WO2021012879A1 - 轨迹重建方法、装置、计算机设备和存储介质 - Google Patents

轨迹重建方法、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2021012879A1
WO2021012879A1 PCT/CN2020/098555 CN2020098555W WO2021012879A1 WO 2021012879 A1 WO2021012879 A1 WO 2021012879A1 CN 2020098555 W CN2020098555 W CN 2020098555W WO 2021012879 A1 WO2021012879 A1 WO 2021012879A1
Authority
WO
WIPO (PCT)
Prior art keywords
model
time point
model parameter
positioning
positioning coordinates
Prior art date
Application number
PCT/CN2020/098555
Other languages
English (en)
French (fr)
Inventor
尹峰
谢昂
崔曙光
艾渤
Original Assignee
香港中文大学(深圳)
深圳市大数据研究院
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 香港中文大学(深圳), 深圳市大数据研究院 filed Critical 香港中文大学(深圳)
Publication of WO2021012879A1 publication Critical patent/WO2021012879A1/zh
Priority to ZA2022/01092A priority Critical patent/ZA202201092B/en

Links

Images

Classifications

    • GPHYSICS
    • G01MEASURING; TESTING
    • G01SRADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S11/00Systems for determining distance or velocity not using reflection or reradiation
    • G01S11/02Systems for determining distance or velocity not using reflection or reradiation using radio waves
    • G01S11/06Systems for determining distance or velocity not using reflection or reradiation using radio waves using intensity measurements
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04WWIRELESS COMMUNICATION NETWORKS
    • H04W4/00Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/02Services making use of location information
    • H04W4/025Services making use of location information using location based information parameters
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04WWIRELESS COMMUNICATION NETWORKS
    • H04W4/00Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/02Services making use of location information
    • H04W4/029Location-based management or tracking services

Definitions

  • This application relates to the field of data processing technology, in particular to a method, device, computer equipment and storage medium for trajectory reconstruction.
  • GPS Global Positioning System
  • Traditional indoor navigation technology is to collect single data information for indoor positioning and trajectory reconstruction. For example, traditional Bluetooth indoor positioning, WiFi positioning, ZigBee positioning technology, etc. all use a single measurement data. If the indoor wireless communication environment is more complex and correct When a single measurement data has large interference, it will directly cause a large deviation in the positioning result, and then affect the reconstructed movement trajectory.
  • an embodiment of the present invention provides a trajectory reconstruction method, the method includes:
  • an embodiment of the present invention provides a trajectory reconstruction device, the device including:
  • the collection module is used to collect the inertial sensor data of the terminal equipment at multiple time points and the signal strength data of the wireless fidelity network;
  • the wireless fidelity network positioning module is configured to generate the initial positioning coordinates corresponding to each time point according to the signal strength data of the wireless fidelity network at each time point;
  • a displacement vector determining module configured to generate a displacement vector corresponding to each time point according to the inertial sensor data at each time point;
  • the model positioning module is used to input the initial positioning coordinates and displacement vectors corresponding to each said time point into the global positioning model to obtain the final positioning coordinates of each said time point;
  • the trajectory reconstruction module is used for trajectory reconstruction according to the final positioning coordinates.
  • an embodiment of the present invention provides a computer device, including a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:
  • an embodiment of the present invention provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:
  • the above-mentioned trajectory reconstruction method, device, computer equipment and storage medium collect the inertial sensor data of the terminal equipment at multiple time points and the signal strength data of the Wi-Fi network; according to the signal strength of the Wi-Fi network at each time point Data, generating the initial positioning coordinates corresponding to each of the time points; generating the displacement vector corresponding to each of the time points according to the inertial sensor data at each time point; and converting the initial positioning coordinates and displacement vectors corresponding to each of the time points
  • the global positioning model is input to obtain the final positioning coordinates of each time point; the trajectory reconstruction is performed according to the final positioning coordinates.
  • the displacement vector and initial positioning coordinates corresponding to each time point are generated, and the global positioning model is used for each time
  • Modeling the initial positioning coordinates and displacement vectors corresponding to the points and the final positioning coordinates of each time point can obtain more accurate position estimates in complex object motion patterns and in complex wireless propagation environments, and improve the accuracy of trajectory reconstruction.
  • FIG. 1 is an implementation environment diagram of a trajectory reconstruction method provided by an embodiment of the application
  • FIG. 3 is a flowchart of another trajectory reconstruction method provided by an embodiment of the application.
  • FIG. 4 is a flowchart of another trajectory reconstruction method provided by an embodiment of the application.
  • FIG. 5 is a flowchart of another trajectory reconstruction method provided by an embodiment of the application.
  • FIG. 6 is a flowchart of another trajectory reconstruction method provided by an embodiment of the application.
  • FIG. 7 is a flowchart of another trajectory reconstruction method provided by an embodiment of the application.
  • FIG. 8 is a flowchart of another trajectory reconstruction method provided by an embodiment of the application.
  • FIG. 9 is a flowchart of another trajectory reconstruction method provided by an embodiment of the application.
  • FIG. 10 is a block diagram of a trajectory reconstruction device provided by an embodiment of this application.
  • FIG. 11 is a block diagram of a computer device provided by an embodiment of this application.
  • the trajectory reconstruction method provided in this application can be applied to the application environment as shown in FIG. 1.
  • the terminal 102 communicates with the server 104 through the network through the network.
  • the terminal 102 and the server 104 communicate in a wireless manner (4G/5G or WiFi) based on the MQTT protocol.
  • the data received by the server 104 is first forwarded through the Mosquitto proxy software running the MQTT protocol, and further received and processed by the JavaScript script running on the NodeJS platform, and the result is returned to the terminal 102.
  • the terminal 102 can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, and portable wearable devices.
  • the terminal 102 has built-in accelerometers, gyroscopes, magnetometers, barometers, etc. Inertial sensors and wireless access modules, the server 104 can be implemented as an independent server or a server cluster composed of multiple servers.
  • FIG. 2 shows a trajectory reconstruction method provided in this embodiment.
  • the method is applied to the terminal in FIG. 1 as an example for description, and includes the following steps:
  • Step 202 Collect inertial sensor data of the terminal device at multiple time points and signal strength data of the wireless fidelity network.
  • inertial sensors such as accelerometers, gyroscopes, magnetometers, and barometers are built in the terminal device.
  • the terminal device has a built-in linear acceleration sensor, which is an acceleration sensor.
  • the terminal device also has a built-in rotation vector sensor.
  • the rotation vector sensor is a composite sensor based on accelerometer, gyroscope, and magnetometer. Specifically, the inertial sensor data of the terminal device at multiple time points is acquired through the foregoing inertial sensor.
  • a WiFi module is built in the terminal device. Specifically, through the WiFi module, the signal strength data of the Wi-Fi network at multiple time points of the terminal device can be obtained, wherein the signal strength data of the Wi-Fi network at each time point includes the current time point, the receiving The signal strength data of the Wi-Fi network of multiple wireless access points.
  • Step 204 Generate initial positioning coordinates corresponding to each time point according to the signal strength data of the Wi-Fi network at each time point.
  • the distance between the terminal device and each wireless access point at the current time point is calculated, and according to each wireless access point
  • the actual position of the terminal device at the current time point can be obtained through the geometric algorithm.
  • Step 206 Generate a displacement vector corresponding to each time point based on the inertial sensor data at each time point.
  • the direction of movement of the terminal device corresponding to each time point is calculated, combined with the preset unit step length, Obtain the displacement vector corresponding to each time point, and the displacement vector includes the step length and the moving direction from the current time point to the next time point.
