namely a mean and variance, or some other estimate and measure of uncertainty of the location, is passed to a lifelong learning module 22 which fine tunes parameters used by the R-PDR layer 10, and it feeds those back from time to time to the the R-PDR layer 10.
- the resulting end-to-end algorithm is called herein a Lifelong Learning Tracker (LL-Tracker), which learns from experience and gradually gets better at handling familiar tracking scenarios. Later below the feedback loop is described in more detail and examples that illustrate the benefits of cross-layer optimisation are provided.
- the symmetrical mode is the class of motions that are impacted in a similar manner by the movement of the left and right leg; the asymmetrical mode includes all other types of motion. Examples of typical symmetrical motion modes include, but are not limited to:
- Phoning The mobile device is held close to the head while walking;
- Heavy bag The mobile device is put in a heavy bag (so the hand of the pedestrian is not swinging) while walking;
- Shirt pocket The mobile device is put in a shirt pocket while walking;
- Static The mobile device is not moving, wherever it is put.
- some typical asymmetrical motion modes include:
- Hand swinging The mobile device is held in a swinging hand while walking; • Trouser pocket: The mobile device is put in a trouser pocket (front or back) while walking;
- the mobile device is fastened to the belt while walking;
- Handshaking The mobile device is (possibly periodically) shaken while not moving.
- Fig. 2 illustrates a few examples of symmetrical and asymmetrical motion modes. Notice that all symmetrical motion modes have similar periodic acceleration signals, with a period corresponding to a single step, while all asymmetrical motion modes have similar periodic orientation patterns, with a period corresponding to two steps.
- the benefits of classifying motions into symmetrical and asymmetrical are three fold. First, it is a comprehensive scheme that leaves no motion out. Second, the motion classification step itself is a simple task, which is implemented with simple cross correlation with sine waves (described later below). Lastly, the fact that symmetrical motions have periodic acceleration signals while asymmetrical motion modes has periodic orientation signals greatly facilitates the task of step detection (again discussed later below). So far a scheme for motion classification has been discussed; its implementation in the motion identification module is described later below.
- IMU inertial measurement unit
- So Figure 16A shows a prior art module for processing orientation information.
- a Kalman filter 161 receives the orientation signal from the gyro (i.e. the base angular velocity signal from the gyro sensor integrated to become the orientation signal) with the orientation signal being provided as the signal being filtered, with a magnetometer signal being applied as the update input signal to the Kalman filter.
- Figure 16B shows a similar part of the present embodiment (which is part of the orientation tracking module 3).
- This also employs a Kalman filter 161 ' but has an additional update signal, which is a signal indicating the direction of the gravity acting on the device. So this signal also is used to update the filtered orientation signal.
- the Kalman filter may be at least any of the variants of Kalman filter mentioned above.
- a gravity signal can be provided in various ways. Some known methods of that per se, as well as, the preferred method in this embodiment, which is of superior accuracy are described below.
- the first approach proposed by Mizell et. al [15] uses the mean over a window of fixed duration to estimate the gravity. This approach is elegant and simple but it is not suitable for mobile devices used by pedestrians. The major reason is that a change in orientation introduces a considerable lag to accurate gravity estimation. To reduce the lag, it is possible to shorten the time window over which the gravity is estimated [16]. However, this comes with a decrease in the accuracy of gravity estimates.
- the second approach proposed by Hemminki et. al [17], estimates gravity only when the variation of the acceleration signals over a period of time is below a certain threshold. This approach reduces the lag introduced by the first approach, but also suffers from two limitations. First, this approach works only when the mobile device is static for a period of time; it does not work when the user is in motion, which is when gravity estimates are mostly needed.
- this embodiment uses a novel algorithm that can accurately estimate the gravity vector without lag and that works all the time.
- the key fact that it exploits is that gyro sensors are reliable over a short period of time.
- the algorithm is as follows. Assume we want to estimate the gravity at time t given a window of historical raw accelerometer data a t - T: t of size T. In a first step (see Figure 14) the acceleration signals from the accelerometer a t - T: t-i are projected to the phone's coordinate system at time stamp t and then in a second step the gravity vector at time t, g t is estimated as the mean of the acceleration readings after the projection:
- FIG 16C shows a more complete version of the details of the orientation tracking module 3 of this embodiment.
- This has the Kalman filter 161' (as in Figure 16B) with the gravity signal supplied from a gravity estimator module 162 that performs the calculation of equation (1) above.
- This module 162 takes as its input a signal indicative of the acceleration of the device (e.g. the raw accelerometer data) and values for the matrix R of Equation 1 provided by a t to ⁇ calculator 63.
- the proposed gravity estimation algorithm can make the long term orientation tracking of the device robust (i.e. the output of the Kalman filter), it is not sufficient to accurately estimate acceleration on the horizontal plane, which is important for step detection.
- the major reason is that even a very small tilt of the device, e.g. two degrees, which may not be corrected by the above algorithm, can significantly impact the acceleration signal on the horizontal plane. For instance, a tilt of two degrees can only change the acceleration reading in Z axis from 9.806 m/s 2 to 9.801 m/s 2 , but it can offset the
- R is the rotation matrix that corrects the orientation drift in the acceleration vector a' from the device's acceleration sensor at time t
- [a x t , a y t , a z t ] are the components of a t at time t expressed in the Earth coordinate system (preferably the NED - North East Down) with z being vertical
- At is the acceleration sampling interval.