  • Step 208 Input the initial positioning coordinates and displacement vectors corresponding to each time point into the global positioning model to obtain the final positioning coordinates of each time point.
  • the global positioning model represents the mapping relationship between input parameters (initial positioning coordinates and displacement vectors corresponding to each time point) and output parameters (final positioning coordinates of each time point).
  • the global positioning model may be a Gaussian process state space model, a variational Gaussian process state space model, a convolutional neural network (Convolutional Neural Networks, CNN) model, and a deep belief network (Deep Belief Networks, DBF) model, one of Restricted Boltzmann Machine (RBM) model and AutoEncoder.
  • a Gaussian process state space model a Gaussian process state space model
  • a variational Gaussian process state space model a convolutional neural network (Convolutional Neural Networks, CNN) model
  • a deep belief network Deep Belief Networks, DBF
  • RBM Restricted Boltzmann Machine
  • Step 210 Perform trajectory reconstruction according to each final positioning coordinate.
  • each final positioning coordinate reflects the positioning coordinates of the terminal device at each time, and the movement trajectory of the terminal device can be reconstructed by connecting the final positioning coordinates in a chronological order.
  • the inertial sensor data of the terminal device at multiple time points and the signal strength data of the Wi-Fi network are collected; according to the signal strength data of the Wi-Fi network at each time point , Generate the initial positioning coordinates corresponding to each of the time points; generate the displacement vector corresponding to each of the time points according to the inertial sensor data of each time point; input the initial positioning coordinates and the displacement vector corresponding to each of the time points
  • the global positioning model obtains the final positioning coordinates of each time point; and the trajectory reconstruction is performed according to the final positioning coordinates.
  • the displacement vector and initial positioning coordinates corresponding to each time point are generated, and the global positioning model is used for each time
  • Modeling the initial positioning coordinates and displacement vectors corresponding to the points and the final positioning coordinates of each time point can obtain more accurate position estimates in complex object motion patterns and in complex wireless propagation environments, and improve the accuracy of trajectory reconstruction.
  • the global positioning model can be determined according to multiple local positioning models established by the multiple mobile terminals. Therefore, please refer to FIG. 3, which shows a flowchart of another trajectory reconstruction method provided by this embodiment, and the trajectory reconstruction method can be applied to the terminal 102 in the implementation environment described above. Based on the embodiment shown in FIG. 2, the above step 208 may specifically include the following steps:
  • Step 302 Training the initial positioning model according to the initial positioning coordinates and displacement vectors corresponding to each time point to obtain a local positioning model; the local positioning model includes multiple model parameters.
  • multiple model parameters of the local positioning model can be obtained, and the multiple model parameters can be directly determined Local positioning model of terminal equipment.
  • the positioning model is determined by a mean function and a covariance function. Among them, some of the model parameters are used to represent the model parameters of the covariance function, and some of the model parameters are used to represent the covariance matrix of noise.
  • Step 304 Upload multiple model parameters to the central node, so that the central node performs an equalization operation on each model parameter of the multiple local positioning models, and obtain multiple equalized model parameters.
  • the central node is the server 104 in the implementation environment of FIG. 1.
  • the central node also communicates with other terminal devices in the current scene.
  • multiple model parameters contained in the model will be uploaded to the central node, and the central node will also receive and save multiple model parameters uploaded by other terminal devices, in order to improve the positioning accuracy of the global positioning model
  • the central node will balance and coordinate multiple model parameters of the local positioning model uploaded by each terminal device, and obtain the balanced multiple model parameters of the global positioning model after balance and coordination.
  • the central node every time a terminal device connected to the central node uploads multiple model parameters of the local positioning model, the central node will update multiple model parameters of the global positioning model and update the updated The multiple model parameters of the global positioning model are sent to all terminal devices connected to the central node.
  • Step 306 Receive multiple equalized model parameters returned by the central node.
  • Step 308 Establish a global positioning model according to the multiple model parameters after equalization.
  • the number of the received balanced multiple model parameters is the same as the multiple model parameters uploaded to the central node, and the only one global positioning model can be determined through the multiple balanced model parameters.
  • Step 310 Input the initial positioning coordinates and displacement vectors corresponding to each time point into the global positioning model to obtain the final positioning coordinates of each time point.
  • the terminal device after receiving the multiple model parameters after the equalization, can directly use the multiple model parameters to re-establish a global positioning model, and perform input on each time point according to the global positioning model.
  • the corresponding initial positioning coordinates and displacement vectors are calculated to obtain the final positioning coordinates of each time point.
  • the initial positioning model is trained according to the initial positioning coordinates and displacement vectors corresponding to each time point to obtain a local positioning model;
  • the local positioning model includes multiple model parameters;
  • the parameters are uploaded to the central node so that the central node can perform equalization operations on the various model parameters of multiple local positioning models, and obtain multiple equalized model parameters; receive the equalized multiple model parameters returned by the central node;
  • a global positioning model is established with multiple model parameters of, and the initial positioning coordinates and displacement vectors corresponding to each time point are input into the global positioning model to obtain the final positioning coordinates of each time point.
  • the multiple model parameters of the local positioning model obtained by uploading are used to enable the central node to perform equalization operations according to the various model parameters uploaded by each terminal device, and obtain the balanced global positioning model.
  • Model parameters Through the above-mentioned distributed model building process, the scale of the model is enlarged, and the accuracy of positioning based on the global positioning model is further improved. Moreover, since only the model parameters of the local model are transmitted, the calculations related to model training and trajectory reconstruction sink to the mobile phone. , To ensure the privacy of user mobile phone data.
  • the local positioning model is a local variational Gaussian process state space model
  • the multiple model parameters include a first model parameter, a second model parameter, a third model parameter, and a first model parameter.
  • the above step 302 may specifically include the following steps:
  • Step 402 Use the Gaussian process to model the mapping relationship between the initial positioning coordinates and the real position coordinates corresponding to each time point to obtain an observation quantum model.
  • the observation quantum model includes a first model parameter and a second model parameter.
  • the wireless fidelity network positioning technology based on the strength of the WiFi received signal can only provide a rough position estimation, and the accuracy of the position estimation depends on the specific indoor location of pedestrians and the distribution of wireless access points. And the indoor complex radio wave propagation environment.
  • the present invention uses the Gaussian process to model the mapping relationship between the initial positioning coordinates and the real position coordinates provided by the WiFi positioning technology.
  • the Gaussian process describing the mapping relationship between the initial positioning coordinates and the real position coordinates can be defined as:
  • g(x) is a Gaussian process with the first model parameter ⁇ g
  • r is a Gaussian distribution
  • the observation noise of, R is the covariance matrix of the observation noise
  • only one observation quantum model can be determined through the first model parameter and the second model parameter.
  • Step 404 Model the mapping relationship between the displacement vector corresponding to each time point and the real position coordinate by using a Gaussian process to obtain a state evolution sub-model, which includes a third model parameter and a fourth model parameter.
  • the state evolution sub-model is also represented by a Gaussian process, and its mean function can be described by pedestrian dead reckoning technology, namely:
  • x t represents the real position coordinates at time t
  • u t represents the displacement vector at time t.
  • only one state evolution sub-model can be determined through the third model parameter and the fourth model parameter.
  • Step 406 Obtain a local variational Gaussian process state space model according to the observation quantum model and the state evolution sub-model.
  • the state space model of the local variational Gaussian process can be jointly represented by the observation quantum model and the state evolution sub-model.
  • the local variational Gaussian process state space model further includes the distribution probability of the initial real position coordinates, expressed as:
  • the state space model of the local variational Gaussian process is divided into the state evolution sub-model and the observation quantum model, and the state evolution sub-model and the observation quantum model are respectively learned in order, which simplifies the local variation
  • the learning complexity of the Gaussian process state space model further improves the efficiency of trajectory reconstruction.
  • the embodiment of the present application also provides another trajectory reconstruction method, which can be applied to the terminal 102 in the implementation environment described above.
  • the step 402 may specifically include the following steps:
  • the embodiment of the present application also provides another trajectory reconstruction method, which can be applied to the terminal 102 in the implementation environment described above.
  • the observation quantum model includes:
  • m g (x) represents the mean function of Gaussian process g(x)
  • k g (x,x′) represents the covariance function of Gaussian process g(x)
  • k g (x,x′) is defined by
  • the first model parameter ⁇ g is determined
  • g t represents the mean value of the probability distribution y t
  • R represents the covariance matrix of the probability distribution y t
  • R is the second model parameter.
  • FIG. 5 shows a flowchart of another trajectory reconstruction method provided by this embodiment, and the trajectory reconstruction method can be applied to the terminal 102 in the implementation environment described above.
  • the above step 402 may specifically include the following steps:
  • Step 502 according to the training data set ⁇ x 1:N , y 1:N ⁇ , generate a corresponding observational logarithmic marginal likelihood function.
  • the logarithmic marginal likelihood function of the observation is expressed as:
  • I N is an identity matrix
  • the training data set ⁇ x 1:N , y 1:N ⁇ is used to describe the mapping relationship between the initial positioning coordinates and the real position corresponding to each time point.
  • Step 504 Obtain the first model parameter and the second model parameter by maximizing the observed logarithmic marginal likelihood function.
  • the conjugate gradient method (CG) and the LBFGS algorithm can be used to maximize the logarithmic marginal likelihood function of the observation, and obtain the first model parameter ⁇ g and the second model parameter ⁇ g. Model parameter R.
  • Step 506 Determine the observation quantum model according to the first model parameter and the second model parameter.
  • the only observation quantum model can be determined by the acquired first model parameter ⁇ g and the second model parameter R.
  • the obtained observational measurement logarithmic marginal likelihood function can be maximized to obtain the first model parameter and the second model parameter that can determine the observation quantum model, and then the observation quantum model. It can reduce the training workload of the local variational Gaussian process state space model and improve the calculation efficiency.
  • the embodiment of the present application also provides another trajectory reconstruction method, which can be applied to the terminal 102 in the implementation environment described above.
  • the state evolution sub-model includes:
  • m f (x, u) represents the mean function of the Gaussian process f(x, u)
  • k f ((x, u), (x′,u′)) represents the Gaussian process f(x, u)
  • the covariance function, k f ((x,u),(x′,u′)) is determined by the third model parameter ⁇ f
  • f t represents the mean value of the probability distribution x t
  • f t is the covariance matrix
  • Q is the fourth model parameter.
  • FIG. 6 shows a flowchart of another trajectory reconstruction method provided by this embodiment.
  • the trajectory reconstruction method can be applied to the terminal 102 in the implementation environment described above.
  • the above step 404 may specifically include the following steps:
  • Step 602 Obtain M induction points ⁇ z 1:M ,v 1:M ⁇ .
  • the present invention adopts the variational sparse Gaussian process framework, by introducing M auxiliary induction points v 1:M to reflect the information contained in the observation. These induction points are the values of the corresponding Gaussian process on the induction input z 1:M , and meet the joint Gaussian distribution with the unknown function value f 1:T .
  • Step 604 Obtain the lower bound of evidence for the state space model of the local variational Gaussian process based on M induction points ⁇ z 1:M ,v 1:M ⁇ .
  • the lower bound of evidence is similar to the posterior q(x 0:T ), q(v 1:M ), the third model parameter ⁇ f , the fourth model parameter Q, and the induced input z 1:M related functional.
  • the evidence lower bound of the initial local variational Gaussian process state space model can be obtained, and the evidence lower bound is expressed as:
  • q(x 0:T ,f 1:T ,v 1:M ) is a variational approximation to the posterior distribution p(x 0:T ,f 1:T ,v 1:M
  • the joint distribution of all variables in the Gaussian process state space model can be written as:
  • the lower bound of evidence for the Gaussian process state space model can be written as a generalization of the posterior approximation q(x 0:T ), q(v 1:M ), model parameters ⁇ f , Q, and induced input z 1:M letter:
  • KL(. ⁇ .) means KL divergence
  • Step 606 fix ⁇ f ,Q,z 1:M ⁇ , and maximize the lower bound of evidence to obtain q(x 0:T ) and q(v 1:M ).
  • the optimal q(x 0:T ) and q(v 1:M ) can be maximized by using the variational method
  • the lower bound of the above evidence is drawn. Therefore, the optimal q(v 1:M ) obtained is Gaussian distribution And its characteristic parameter (natural parameter) is:
  • the optimal q * (x 0:T ) can be obtained through the above formula, and the optimal q * (x 0:T ) is used as the posterior approximation q(x 0:T ).
  • Step 608 fix q(x 0:T ) and q(v 1:M ), and maximize the lower bound of evidence to obtain the third model parameter ⁇ f and the fourth model parameter Q.
  • the model parameters ⁇ f and Q of the model and the induced input z 1:M can be obtained by maximizing the lower bound of evidence.
  • the present invention adopts a gradient-based optimization scheme.
  • the gradient of can be calculated by the following formula:
  • Step 610 Determine the state evolution sub-model according to the third model parameter and the fourth model parameter.
  • only one state evolution sub-model can be determined through the acquired third model parameter ⁇ f and fourth model parameter Q.
  • the lower bound of evidence of the obtained Gaussian process state space model can be maximized, and the parameters ⁇ f ,Q,z 1:M ⁇ and q(x 0:T) can be fixed respectively . ), q(v 1:M ) obtains the third model parameter and the fourth model parameter that can determine the state evolution sub-model, and then obtains the state evolution sub-model. It can reduce the training workload of the local variational Gaussian process state space model and improve the calculation efficiency.
  • FIG. 7 shows a flowchart of another trajectory reconstruction method provided by this embodiment.
  • the trajectory reconstruction method can be applied to the terminal 102 in the implementation environment described above.
  • the above step 204 may specifically include the following steps:
  • Step 702 Establish a corresponding positioning log-likelihood function according to the signal strength data of the Wi-Fi network at each time point.
  • the signal strength data of the Wi-Fi network at each time point includes the signal strength of the Wi-Fi network of multiple wireless access points received by the terminal device at the current time point.
  • the mapping relationship between the signal strength data of the wireless fidelity network and the scanning distance can be expressed by a linear logarithmic model:
  • R & lt signal intensity i Wi-Fi network is obtained by collecting a plurality of times i th wireless access points, A and B are signal path loss parameter, D i is the reference distance, d 0 is the distance from the scanning, the The scanning position indicates the distance between the wireless access point and the position of the terminal device.
  • the noise distribution is assumed to be independent and identically distributed Gaussian noise, with a mean of 0 and a variance of ⁇ 2 .
  • the unknown parameters A, B, and ⁇ of the linear logarithmic model of each wireless access point can be estimated by linear least square method.
  • the present invention obtains the initial positioning coordinates based on maximum likelihood estimation (MLE). Assuming that a scan at any location can get the signal strength of the Wi-Fi network of M AP wireless access points, where the signal strength of the Wi-Fi network of each wireless access point belongs to an independent wireless access point .