- R is similarly expressed in those coordinates.
- a' t R T .a t , where a' is the corrected acceleration vector, and R T is the one calculated from equations (3) and (4) for the interval t 0 to t e containing a t .
- Fig. 4 shows a typical example of acceleration signals before and after the REPUT algorithm. It can observed that the acceleration signals along X and Y axes drift significantly over time, if we do not exploit the repetitive pattern assumption; the drift is successfully corrected by REPUT.
- Figure 17 shows the preferred form of the acceleration module 2 in which the R matrix is used to correct the acceleration signals to provide acceleration signals output from module 2.
- the raw acceleration vector signals have their components converted from those in the device frame to those in the Earth frame by acceleration projector module 171 , using orientation signals.
- the acceleration signals are then applied to the R optimiser module 172 to derive R of Equations 3 and 4, which R is then applied to in turn to the acceleration signals, by an acceleration corrector 173 to correct them as noted above.
- the R found is applied to the acceleration vectors in the interval t 0 to t e , which were used to calculate that value of R.
- the motion identification module 4 exploits the novel classification scheme described above, which classification consists of two motion classes only - symmetrical and asymmetrical. In addition, it uses the corrected orientation and acceleration (vertical component) signals in order to identify the motion class.
- the orientation signal in this embodiment is output of Figure 16C (i.e. the Kalman filtered one), for example the pitch component of that signal is used but since the mobile device can be held in various orientations the roll and yaw are preferably also used, and the appropriate one selected at a particular time. .
- the acceleration signal is, in this embodiment, the raw signal fro the accelerometer, with gravity subtracted.. As can be seen from Figure 2, in symmetrical motion modes, e.g.
- FIG. 14A shows the motion identification module 4.
- DFT modules 141 and 142 respectively compute a discrete Fourier transform (DFT) of each of the acceleration and orientation signals, from the accelerometer (less gravity) and the robust orientation tracking module 3, over the frequency range that corresponds to human walking, typically between 1.2 Hz and 2.5 Hz, thereby detecting the energy in that range.
- a Bayesian classification module 143 receives those DFTs respectively and decides whether the device placement is resulting in symmetric (vertical acceleration has stronger periodicity than orientation) or asymmetric motion (orientation has stronger periodicity than vertical acceleration), outputting a signal indicting that decision.
- the energy detector module uses a normalised crossed relation between the DFTs and a two state hidden Markov model to make the decision. So it assesses which one of the vertical acceleration or orientation waveforms has the strongest periodic component.
- the motion identification module could use other signals indicative of the orientation and acceleration, in particular raw data from accelerometers commonly used can be used. Also other methods of assessing which has the strongest periodic component may also be used.
- the next task performed is to detect human steps.
- Typical existing step detection algorithms detect steps by identifying the cyclic patterns in acceleration signal, e.g. peak detection [5], [7], [19], zero-crossings [20]- [22], spectral analysis like Fourier transform [23], [24], short time Fourier transform [25], and continuous wavelet transform [8], and auto/cross correlation [26], [27].
- the aforementioned techniques assume that people generate periodic acceleration signals only when walking but this is not true. Hence, they cannot distinguish between a real step and a simple hand swing without the user really moving forward. This is because the two actions can actually generate very similar acceleration signal patterns.
- step pattern extraction To summarise, existing techniques only focus on identifying the boundaries of step-like signal segments, referred to as step pattern extraction.
- the present embodiment not only proposes a novel way of performing step pattern extraction, but also introduces an additional classification step for distinguishing whether the extracted step is real or fake.
- a real step is herein defined as a step with locomotion, as opposed to a fake step without locomotion.
- An example of a fake step is one detected when a person swings the phone in her hand without actually moving her legs. Details of the novel step pattern extraction and step classification modules described below.
- the step identification approach of this embodiment aims to identify the boundaries of step-like signal segments in a lightweight online fashion.
- the approach is to leverage motion identification method described above to extract steps very accurately using time domain analysis. The analysis used depends on whether the motion is identified as symmetrical or asymmetrical.
- Figure 15 shows the details of the step detection module 5.
- the module 5 uses the vertical component of the acceleration signal for step detection, because it has a simple sine-wave-like pattern in symmetrical motion mode. Step boundaries here are detected with enhanced zero crossing detector 151 similar to [22] applied to the acceleration signal, which produces step boundary indications.
- step boundaries are the peaks and troughs of the device orientation signal - e.g. first step is from trough to peak, second step is from peak to trough etc.
- the orientation signal is filtered by low-pass filter 152.
- the orientation signal can have some non-fundamental frequency components complicating the shapes of the peaks and troughs, so the filter 153 improves the reliability and accuracy of the peal and trough detection.
- the acceleration signal is also filtered by a low-pass filter 155 before it is applied to zero crossing detector 151.
- a multiplexer 154 is shown as part of the step detection module 5, connected to receive the step boundary indications form the zero-crossing detector 151 and the peak and trough detector 153 and to choose between them based on the indication, from the motion identification module 4, of whether the motion is symmetric or asymmetric.
- the modules are implemented in software and the multiplexer is only symbolic - the software first checks the indication of whether the motion is symmetric or asymmetric and only performs the detection of step boundaries on the acceleration or orientation signals accordingly, to save computing resources.
- the module selects which a component of the orientation signal to make the detection of peaks and troughs.
- the step pattern extraction step does not distinguish between real and fake steps.
- a further classification step is provided that takes into account informative features of the orientation and acceleration signals.
- the horizontal and vertical displacements are informative features in telling whether real locomotion happens to the mobile device. In this embodiment they are derived from the double integration of the acceleration signal during one step (in the x-y plane / z axis respectively) but other methods are possible. These two features are extremely informative because they can capture the fundamental difference between real and fake steps. In real steps, the horizontal displacement is higher and the vertical displacement is lower than in fake steps.
- the cross correlation between the acceleration signal and the orientation signal is also identified as an effective feature to reject fake steps because the acceleration and orientation signals tend to have the same period if we shake the device, which is not true for normal walking behaviour.
- the orientation offset is another feature used for real/false step classification.
- the orientation offset measures how much the orientation has changed during one step. This feature is proposed to prevent the
- the module provided for disguising real and fake steps from those signals can be, for example, any machine learning or Bayesian classifier (decision tree, svm, neural networks, etc.).
- R-Step Length Estimation Unlike for the foot-mounted inertial navigation system [32], double integration of the acceleration to get the step length is not applicable to handheld devices, reasons being zero velocity update (ZUPT) is not available for handheld devices because only the heel can be regarded as (almost) static when one foot is planted on the floor during walking period.
- ZUPT zero velocity update
- Various models have been proposed to address the step length estimation problem. The simplest approach is to make it a constant [33] because
- step length [29].
- the most common approaches used to estimate the step length are linear [6], [8] or nonlinear [5], [7] model to relate the step length to variables like pedestrian height, step frequency, acceleration variance, etc. These models are easy to implement and thus widely adopted in practice.
- An alternative of the linear/nonlinear model is regression. Machine learning regression techniques like support vector machine [3], [28] and neural networks [34] have also been applied based on various features extracted from the sensor data.
- This embodiment uses a simple step length model [35], which uses accelerometer data to detect the step frequency (denoted with f t at time t) which is later used to estimate step length l t as follows:
- h is the pedestrian height
- a is the step frequency coefficient
- ⁇ is the step constant given pedestrian height
- ⁇ is the step constant
- the accuracy of this model is not satisfactory by itself because step length parameters vary across different users and environments.
- the R-Step Length Estimation method of this embodiment however benefits from being fed with carefully configured parameter values thanks to the feedback loop between R-PDR and Map Matching. This is referred to here as unsupervised lifelong learning approach as LL-Tracker, and describe in detail later below.
- WalkCompass samples the acceleration signal during the segment of a step with maximum deceleration. After removing noise and rotating samples to a stable coordinate system, it averages them to obtain the walking direction in 3D. It then projects the estimated direction to the horizontal walk plane. This approach works for various device placements, but is sensitive to noise as shown below.
- a straight line that best matches the acceleration readings of a single step in the horizontal plane (the two components X and Y) is fitted using simple least squares.
- Bayesian regression can be used instead of least squares.
- Figs. 6(b) and 6(c) compares the technique of this embodiment and competing techniques for tracking a user in a sports centre for 1 10 seconds. Notice that the proposed R-PDR approach can detect and avoid noisy heading estimates, thus yielding higher accuracy than the competing approaches.
- the heading estimation of this embodiment relies on being able to correctly rotate acceleration data to the earth coordinate system. This requires accurate device orientation estimates, which in turn depends on the quality of gyro data as discussed above.
- This embodiment addresses the issue of long term drift of the gyro sensors as discussed above, and further improves the accuracy of R-Heading Estimation, by learning the sensor bias via the lifelong learning approach of LL-Tracker, as discussed below.
- An illustrative example is the implementation of map matching using a Simultaneous Localisation And Mapping approach, such as WiFiSLAM [37].
- Bayesian filters are also adapted in map matching, e.g. Zee [27], [36], where the radio map or the magnetic distortion map [38] are learned after the trajectories have been matched to a given map.
- map matching e.g. Zee [27], [36]
- LL-Tracker This system of a end-to-end lifelong learning tracking system is referred to here as LL-Tracker.
- This consists of R-PDR 10, map matching 20 and a feedback loop from map matching to R-PDR, as shown in Fig. 1.
- the parameter learning process which is central to the feedback loop, and then followed by a couple of illustrative examples.
- R-PDR The accuracy of R-PDR depends on a number of parameters employed in its four modules 4, 5, 6 and 7. These are hard to tune manually because they may depend on the device, attachment, user and environment. To address this issue, to the present
- R-PDR parameter tuning becomes the following optimization problem:
- x is the parameter (or vector of parameters) of R-PDR (or indeed another PDR implementation) that requires tuning
- S(x) is the matched trajectory
- Z(x) is the raw trajectory fed from R-PDR to map matching.
- EM expectation maximization
- the soft EM approach does not work because the optimization of parameters x in the M-step actually changes the state space of the model, which makes the E-step unable to evaluate the expectation in the next iteration [1]. Therefore, the hard EM approach, also known as Viterbi training is employed to solve the optimization problem [39].
- the learning of sensor bias is crucial to the performance of PDR algorithms.
- the motion sensor bias is the bottleneck that stops the inertial tracking from being widely used because the sensor bias is accumulated, leading to significant errors in the trajectory, especially when used in open space where no additional constraints can be applied.