  • MLE maximum likelihood estimation
  • erf(.) represents the standard Gaussian error function
  • P dec is the cutoff limit of the signal strength of the wireless fidelity network
  • step 704 the initial positioning coordinates corresponding to each time point are obtained by maximizing the positioning log likelihood function.
  • the initial positioning coordinates corresponding to each time point can be obtained by maximizing the log likelihood function, namely:
  • y represents the initial positioning coordinates.
  • the linear logarithmic model of each wireless access point and the signal strength data of the wireless fidelity network at each time point can be used to establish a positioning log likelihood function, and by maximizing the positioning The log likelihood function obtains the initial positioning coordinates of each time point. It is possible to improve the accuracy of the initial positioning coordinates describing the true position coordinates, thereby ensuring the accuracy of the subsequent final position coordinates, and improving the accuracy of the reconstructed trajectory.
  • FIG. 8 shows a flowchart of another trajectory reconstruction method provided by this embodiment.
  • the trajectory reconstruction method can be applied to the terminal 102 in the implementation environment described above.
  • the displacement vector is used to indicate the moving direction and step length of the terminal device at the time point.
  • the above step 206 may specifically include the following steps:
  • Step 802 Obtain the rotation vector signal of the terminal device through the rotation vector sensor.
  • Step 804 Obtain the movement direction at each time point according to the rotation vector signal, where the movement direction is the deflection angle between the orientation of the terminal device and the geomagnetic north pole.
  • step 806 the displacement vector at each time point is determined by the preset step length and the movement direction at each time point.
  • the terminal device is kept flat, and the front end of the terminal device faces the walking direction.
  • the rotation vector sensor is a composite sensor based on accelerometer, gyroscope, and magnetometer.
  • the yuan is processed, and the declination angle between the terminal device's orientation and the geomagnetic north pole at each time point can be obtained, and this can be used as the direction of movement.
  • the movement direction at each time point can be obtained to determine the displacement vector at each time point.
  • the displacement vector at each time point obtained by this embodiment can more accurately reflect the movement direction and movement distance from each time point to the next time point, thereby ensuring the accuracy of subsequent final position coordinates and improving the reconstruction of the trajectory. Accuracy.
  • FIG. 9 shows a flowchart of another trajectory reconstruction method provided by this embodiment, and the trajectory reconstruction method can be applied to the terminal 102 in the implementation environment described above. Based on the above embodiment shown in FIG. 2, the following steps may also be included:
  • Step 902 Obtain a linear acceleration signal of the terminal device relative to the direction of gravity through a linear acceleration sensor.
  • Step 904 Perform moving average processing on the linear acceleration signal.
  • Step 906 Perform local peak detection on the linear acceleration signal after moving average processing to obtain multiple time points, each time point being the time when the linear acceleration signal is at the local peak.
  • the terminal device is kept flat, and the front end of the terminal device faces the walking direction.
  • the linear acceleration sensor is a composite sensor based on accelerometer and gyroscope.
  • the terminal device output by the linear acceleration sensor is relatively The linear acceleration signal in the direction of gravity, the linear acceleration signal is subjected to sliding average processing, and the local peak detection of the linear acceleration signal after the sliding average processing is performed to obtain multiple time points, each time point is the time when the linear acceleration signal is at the local peak , The multiple time points are multiple time when the footsteps fall.
  • the multiple time points obtained by this embodiment can more accurately reflect the multiple moments of the footsteps falling, thereby ensuring the accuracy of the subsequent final position coordinates and improving the accuracy of reconstructing the trajectory.
  • the trajectory reconstruction device 1000 may include: an acquisition module 1001, a wireless fidelity network positioning module 1002, a displacement vector determination module 1003, a model positioning module 1004, and a trajectory reconstruction module 1005, wherein:
  • the collection module 1001 is used to collect the inertial sensor data of the terminal equipment at multiple time points and the signal strength data of the wireless fidelity network.
  • the Wi-Fi network positioning module 1002 is configured to generate initial positioning coordinates corresponding to each time point according to the signal strength data of the Wi-Fi network at each time point.
  • the displacement vector determining module 1003 is configured to generate a displacement vector corresponding to each time point according to the inertial sensor data at each time point.
  • the model positioning module 1004 is used to input the initial positioning coordinates and displacement vectors corresponding to each time point into the global positioning model to obtain the final positioning coordinates of each time point.
  • the trajectory reconstruction module 1005 is used for trajectory reconstruction according to the final positioning coordinates.
  • the model positioning module 1004 is specifically configured to: train the initial positioning model according to the initial positioning coordinates and displacement vectors corresponding to each time point to obtain a local positioning model;
  • the model includes a plurality of model parameters; uploading the plurality of model parameters to a central node, so that the central node performs an equalization operation on each model parameter of the plurality of local positioning models, and obtains a plurality of equalized model parameters ;
  • the global positioning model obtains the final positioning coordinates of each time point.
  • the local positioning model is a local variational Gaussian process state space model
  • the multiple model parameters include a first model parameter, a second model parameter, a third model parameter, and a fourth model parameter
  • the model positioning module 1004 is also specifically configured to: use a Gaussian process to model the mapping relationship between the initial positioning coordinates corresponding to each time point and the real position coordinates to obtain an observation quantum model, the observation quantum model including The first model parameter and the second model parameter; using a Gaussian process to model the mapping relationship between the displacement vector corresponding to each time point and the real position coordinate to obtain a state evolution sub-model, the state evolution sub-model
  • the model includes the third model parameter and the fourth model parameter; the local variational Gaussian process state space model is obtained according to the observation quantum model and the state evolution sub-model.
  • the multiple model parameters include a first model parameter, a second model parameter, a third model parameter, and a fourth model parameter.
  • the model positioning module 1004 is further specifically configured to: The first model parameter, the second model parameter, the third model parameter, and the fourth model parameter are uploaded to the central node, so that the central node can perform the state-space model of a plurality of the local variational Gaussian processes Perform an equalization operation on each model parameter of, and obtain the equalized first model parameter, the second model parameter, the third model parameter, and the fourth model parameter.
  • the observation quantum model includes:
  • m g (x) represents the mean function of Gaussian process g(x)
  • k g (x,x′) represents the covariance function of Gaussian process g(x)
  • k g (x,x′) is defined by
  • the first model parameter ⁇ g is determined
  • g t represents the mean value of the probability distribution y t
  • R represents the covariance matrix of the probability distribution y t
  • R is the second model parameter.
  • the model positioning module 1004 is further specifically configured to: generate a corresponding observation logarithmic marginal likelihood function according to the training data set ⁇ x 1:N ,y 1:N ⁇ :
  • I N is the identity matrix
  • the training data set ⁇ x 1:N , y 1:N ⁇ is used to describe the mapping relationship between the initial positioning coordinates and the real position corresponding to each time point
  • the observed logarithmic marginal likelihood function is used to obtain the first model parameter and the second model parameter
  • the observation quantum model is determined according to the first model parameter and the second model parameter.
  • the state evolution sub-model includes:
  • m f (x, u) represents the mean function of the Gaussian process f(x, u)
  • k f ((x, u), (x′,u′)) represents the Gaussian process f(x, u)
  • the covariance function, k f ((x,u),(x′,u′)) is determined by the third model parameter ⁇ f
  • f t represents the mean value of the probability distribution x t
  • f t is the covariance matrix
  • Q is the fourth model parameter.