- motion sensors also suffer from both time drift and thermal drift, which makes the bias varying with time and temperature. Therefore, the lifelong learning of sensor bias is essential to accurate PDR tracking. In practice all three motion sensors that are usually available, the accelerometer,
- Fig. 1 1 shows the trajectory with one Nexus 5 device in an indoor basketball court without any floorplan or other map constraints. It is observed that the trajectory largely deviates from the ground truth without sensor bias learning.
- the second example shows how we can use the feedback loop between R-PDR and map matching to improve the performance of R-Step Length Estimation, module 6 of the R-PDR 10.
- the method of this embodiment was used to learn the step constant ⁇ (Eqn. (5)) for different individuals because 1) the average step length plays a crucial role in the tracking accuracy due to the high consistency of step length in human walking patterns; and 2) the parameters a and ⁇ are very similar for different individuals in our experiments.
- step constants ⁇ are very different for different pedestrians (U 1 to U4), and very different even for the same pedestrian in different environments (office, museum, and market) as shown in Fig. 8(b).
- Fig. 8(c) shows that the step constant that maximizes the conditional probability p(S
- the software has the module structure has described above. In alternative embodiments, some or all of the calculations described may be performed by hard-wired circuits.
- the calculations may be carried out on other devices, e.g. computers, connected to the device by a communications link.
- the computer be local, for example in a building that the user uis navigating, or distant, for example connected to the device the Internet.
- LL-Tracker has been evaluated and compared against competing approaches in three real-world settings with known floor plans: an office building (65x35m 2 ), a museum (109x89m 2 ), and a sports center hall (30 ⁇ 20m 2 ). For the majority of the tests, the office building was used as it has the most distorted trajectories. Overall, 224 trajectories of average length over 400m were collected over 30 days. Error is expressed in [m] RMS.
- Ground Truth To provide accurate ground truth, numbered labels were placed along corridors and within rooms on a 2m grid. During the experiments, the test subject always held one camera at hand. These labels were filmed at the same time experiments were conducted. The time-synchronized video streams were then mapped to locations on the floorplan, and intermediate locations interpolated using footstep timing, also obtained from the video.
- LL-Tracker combines R-PDR and map matching with lifelong learning, i.e. with the feedback loop from map matching to fine tune the parameters of R-PDR.
- R-Tracker consists of R-PDR and map matching without lifelong learning.
- Competing Algorithms Practical indoor positioning algorithms were required to be infrastructure-free or use existing infrastructures like WiFi and Bluetooth low energy (BLE).
- Existing practical algorithms fall into two categories: RF category where only WiFi/BLE are used and fusion category where inertial data and WiFi/BLE data are fused to perform positioning.
- RF category HORUS [40], RADAR [41], and EZ [42].
- the state-of-the-art algorithms in the fusion category are MapCraft [1 ], Zee [27], UnLoc [43], WifiSLAM [37], and the algorithm in [36].
- HORUS 1 takes a set of RSS r, 2) estimates the likelihood of this set given RSS-location pairs in the radio map as
- MapCraft [1 ] is a recently proposed lightweight indoor positioning algorithm, which employs conditional random fields for map matching. It has reported an RMS error of around 1 ⁇ 2 meters but, so far, it has been designed and implemented with the assumption of mobile devices held by the user in text mode. To avoid penalising it for this reason, we compare the performance of the proposed algorithms using a variety of mobile device attachments, to that of MapCraft using a single device attachment (text mode).
- Fig. 9 compares the error cumulative distribution function of the four approaches. It is observed from Fig. 9(a) that LL- Tracker outperforms all other approaches, followed by R-Tracker, and lastly MapCraft when only IMU sensors are used. The RMS errors for these three approaches are 0.87m (LL- Tracker), 1.36m (R-Tracker), and 1.69m (MapCraft).
- LL-Tracker, R-Tracker, and MapCraft use inertial measurements at the pedestrian dead reckoning layer and perform map matching using information about the environment's floor plan - they do not assume knowledge of a radio map, although they can flexibly exploit this information if it becomes available.
- the performance of these three approaches and HORUS were then compared, when a WiFi radio map was available and used by all approaches. It can be observed from Fig. 9(b) that LLTracker also has the best accuracy with a RMS error of 0.86m, followed by R-Tracker (1.07m), MapCraft (1.61 m), and HORUS (4.62m).
- LL-Tracker, R- Tracker, and MapCraft fuses IMU data, radio map and floor plan constraints while HORUS only exploits knowledge of the radio map, but does not perform map matching.
- the RMS tracking error can be as high as 15m without the bias learning algorithms while the error can decrease to around only 2m if the gyro bias has been learned in the office environments before it is used in the basketball court.
- the sensor bias learning still improves the tracking accuracy by over 20% (from 0.96m to 0.75m).
- Attachment Variability Five typical attachments were tested in this experiment to evaluate the robustness of LL-Tracker, including handheld, watch, glasses, shirt pocket, and trousers pocket. This experiment was also conducted by User 1 with Nexus 5 phone in the office environment. It is observed from Fig. 1 1 (c) that the RMS errors of LL-Tracker are extremely similar for different attachments. Please note that MapCraft only works for handheld attachment.
- the performance of LL-Tracker was evaluated in a variety of environments, namely an office environment, a museum, and a sports center. All of these environments have different floor plans and methods of construction which affect the obtained sensor data.
- the museum was a multi-storey stone building with large, open spaces. Testing was conducted on the ground floor.
- the office environment (where the majority of the tests have been conducted) was a multi-storey office building with a stone and brick construction, reinforced with metal bars; testing was conducted on the fourth floor.
- the sports center was a big multi-functional hall where the experiments were conducted in the basketball court for the convenience of ground truth collection. In this set of experiments, data from all 15 experiment subjects, 5 different devices, and 5 different attachments were taken into account.
- Fig. 13 shows a couple of illustrative examples of how LL-Tracker succeeds in accurately tracking a pedestrian through the office and museum environments.
- the cumulative distribution functions of these environments are shown in Fig. 12(a).
- the RMS errors are 0.86m in the office and 0.90m in the museum.
- LL-Tracker starts tuning R-PDR parameters after it converges to a single trajectory on the map with high probability; the CDF of the convergence distance in the office and museum environments is presented in Fig. 12(b).
- Fig. 1 System architecture of LL-Tracker.
- FIG. 2 Acceleration and orientation signals of some typical motions. Note how symmetrical motions have strong periodicity in acceleration, whereas asymmetrical motion has strong periodicity in pitch.
- FIG. 3 Illustrative examples of robust orientation estimation.
- Fig. 4 Acceleration signals before and after REPUT.
- the X and Y acceleration signals are expected to have mean values close to zero in a window.
- the mean acceleration drifts significantly away from zero without REPUT.
- Fig. 6(a) Least squares heading estimation.
- Fig. 7 Inertial trajectories of 0.5 km in a basketball court without any map constraints before and after sensor bias learning, showing an improvement in loop closing loop error from 17.3m to 2.3m, and demonstrating the importance and effectiveness of the learning algorithm.
- Fig. 8(a) Step constants of different pedestrians in an office environment.
- Fig. 9 Accuracy comparison of various tracking algorithms.
- Fig. 11 Tracking accuracy of LL-Tracker with different devices and attachments, showing the effectiveness of the lifelong bias learning algorithms.
- the bias was learned in the office environment for a period of time, and then tested both in the same office and a completely different sports centre environment.
- Fig. 12(a) Multi-site performance to validate the robustness of LL-Tracker against environmental variations.
Landscapes
- Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Navigation (AREA)
- User Interface Of Digital Computer (AREA)
Abstract
Description
Claims
Priority Applications (5)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910651795.5A CN110501011B (en) | 2014-09-15 | 2015-09-10 | Determining a location of a mobile device in a geographic area |
EP15766567.0A EP3194889B1 (en) | 2014-09-15 | 2015-09-10 | Inertial tracking based determination of the position of a mobile device carried by a user in a geographical area |
US15/509,771 US10502576B2 (en) | 2014-09-15 | 2015-09-10 | Determining the position of a mobile device in a geographical area |
CN201580049724.8A CN107250727A (en) | 2014-09-15 | 2015-09-10 | Determine position of the mobile device in geographic area |
AU2015316575A AU2015316575B2 (en) | 2014-09-15 | 2015-09-10 | Inertial tracking based determination of the position of a mobile device carried by a user in a geographical area |
Applications Claiming Priority (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
GR20140100462 | 2014-09-15 | ||
GR20140100462 | 2014-09-15 | ||
GB1500411.2 | 2015-01-12 | ||
GBGB1500411.2A GB201500411D0 (en) | 2014-09-15 | 2015-01-12 | Determining the position of a mobile device in a geographical area |
Publications (2)
Publication Number | Publication Date |
---|---|
WO2016042296A2 true WO2016042296A2 (en) | 2016-03-24 |
WO2016042296A3 WO2016042296A3 (en) | 2016-06-30 |
Family
ID=52597465
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/GB2015/052615 WO2016042296A2 (en) | 2014-09-15 | 2015-09-10 | Determining the position of a mobile device in a geographical area |
Country Status (6)
Country | Link |
---|---|
US (1) | US10502576B2 (en) |
EP (1) | EP3194889B1 (en) |
CN (2) | CN110501011B (en) |
AU (1) | AU2015316575B2 (en) |
GB (1) | GB201500411D0 (en) |
WO (1) | WO2016042296A2 (en) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2016206119A1 (en) | 2015-06-26 | 2016-12-29 | Intel Corporation | Technologies for pedestrian dead reckoning |
WO2017202948A1 (en) * | 2016-05-24 | 2017-11-30 | Thales | Device for locating moving objects or people |
WO2017215024A1 (en) * | 2016-06-16 | 2017-12-21 | 东南大学 | Pedestrian navigation device and method based on novel multi-sensor fusion technology |
WO2020188275A2 (en) | 2019-03-20 | 2020-09-24 | Navenio Ltd. | Transition detection |
EP3812706A4 (en) * | 2018-08-08 | 2021-07-07 | Huawei Technologies Co., Ltd. | Method for obtaining movement track of user and terminal |