  • the model positioning module 1004 is further specifically configured to: obtain M induction points ⁇ z 1:M , v 1:M ⁇ ; based on the M induction points ⁇ z 1:M ,v 1:M ⁇ Obtain the lower bound of evidence of the state space model of the local variational Gaussian process, and the lower bound of evidence is approximate to the posterior q(x 0:T ), q(v 1:M ), the third Model parameter ⁇ f , the fourth model parameter Q, and induced input z 1: M- related functional; fix ⁇ f ,Q,z 1:M ⁇ , maximize the lower bound of the evidence to obtain q (x 0:T ) and q(v 1:M ); fix q(x 0:T ) and q(v 1:M ), maximize the lower bound of evidence, and obtain the third model parameter ⁇ f , the fourth model parameter Q; the state evolution sub-model is determined according to the third model parameter and the fourth model parameter.
  • the Wi-Fi network positioning module 1002 is specifically configured to: establish a corresponding positioning log likelihood function according to the signal strength data of the Wi-Fi network at each time point; Maximize the positioning log likelihood function to obtain the initial positioning coordinates corresponding to each time point.
  • the displacement vector determining module 1003 is specifically configured to: obtain a rotation vector signal of the terminal device through a rotation vector sensor; obtain the direction of motion at each time point according to the rotation vector signal ,
  • the movement direction is the deflection angle between the orientation of the terminal device and the geomagnetic north pole; the displacement vector at each time point is determined by a preset step length and the movement direction of each time point.
  • the displacement vector determining module 1003 is further specifically configured to: obtain a linear acceleration signal of the terminal device relative to the direction of gravity through a linear acceleration sensor; perform sliding average processing on the linear acceleration signal; Performing local peak detection on the linear acceleration signal after the moving average processing to obtain the multiple time points, each of the time points being the time when the linear acceleration signal is at the local peak.
  • Each module in the above-mentioned trajectory reconstruction device can be implemented in whole or in part by software, hardware, and a combination thereof.
  • the foregoing modules may be embedded in the form of hardware or independent of the processor in the computer device, or may be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the foregoing modules.
  • a computer device is provided.
  • the computer device may be a terminal, and its internal structure diagram may be as shown in FIG. 11.
  • the computer equipment includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus.
  • the processor of the computer device is used to provide calculation and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium and an internal memory.
  • the non-volatile storage medium stores an operating system and a computer program.
  • the internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium.
  • the network interface of the computer device is used to communicate with an external terminal through a network connection.
  • the computer program is executed by the processor to realize a trajectory reconstruction method.
  • the display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen
  • the input device of the computer equipment can be a touch layer covered on the display screen, or it can be a button, a trackball or a touchpad set on the housing of the computer equipment , It can also be an external keyboard, touchpad, or mouse.
  • FIG. 11 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied.
  • the specific computer device may Including more or fewer parts than shown in the figure, or combining some parts, or having a different arrangement of parts.