WO2022018399A1 (en) | 2020-07-20 | 2022-01-27 | Navenio Ltd. | Map matching trajectories |
DE102022101492A1 (en) | 2022-01-24 | 2023-07-27 | Bayerische Motoren Werke Aktiengesellschaft | Method and device for determining pose data relating to the pose of a vehicle |
Families Citing this family (38)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2015184160A (en) * | 2014-03-25 | 2015-10-22 | セイコーエプソン株式会社 | Reference value generation method, motion analysis method, reference value generation device, and program |
US9964409B1 (en) * | 2014-05-27 | 2018-05-08 | Apple Inc. | Localized map generation |
US10488222B2 (en) * | 2015-11-02 | 2019-11-26 | David Martin | Mobile device control leveraging user kinematics |
WO2016098457A1 (en) * | 2014-12-17 | 2016-06-23 | ソニー株式会社 | Information processing device, information processing method, and program |
EP3076128B1 (en) * | 2015-04-01 | 2021-09-22 | Tata Consultancy Services Limited | Backtracking indoor trajectories using mobile sensors |
AT517225B1 (en) * | 2015-06-12 | 2016-12-15 | Smartbow Gmbh | Method for locating animals using radio waves |
US10129691B2 (en) | 2016-10-14 | 2018-11-13 | OneMarket Network LLC | Systems and methods to determine a location of a mobile device |
US20200158517A1 (en) * | 2017-01-19 | 2020-05-21 | Mindmaze Holding Sa | System, methods, device and apparatuses for preforming simultaneous localization and mapping |
US11199411B2 (en) | 2017-05-15 | 2021-12-14 | Fujifilm Business Innovation Corp. | Systems and methods for utilizing graph based map information as priors for localization using particle filter |
US10812877B2 (en) * | 2017-05-15 | 2020-10-20 | Fuji Xerox Co., Ltd. | System and method for calibration-lessly compensating bias of sensors for localization and tracking |
GB201718507D0 (en) | 2017-07-31 | 2017-12-27 | Univ Oxford Innovation Ltd | A method of constructing a model of the motion of a mobile device and related systems |
WO2019084918A1 (en) * | 2017-11-03 | 2019-05-09 | Beijing Didi Infinity Technology And Development Co., Ltd. | System and method for determining a trajectory |
WO2019104188A1 (en) * | 2017-11-22 | 2019-05-31 | DeepMap Inc. | Improving accuracy of global navigation satellite system based positioning using high definition map based localization |
CN108413957A (en) * | 2017-12-06 | 2018-08-17 | 上海交通大学 | The method for carrying out pedestrian's course estimation under multiple carrying mode using mobile terminal |
CN108253992B (en) * | 2017-12-31 | 2021-07-02 | 深圳市易景空间智能科技有限公司 | Step counting method based on walking state |
CN108537101B (en) * | 2018-01-05 | 2022-03-25 | 浙江大学 | Pedestrian positioning method based on state recognition |
CN108871333B (en) * | 2018-05-31 | 2021-12-03 | 金华航大北斗应用技术有限公司 | Step length real-time calculation method based on indoor plan topology information |
US10598499B2 (en) * | 2018-07-25 | 2020-03-24 | Kabushiki Kaisha Toshiba | Method and device for accelerated map-matching |
RU2685767C1 (en) * | 2018-08-13 | 2019-04-23 | Самсунг Электроникс Ко., Лтд. | Method and device for strap down inertial navigation |
CN109341682B (en) * | 2018-11-12 | 2021-04-06 | 浙江工业大学 | Method for improving geomagnetic field positioning accuracy |
JP2020085783A (en) * | 2018-11-29 | 2020-06-04 | 一般財団法人生産技術研究奨励会 | Pedestrian-purpose positioning device, pedestrian-purpose positioning system, and pedestrian-purpose positioning method |
CN109856596B (en) * | 2019-02-21 | 2021-05-07 | 上海迹寻科技有限公司 | Method for collecting wireless node signal position fingerprint in high-speed moving state, computer readable storage medium and equipment thereof |
CN109870172B (en) * | 2019-02-25 | 2020-12-18 | 广州市香港科大霍英东研究院 | Step counting detection method, device, equipment and storage medium |
CN111693041B (en) * | 2019-03-12 | 2022-03-08 | 腾讯科技(深圳)有限公司 | Direction positioning method, device and storage medium |
CN109883429A (en) * | 2019-04-15 | 2019-06-14 | 山东建筑大学 | Zero-speed detection method and indoor pedestrian's inertial navigation system based on Hidden Markov Model |
US11321566B2 (en) * | 2019-08-22 | 2022-05-03 | Jpmorgan Chase Bank, N.A. | Systems and methods for self-learning a floorplan layout using a camera system |
CN110766154B (en) * | 2019-09-18 | 2022-05-10 | 北京邮电大学 | Pedestrian track inference method, device, equipment and storage medium |
CN112468965B (en) * | 2020-11-30 | 2023-03-14 | 成都新潮传媒集团有限公司 | Method and device for verifying authenticity of field visit and computer equipment |
CN112683269B (en) * | 2020-12-07 | 2022-05-03 | 电子科技大学 | MARG attitude calculation method with motion acceleration compensation |
US11815568B2 (en) | 2020-12-28 | 2023-11-14 | Stmicroelectronics, Inc. | System and method for fast magnetometer calibration using gyroscope |
CN112766218B (en) * | 2021-01-30 | 2023-05-30 | 上海工程技术大学 | Cross-domain pedestrian re-recognition method and device based on asymmetric combined teaching network |
CN113805214B (en) * | 2021-08-24 | 2023-03-24 | 广东汇天航空航天科技有限公司 | Combined navigation method, device, travelable equipment and computer storage medium |