  • a computer device including a memory and a processor, and a computer program is stored in the memory, and the processor implements the following steps when executing the computer program:
  • a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:
  • Non-volatile memory may include read only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Remote Sensing (AREA)
  • Position Fixing By Use Of Radio Waves (AREA)
  • Navigation (AREA)

Abstract

本申请涉及一种轨迹重建方法、装置、计算机设备和存储介质。所述方法包括:采集终端设备在多个时间点的惯性传感器数据及无线保真网络的信号强度数据;根据各时间点的无线保真网络的信号强度数据,生成各时间点的初始定位坐标;根据各时间点的惯性传感器数据,生成各时间点的位移向量;将各时间点对应的初始定位坐标及位移向量输入全局定位模型,得到各时间点的最终定位坐标;根据各最终定位坐标进行轨迹重建。采用本方法能够在复杂的物体运动模式以及无线传播环境下得到更加准确的位置估计,提升了轨迹重建的精度。

Description

轨迹重建方法、装置、计算机设备和存储介质 技术领域
本申请涉及数据处理技术领域,特别是涉及一种轨迹重建方法、装置、计算机设备和存储介质。
背景技术
随着无线通信网络技术的快速发展,移动智能终端正逐渐深入生活的各个层面,导航与移动智能终端也在进行紧密的结合,全球定位系统(Global Positioning System,GPS)卫星定位是获取位置信息的最常用方式,但是由于卫星信号容易受到各种建筑物遮挡或其它因素干扰,GPS定位技术并不适用于室内场合或高楼林立的复杂环境。随着无线电子移动通信技术的发展,室内定位技术已成为科研工作者日益重视的热点研究领域。
传统的室内导航技术都是通过收集单一的数据信息进行室内定位及轨迹重建,例如传统的蓝牙室内定位、WiFi定位、ZigBee定位技术等均采用单一的测量数据,若室内无线传播环境较为复杂且对单一测量数据有较大干扰时,会直接导致定位结果出现较大偏差,进而对重建的移动轨迹造成影响。
因此,传统的室内导航技术由于难以有效地表示复杂的无线传播环境下搜集到的观测数据,无法取得更高定位精度,进而无法还原较为准确的室内运动轨迹。
发明内容
基于此,有必要针对上述技术问题,提供一种定位精度高且轨迹重建准确率高的轨迹重建方法、装置、计算机设备和存储介质。
第一方面,本发明实施例提供一种轨迹重建方法,所述方法包括:
采集终端设备在多个时间点的惯性传感器数据及无线保真网络的信号强度数据;
根据各所述时间点的无线保真网络的信号强度数据,生成各所述时间点对 应的初始定位坐标;
根据各所述时间点的惯性传感器数据,生成各所述时间点对应的位移向量;
将各所述时间点对应的初始定位坐标及位移向量输入全局定位模型,得到各所述时间点的最终定位坐标;
根据各最终定位坐标进行轨迹重建。
第二方面,本发明实施例提供一种轨迹重建装置,所述装置包括:
采集模块,用于采集终端设备在多个时间点的惯性传感器数据及无线保真网络的信号强度数据;
无线保真网络定位模块,用于根据各所述时间点的无线保真网络的信号强度数据,生成各所述时间点对应的初始定位坐标;
位移向量确定模块,用于根据各所述时间点的惯性传感器数据,生成各所述时间点对应的位移向量;
模型定位模块,用于将各所述时间点对应的初始定位坐标及位移向量输入全局定位模型,得到各所述时间点的最终定位坐标;
轨迹重建模块,用于根据各最终定位坐标进行轨迹重建。
第三方面,本发明实施例提供一种计算机设备,包括存储器和处理器,所述存储器存储有计算机程序,所述处理器执行所述计算机程序时实现以下步骤:
采集终端设备在多个时间点的惯性传感器数据及无线保真网络的信号强度数据;
根据各所述时间点的无线保真网络的信号强度数据,生成各所述时间点对应的初始定位坐标;
根据各所述时间点的惯性传感器数据,生成各所述时间点对应的位移向量;
将各所述时间点对应的初始定位坐标及位移向量输入全局定位模型,得到各所述时间点的最终定位坐标;
根据各最终定位坐标进行轨迹重建。
第四方面,本发明实施例提供一种计算机可读存储介质,其上存储有计算机程序,所述计算机程序被处理器执行时实现以下步骤:
采集终端设备在多个时间点的惯性传感器数据及无线保真网络的信号强度 数据;
根据各所述时间点的无线保真网络的信号强度数据,生成各所述时间点对应的初始定位坐标;
根据各所述时间点的惯性传感器数据,生成各所述时间点对应的位移向量;
将各所述时间点对应的初始定位坐标及位移向量输入全局定位模型,得到各所述时间点的最终定位坐标;
根据各最终定位坐标进行轨迹重建。
上述轨迹重建方法、装置、计算机设备和存储介质,通过采集终端设备在多个时间点的惯性传感器数据及无线保真网络的信号强度数据;根据各所述时间点的无线保真网络的信号强度数据,生成各所述时间点对应的初始定位坐标;根据各所述时间点的惯性传感器数据,生成各所述时间点对应的位移向量;将各所述时间点对应的初始定位坐标及位移向量输入全局定位模型,得到各所述时间点的最终定位坐标;根据各最终定位坐标进行轨迹重建。根据本申请实施例提供的轨迹重建方法,根据各时间点的惯性传感器数据及无线保真网络的信号强度数据,生成各时间点对应的位移向量及初始定位坐标,并通过全局定位模型对各时间点对应的初始定位坐标、位移向量与各时间点的最终定位坐标进行建模,可以在复杂的物体运动模式以及在复杂的无线传播环境下得到更加准确的位置估计,提升了轨迹重建的精度。
附图说明
图1为本申请实施例提供的轨迹重建方法的实施环境图;
图2为本申请实施例提供的一种轨迹重建方法的流程图;
图3为本申请实施例提供的另一种轨迹重建方法的流程图;
图4为本申请实施例提供的另一种轨迹重建方法的流程图;
图5为本申请实施例提供的另一种轨迹重建方法的流程图;
图6为本申请实施例提供的另一种轨迹重建方法的流程图;
图7为本申请实施例提供的另一种轨迹重建方法的流程图;
图8为本申请实施例提供的另一种轨迹重建方法的流程图;
图9为本申请实施例提供的另一种轨迹重建方法的流程图;
图10为本申请实施例提供的一种轨迹重建装置的框图;
图11为本申请实施例提供的一种计算机设备的框图。
具体实施方式
为了使本申请的目的、技术方案及优点更加清楚明白,以下结合附图及实施例,对本申请进行进一步详细说明。应当理解,此处描述的具体实施例仅仅用以解释本申请,并不用于限定本申请。
本申请提供的轨迹重建方法,可以应用于如图1所示的应用环境中。其中,终端102通过网络与服务器104通过网络进行通信,在一个实施例中,终端102与服务器104间基于MQTT协议以无线方式(4G/5G或WiFi)进行通信。服务器104接收到的数据首先通过运行MQTT协议的Mosquitto代理软件进行转发,进一步由运行在NodeJS平台上的JavaScript脚本进行接收和处理,并将结果返回至终端102。其中,终端102可以但不限于是各种个人计算机、笔记本电脑、智能手机、平板电脑和便携式可穿戴设备,在一个实施例中,该终端102内置加速度计、陀螺仪、磁力计、气压计等惯性传感器及无线接入模块,服务器104可以用独立的服务器或者是多个服务器组成的服务器集群来实现。
请参考图2,其示出了本实施例提供的一种轨迹重建方法,以该方法应用于图1中的终端为例进行说明,包括以下步骤:
步骤202,采集终端设备在多个时间点的惯性传感器数据及无线保真网络的信号强度数据。
在本申请的一个实施例中,终端设备中内置加速度计、陀螺仪、磁力计、气压计等惯性传感器,在本申请的另一个实施例中,终端设备内置线性加速度传感器,该加速度传感器是一种基于加速度计和陀螺仪的复合传感器,终端设备还内置旋转向量传感器,该旋转向量传感器是一种基于加速度计,陀螺仪,和磁力计的复合传感器。具体的,通过上述惯性传感器获取终端设备的在多个时间点时的惯性传感器数据。
在本申请的一个实施例中,终端设备中内置WiFi模块。具体的,通过该WiFi 模块可以获取终端设备的在多个时间点时的无线保真网络的信号强度数据,其中,每一时间点的无线保真网络的信号强度数据包括当前时间点时,接收到的多个无线接入点的无线保真网络的信号强度数据。
步骤204,根据各时间点的无线保真网络的信号强度数据,生成各时间点对应的初始定位坐标。
在本申请的一个实施例中,根据每一时间点对应的多个无线接入点的WiFi信号强度值,计算当前时间点下终端设备距离各个无线接入点的距离,根据各个无线接入点的实际位置,通过几何算法可以得到当前时间点下的终端设备的初始定位坐标。
步骤206,根据各时间点的惯性传感器数据,生成各时间点对应的位移向量。
在本申请的一个实施例中,根据每一时间点对应的终端设备的加速度值、角速度值及方向值,计算每一时间点对应的终端设备移动的方向,结合预设的单位步长,可以得到各时间点对应的位移向量,该位移向量包括当前时间点到下一时间点的步长及移动方向。
步骤208,将各时间点对应的初始定位坐标及位移向量输入全局定位模型,得到各时间点的最终定位坐标。
在本申请的一个实施例中,该全局定位模型表示输入参数(各时间点对应的初始定位坐标及位移向量)与输出参数(各时间点的最终定位坐标)的映射关系。
在本申请的一个实施例中,该全局定位模型可以为高斯过程状态空间模型、变分高斯过程状态空间模型、卷积神经网络(Convolutional Neural Networks,CNN)模型、深信度网络(Deep Belief Networks,DBF)模型,限制波尔兹曼机(Restricted Boltzmann Machine,RBM)模型及自动编码器(AutoEncoder)中的一种。
步骤210,根据各最终定位坐标进行轨迹重建。
在本申请的一个实施例中,各最终定位坐标反映了终端设备在各时间的定位坐标,可以按照时间顺序依次连接各最终定位坐标的方式对终端设备的移动轨迹进行重建。
在本申请实施例提供的轨迹重建方法中,通过采集终端设备在多个时间点的惯性传感器数据及无线保真网络的信号强度数据;根据各所述时间点的无线保真网络的信号强度数据,生成各所述时间点对应的初始定位坐标;根据各所述时间点的惯性传感器数据,生成各所述时间点对应的位移向量;将各所述时间点对应的初始定位坐标及位移向量输入全局定位模型,得到各所述时间点的最终定位坐标;根据各最终定位坐标进行轨迹重建。根据本申请实施例提供的轨迹重建方法,根据各时间点的惯性传感器数据及无线保真网络的信号强度数据,生成各时间点对应的位移向量及初始定位坐标,并通过全局定位模型对各时间点对应的初始定位坐标、位移向量与各时间点的最终定位坐标进行建模,可以在复杂的物体运动模式以及在复杂的无线传播环境下得到更加准确的位置估计,提升了轨迹重建的精度。
在实施环境中若存在多个移动终端时,可以根据多个移动终端建立的多个局部定位模型确定全局定位模型。因此请参考图3,其示出了本实施例提供的另一种轨迹重建方法的流程图,该轨迹重建方法可以应用于上文所述的实施环境中的终端102中。在上述图2所示实施例的基础上,上述步骤208具体可以包括以下步骤:
步骤302,根据各时间点对应的初始定位坐标及位移向量对初始定位模型进行训练,得到局部定位模型;局部定位模型包括多个模型参数。
在本申请的一个实施例中,根据各时间点对应的初始定位坐标及位移向量对初始定位模型进行训练后,可以得到局部定位模型的多个模型参数,依据该多个模型参数,可以直接确定终端设备的局部定位模型。该定位模型由均值函数及协方差函数决定,其中,一部分模型参数用于表示协方差函数的模型参数,一部分模型参数用于表示噪声的协方差矩阵。
步骤304,将多个模型参数上传至中心节点,以使中心节点对多个局部定位模型的各个模型参数进行均衡操作,并得到均衡后的多个模型参数。
在本申请的一个实施例中,所述中心节点为图1实施环境中的服务器104,中心节点除了与终端设备102通信连接之外,还与当前场景中其他终端设备进 行通信连接,在终端设备102建立好局部定位模型后,会将该模型包含的多个模型参数上传至中心节点中,中心节点同时也会接收并保存其他终端设备上传的多个模型参数,为了提高全局定位模型的定位准确性,中心节点会将每一个终端设备上传的局部定位模型多个模型参数进行均衡协调,并得到经过均衡协调后的全局定位模型的均衡后的多个模型参数。在本申请的一个事实例中,每当与中心节点连接的一个终端设备上传局部定位模型的多个模型参数时,中心节点都会对全局定位模型的多个模型参数进行更新,并将更新后的全局定位模型的多个模型参数发送至与该中心节点连接的所有终端设备。
步骤306,接收中心节点返回的均衡后的多个模型参数。
步骤308,根据均衡后的多个模型参数建立全局定位模型。
在本申请的一个实施例中,接收到的均衡后的多个模型参数与上传至中心节点的多个模型参数的数量相同,通过该均衡后的多个模型参数可以确定唯一一个全局定位模型。
步骤310,将各时间点对应的初始定位坐标及位移向量输入全局定位模型,得到各时间点的最终定位坐标。
在本申请的一个实施例中,终端设备在接收到该均衡后的多个模型参数后,可以直接利用该多个模型参数重新建立全局定位模型,并根据该全局定位模型对输入的各时间点对应的初始定位坐标及位移向量进行运算,以得到各时间点的最终定位坐标。
在本申请实施例提供的轨迹重建方法中,通过根据各时间点对应的初始定位坐标及位移向量对初始定位模型进行训练,得到局部定位模型;局部定位模型包括多个模型参数;将多个模型参数上传至中心节点,以使中心节点对多个局部定位模型的各个模型参数进行均衡操作,并得到均衡后的多个模型参数;接收中心节点返回的均衡后的多个模型参数;根据均衡后的多个模型参数建立全局定位模型;将各时间点对应的初始定位坐标及位移向量输入全局定位模型,得到各时间点的最终定位坐标。根据本申请实施例提供的轨迹重建方法,通过上传得到的局部定位模型的多个模型参数,以使中心节点根据各终端设备上传的各个模型参数进行均衡操作,并获取均衡后的全局定位模型的模型参数。通 过上述分布式的模型建立过程,扩大了模型规模,进而提升了基于全局定位模型定位的准确性,并且,由于只传输局部模型的模型参数,模型训练与轨迹重建的相关计算下沉到手机端,保证了用户手机数据的隐私。
请参考图4,其示出了本实施例提供的另一种轨迹重建方法的流程图,该轨迹重建方法可以应用于上文所述的实施环境中的终端102中。在上述图2所示实施例的基础上,所述局部定位模型为局部变分高斯过程状态空间模型,所述多个模型参数包括第一模型参数、第二模型参数、第三模型参数及第四模型参数,上述步骤302,具体可以包括以下步骤:
步骤402,利用高斯过程对各时间点对应的初始定位坐标与真实位置坐标间的映射关系进行建模,得到观测量子模型,观测量子模型包括第一模型参数及第二模型参数。
在本申请的一个实施例中,基于WiFi接收信号强度的无线保真网络定位技术仅能提供粗略的位置估计,且位置估计的精度取决于行人在室内的具体位置,无线接入点的分布,以及室内复杂的电波传播环境。本发明利用高斯过程对WiFi定位技术提供的初始定位坐标与真实位置坐标间的映射关系进行建模。
具体的,对于该描述初始定位坐标与真实位置坐标间的映射关系的高斯过程可以定义为:
y=g(x)+r
其中g(x)为具有第一模型参数θ g的高斯过程,r为服从高斯分布
Figure PCTCN2020098555-appb-000001
的观测噪声,R为观测噪声的协方差矩阵,高斯过程的均值函数为m g(x)=x。
在本申请的一个实施例中,通过该第一模型参数及第二模型参数可以确定唯一一个观测量子模型。
步骤404,利用高斯过程对各时间点对应的位移向量与真实位置坐标间的映射关系进行建模,得到状态演进子模型,状态演进子模型包括第三模型参数及第四模型参数。
在本申请的一个实施例中,该状态演进子模型同样由高斯过程表示,且其均值函数可以由行人航位推算技术刻画,即:
m f(x t-1,u t)=x t-1+u t
其中,x t表示t时间点的真实位置坐标,u t表示t时间点的位移向量。
在本申请的一个实施例中,通过该第三模型参数及第四模型参数可以确定唯一一个状态演进子模型。
步骤406,根据观测量子模型及状态演进子模型得到局部变分高斯过程状态空间模型。
在本申请的一个实施例中,该局部变分高斯过程状态空间模型可以由该观测量子模型及状态演进子模型共同表示。在一个可选的实施例中,该局部变分高斯过程状态空间模型还包括初始真实位置坐标的分布概率,表示为:
Figure PCTCN2020098555-appb-000002
根据本申请实施例提供的轨迹重建方法,将局部变分高斯过程状态空间模型分为状态演进子模型与观测量子模型,并按顺序分别学习状态演进子模型与观测量子模型,简化了局部变分高斯过程状态空间模型学习的复杂度,进而提高了轨迹重建的效率。
本申请实施例还提供了另一种轨迹重建方法,该轨迹重建方法可以应用于上文所述的实施环境中的终端102中。在上文所述的实施例的基础上,所述步骤402,具体还可以包括以下步骤:
将第一模型参数、第二模型参数、第三模型参数及第四模型参数上传至中心节点,以使中心节点对多个局部变分高斯过程状态空间模型的各个模型参数进行均衡操作,并得到均衡后的第一模型参数、第二模型参数、第三模型参数及第四模型参数。
本申请实施例还提供了另一种轨迹重建方法,该轨迹重建方法可以应用于上文所述的实施环境中的终端102中。在上文所述的实施例的基础上,所述观测量子模型包括:
Figure PCTCN2020098555-appb-000003
g t=g(x t)
Figure PCTCN2020098555-appb-000004