CN114526727B (en) * | 2022-01-11 | 2023-11-21 | 重庆邮电大学 | Decision tree-based step number judging method |
CN114710744B (en) * | 2022-04-12 | 2023-03-21 | 电子科技大学 | Indoor positioning method integrating WiFi ranging and PDR calculation in depth |
CN115420288B (en) * | 2022-07-19 | 2024-05-17 | 北京航空航天大学 | Indoor positioning method for lightweight map information robust constraint PDR |
US20240045073A1 (en) * | 2022-08-04 | 2024-02-08 | Qualcomm Incorporated | Enhanced navigation mode with location detection and map layer switching |
WO2024077237A1 (en) * | 2022-10-07 | 2024-04-11 | Ceva Technologies, Inc. | Methods and systems for robust heading estimation in pedestrian dead reckoning navigation |
CN115423053B (en) * | 2022-11-07 | 2023-04-07 | 中国空气动力研究与发展中心计算空气动力研究所 | Method and related equipment for classifying unstructured flow field data suitable for airfoil section |
Family Cites Families (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
NL1014658C2 (en) * | 2000-03-16 | 2001-09-19 | Xsens Sports Technologies B V | Speedometer. |
US6826477B2 (en) | 2001-04-23 | 2004-11-30 | Ecole Polytechnique Federale De Lausanne (Epfl) | Pedestrian navigation method and apparatus operative in a dead reckoning mode |
US8751151B2 (en) * | 2012-06-12 | 2014-06-10 | Trx Systems, Inc. | System and method for localizing a trackee at a location and mapping the location using inertial sensor information |
US8204684B2 (en) | 2007-06-28 | 2012-06-19 | Apple Inc. | Adaptive mobile device navigation |
WO2009021068A1 (en) | 2007-08-06 | 2009-02-12 | Trx Systems, Inc. | Locating, tracking, and/or monitoring personnel and/or assets both indoors and outdoors |
JP5436574B2 (en) * | 2008-12-22 | 2014-03-05 | インテリジェント スペイシャル テクノロジーズ,インク. | System and method for linking real world objects and object representations by pointing |
US20110087450A1 (en) * | 2009-04-03 | 2011-04-14 | University Of Michigan | Heading Error Removal System for Tracking Devices |
US8380458B2 (en) * | 2009-11-20 | 2013-02-19 | Qualcomm Incorporated | In flight detection |
US9103917B2 (en) | 2010-02-12 | 2015-08-11 | Broadcom Corporation | Method and system for determining location within a building based on historical location information |
US8594971B2 (en) | 2010-09-22 | 2013-11-26 | Invensense, Inc. | Deduced reckoning navigation without a constraint relationship between orientation of a sensor platform and a direction of travel of an object |
US8694251B2 (en) * | 2010-11-25 | 2014-04-08 | Texas Instruments Incorporated | Attitude estimation for pedestrian navigation using low cost mems accelerometer in mobile applications, and processing methods, apparatus and systems |
WO2012072957A1 (en) | 2010-12-01 | 2012-06-07 | Commissariat à l'énergie atomique et aux énergies alternatives | Method and system for estimating a path of a mobile element or body |
US8498811B2 (en) | 2011-02-09 | 2013-07-30 | SenionLab AB | Method and device for indoor positioning |
US10145707B2 (en) * | 2011-05-25 | 2018-12-04 | CSR Technology Holdings Inc. | Hierarchical context detection method to determine location of a mobile device on a person's body |
EP2740256B1 (en) | 2011-08-04 | 2017-12-13 | Google LLC | Moving direction determination with noisy signals from inertial navigation systems on mobile devices |
US10330491B2 (en) * | 2011-10-10 | 2019-06-25 | Texas Instruments Incorporated | Robust step detection using low cost MEMS accelerometer in mobile applications, and processing methods, apparatus and systems |
US9454245B2 (en) | 2011-11-01 | 2016-09-27 | Qualcomm Incorporated | System and method for improving orientation data |
US9179266B2 (en) | 2012-03-17 | 2015-11-03 | MCube Inc. | Augmentation of indoor navigation methods and apparatus with map matching constraints |
US10215587B2 (en) * | 2012-05-18 | 2019-02-26 | Trx Systems, Inc. | Method for step detection and gait direction estimation |
US9297658B2 (en) | 2012-06-12 | 2016-03-29 | Trx Systems, Inc. | Wi-Fi enhanced tracking algorithms |
CN103827632B (en) * | 2012-09-06 | 2017-02-15 | 株式会社东芝 | Position detection device and position detection method |
US9999376B2 (en) * | 2012-11-02 | 2018-06-19 | Vital Connect, Inc. | Determining body postures and activities |
US9730029B2 (en) * | 2012-11-08 | 2017-08-08 | Duke University | Unsupervised indoor localization and heading directions estimation |
-
2015
- 2015-01-12 GB GBGB1500411.2A patent/GB201500411D0/en not_active Ceased
- 2015-09-10 WO PCT/GB2015/052615 patent/WO2016042296A2/en active Application Filing
- 2015-09-10 CN CN201910651795.5A patent/CN110501011B/en active Active
- 2015-09-10 AU AU2015316575A patent/AU2015316575B2/en active Active
- 2015-09-10 CN CN201580049724.8A patent/CN107250727A/en active Pending
- 2015-09-10 EP EP15766567.0A patent/EP3194889B1/en active Active
- 2015-09-10 US US15/509,771 patent/US10502576B2/en active Active
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2016206119A1 (en) | 2015-06-26 | 2016-12-29 | Intel Corporation | Technologies for pedestrian dead reckoning |
EP3314205A4 (en) * | 2015-06-26 | 2019-05-22 | INTEL Corporation | Technologies for pedestrian dead reckoning |
US10393540B2 (en) | 2015-06-26 | 2019-08-27 | Intel Corporation | Technologies for pedestrian dead reckoning |
US11199421B2 (en) | 2015-06-26 | 2021-12-14 | Intel Corporation | Technologies for pedestrian dead reckoning |
US11181378B2 (en) | 2016-05-24 | 2021-11-23 | Thales | Device for locating moving objects or people in a space equipped with at least one inductive surface |
WO2017202948A1 (en) * | 2016-05-24 | 2017-11-30 | Thales | Device for locating moving objects or people |
FR3051952A1 (en) * | 2016-05-24 | 2017-12-01 | Thales Sa | METHOD AND DEVICE FOR LOCATING OBJECTS OR MOBILE PERSONS |
WO2017215024A1 (en) * | 2016-06-16 | 2017-12-21 | 东南大学 | Pedestrian navigation device and method based on novel multi-sensor fusion technology |
EP3812706A4 (en) * | 2018-08-08 | 2021-07-07 | Huawei Technologies Co., Ltd. | Method for obtaining movement track of user and terminal |
US12085390B2 (en) | 2018-08-08 | 2024-09-10 | Huawei Technologies Co., Ltd. | Method for obtaining movement track of user and terminal |
WO2020188275A2 (en) | 2019-03-20 | 2020-09-24 | Navenio Ltd. | Transition detection |
WO2022018399A1 (en) | 2020-07-20 | 2022-01-27 | Navenio Ltd. | Map matching trajectories |
DE102022101492A1 (en) | 2022-01-24 | 2023-07-27 | Bayerische Motoren Werke Aktiengesellschaft | Method and device for determining pose data relating to the pose of a vehicle |
Also Published As
Publication number | Publication date |
---|---|
EP3194889B1 (en) | 2021-04-14 |
US10502576B2 (en) | 2019-12-10 |
GB201500411D0 (en) | 2015-02-25 |
AU2015316575B2 (en) | 2020-11-05 |
WO2016042296A3 (en) | 2016-06-30 |
CN110501011B (en) | 2023-12-05 |
AU2015316575A1 (en) | 2017-03-30 |
CN107250727A (en) | 2017-10-13 |
EP3194889A2 (en) | 2017-07-26 |
US20170241787A1 (en) | 2017-08-24 |
CN110501011A (en) | 2019-11-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
AU2015316575B2 (en) | Inertial tracking based determination of the position of a mobile device carried by a user in a geographical area | |
Xiao et al. | Robust indoor positioning with lifelong learning | |
Tian et al. | A multi-mode dead reckoning system for pedestrian tracking using smartphones | |
Xiao et al. | Robust pedestrian dead reckoning (R-PDR) for arbitrary mobile device placement | |
Huang et al. | Exploiting cyclic features of walking for pedestrian dead reckoning with unconstrained smartphones | |
TWI457539B (en) | Multi-posture step length calibration system and method for indoor positioning | |
Kang et al. | Improved heading estimation for smartphone-based indoor positioning systems | |
Zhang et al. | Indoor localization using a smart phone | |
US10508920B2 (en) | Pedestrian dead reckoning position tracker | |
US9292936B2 (en) | Method and apparatus for determining location | |
Zhang et al. | SmartMTra: Robust indoor trajectory tracing using smartphones | |
Jiang et al. | Ptrack: Enhancing the applicability of pedestrian tracking with wearables | |
US10302434B2 (en) | Method and apparatus for determining walking direction for a pedestrian dead reckoning process | |
JP5750742B2 (en) | Mobile object state estimation device | |
Park et al. | Accelerometer-based smartphone step detection using machine learning technique | |
Park et al. | Robust pedestrian dead reckoning for multiple poses in smartphones | |
Marron et al. | Multi sensor system for pedestrian tracking and activity recognition in indoor environments | |
Wu et al. | Indoor positioning system based on inertial MEMS sensors: Design and realization | |
Gobana | Survey of Inertial/magnetic Sensors Based pedestrian dead reckoning by multi-sensor fusion method | |
Yan et al. | FlexPDR: Fully flexible pedestrian dead reckoning using online multimode recognition and time-series decomposition | |
Abadleh et al. | Noise segmentation for step detection and distance estimation using smartphone sensor data | |
Qian et al. | RPNOS: Reliable pedestrian navigation on a smartphone | |
Gowda et al. | UMOISP: Usage mode and orientation invariant smartphone pedometer | |
Hoseinitabatabaei et al. | Towards a position and orientation independent approach for pervasive observation of user direction with mobile phones | |
US9069003B2 (en) | Methods, apparatuses and computer program products for determining speed of movement of a device and device pose classification |
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: 15766567 Country of ref document: EP Kind code of ref document: A2 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 15509771 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
REEP | Request for entry into the european phase |
Ref document number: 2015766567 Country of ref document: EP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2015766567 Country of ref document: EP |
|
ENP | Entry into the national phase |
Ref document number: 2015316575 Country of ref document: AU Date of ref document: 20150910 Kind code of ref document: A |