式中,m g(x)表示高斯过程g(x)的均值函数,k g(x,x′)表示高斯过程g(x)的协方差函数,k g(x,x′)由所述第一模型参数θ g确定,g t表示概率分布y t|g t的均值,R表示概率分布y t|g t的协方差矩阵,R为所述第二模型参数。
请参考图5,其示出了本实施例提供的另一种轨迹重建方法的流程图,该轨迹重建方法可以应用于上文所述的实施环境中的终端102中。在上述图4所示实施例的基础上,上述步骤402,具体可以包括以下步骤:
步骤502,根据训练数据集{x 1:N,y 1:N},生成对应的观测量对数边际似然函数。
在本申请的一个实施例中,该观测量对数边际似然函数表示为:
Figure PCTCN2020098555-appb-000005
式中,I N为单位矩阵,所述训练数据集{x 1:N,y 1:N}用于描述各所述时间点对应的初始定位坐标与真实位置间的映射关系。
步骤504,通过最大化观测量对数边际似然函数,得到第一模型参数及第二模型参数。
在本申请的一个实施例中,可以通过共轭梯度法(conjugate gradient,CG)和LBFGS算法对该观测量对数边际似然函数进行最大化处理,并得到第一模型参数θ g及第二模型参数R。
步骤506,根据第一模型参数及第二模型参数确定观测量子模型。
在本申请的一个实施例中,通过获取到的第一模型参数θ g及第二模型参数R可以确定唯一一个观测量子模型。
通过本实施例提供的轨迹重建方法,可以通过对得到的观测量对数边际似然函数进行最大化处理,得到可以确定观测量子模型的第一模型参数及第二模型参数,进而得到该观测量子模型。可以降低局部变分高斯过程状态空间模型的训练工作量,提升计算效率。
本申请实施例还提供了另一种轨迹重建方法,该轨迹重建方法可以应用于上文所述的实施环境中的终端102中。在上文所述的实施例的基础上,所述状 态演进子模型包括:
Figure PCTCN2020098555-appb-000006
f t=f(x t-1,u t)
Figure PCTCN2020098555-appb-000007
式中,m f(x,u)表示高斯过程f(x,u)的均值函数,k f((x,u),(x′,u′))表示高斯过程f(x,u)的协方差函数,k f((x,u),(x′,u′))由所述第三模型参数θ f确定,f t表示概率分布x t|f t的均值,Q表示概率分布x t|f t的协方差矩阵,Q为所述第四模型参数。
请参考图6,其示出了本实施例提供的另一种轨迹重建方法的流程图,该轨迹重建方法可以应用于上文所述的实施环境中的终端102中。在上述图4所示实施例的基础上,上述步骤404,具体可以包括以下步骤:
步骤602,获取M个诱导点{z 1:M,v 1:M}。
在本申请的一个实施例中,为了处理状态演进子模型中输入与输出均不可观测的情况,同时为了降低学习状态演进子模型的计算复杂度,本发明采用变分稀疏高斯过程框架,通过引入M个辅助的诱导点v 1:M来反映观测量所包含的信息。这些诱导点是相应高斯过程在诱导输入z 1:M上的取值,且与未知函数值f 1:T满足联合高斯分布。
步骤604,基于M个诱导点{z 1:M,v 1:M}得到局部变分高斯过程状态空间模型的证据下界,证据下界为与后验近似q(x 0:T)、q(v 1:M)、第三模型参数θ f、第四模型参数Q、以及诱导输入z 1:M相关的泛函。
在本申请的一个实施例中,基于引入的M个诱导点,可以得到初始的局部变分高斯过程状态空间模型的证据下界,该证据下界表示为:
Figure PCTCN2020098555-appb-000008
其中q(x 0:T,f 1:T,v 1:M)是对后验分布p(x 0:T,f 1:T,v 1:M|y 1:T)的变分近似。高斯过程状态空间模型中所有变量的联合分布可以写为:
Figure PCTCN2020098555-appb-000009
其中p(f 1:T|v 1:M)是在
Figure PCTCN2020098555-appb-000010
位置上的高斯过程后验,
Figure PCTCN2020098555-appb-000011
为状态演进子模型在t时刻的输入。为了使上述证据下界易处理,本发明选取如下的后验分布近似形式:
q(x 0:T,f 1:T,v 1:M)=q(x 0:T)p(f 1:T|v 1:M)q(v 1:M)
由此,高斯过程状态空间模型的证据下界可以被写成关于后验近似q(x 0:T),q(v 1:M),模型参数θ f,Q,以及诱导输入z 1:M的泛函:
Figure PCTCN2020098555-appb-000012
其中KL(.‖.)表示KL散度,
Figure PCTCN2020098555-appb-000013
表示信息熵。
步骤606,固定{θ f,Q,z 1:M},对证据下界进行最大化处理,得到q(x 0:T)及q(v 1:M)。
在本申请的一个实施例中,当固定{θ f,Q,z 1:M}时,最优的q(x 0:T)及q(v 1:M)可以通过使用变分法最大化上述证据下界得出。因此,得到的最优的q(v 1:M)为高斯分布
Figure PCTCN2020098555-appb-000014
且其特征参数(natural parameter)为:
Figure PCTCN2020098555-appb-000015
Figure PCTCN2020098555-appb-000016
其中
Figure PCTCN2020098555-appb-000017
Figure PCTCN2020098555-appb-000018
其中,
Figure PCTCN2020098555-appb-000019
通过上述公式可以得到最优的q *(x 0:T),将最优的q *(x 0:T)作为后验近似q(x 0:T)。
步骤608,固定q(x 0:T)和q(v 1:M),对证据下界进行最大化处理,得到第三模型参数θ f,第四模型参数Q。
在本申请的一个实施例中,模型的模型参数θ f、Q,以及诱导输入z 1:M可以通过最大化证据下界得到。本发明采用基于梯度的优化方案,证据下界关于参数
Figure PCTCN2020098555-appb-000020
的梯度可通过下式计算:
Figure PCTCN2020098555-appb-000021
步骤610,根据第三模型参数及第四模型参数确定状态演进子模型。
在本申请的一个实施例中,通过获取到的第三模型参数θ f及第四模型参数Q可以确定唯一一个状态演进子模型。
通过本实施例提供的轨迹重建方法,可以通过对得到的高斯过程状态空间 模型的证据下界进行最大化处理,通过分别固定参数{θ f,Q,z 1:M}及q(x 0:T),q(v 1:M)得到可以确定状态演进子模型的第三模型参数及第四模型参数,进而得到该状态演进子模型。可以降低局部变分高斯过程状态空间模型的训练工作量,提升计算效率。
请参考图7,其示出了本实施例提供的另一种轨迹重建方法的流程图,该轨迹重建方法可以应用于上文所述的实施环境中的终端102中。在上述图2所示实施例的基础上,上述步骤204,具体可以包括以下步骤:
步骤702,根据各时间点的无线保真网络的信号强度数据,建立相应的定位对数似然函数。
在本申请的一个实施例中,每一个时间点的无线保真网络的信号强度数据包括在当前时间点时,终端设备接收到的多个无线接入点的无线保真网络的信号强度,对于各无线接入点,其无线保真网络的信号强度数据与扫描距离的映射关系可以通过线性对数模型表示:
Figure PCTCN2020098555-appb-000022
式中,r i是通过多次采集得到的第i个无线接入点的无线保真网络的信号强度,A和B是信号路损参数,d i是参考距离,d 0是扫描距离,该扫描位置表示无线接入点与终端设备位置间的距离。噪声分布被假设为独立同分布的高斯噪声,均值为0方差为σ 2。每个无线接入点的线性对数模型的未知参数A、B及σ可通过线性最小二乘法(linear least square)估计得出。
在本申请的一个实施例中,在得到所有无线接入点的线性对数模型的参数后,本发明基于最大似然估计(maximum likelihood estimation,MLE)得到初始定位坐标。假设在任意位置的一次扫描可以得到M AP个无线接入点的无线保真网络的信号强度,其中每个无线接入点的无线保真网络的信号强度都归属于一个独立的无线接入点。该定位对数似然函数可以被写为:
Figure PCTCN2020098555-appb-000023
式中,erf(.)代表标准高斯误差函数,P dec是无线保真网络的信号强度的截断极限,
Figure PCTCN2020098555-appb-000024
步骤704,通过最大化定位对数似然函数,得到各时间点对应的初始定位坐标。
在本申请的一个实施例中,各时间点对应的初始定位坐标可以通过最大化该对数似然函数得到,即:
y=arg max xl(x)
式中,y表示初始定位坐标。
通过本实施例提供的轨迹重建方法,可以通过各无线接入点的线性对数模型及各时间点的无线保真网络的信号强度数据,建立定位对数似然函数,并通过最大化该定位对数似然函数得到各时间点的初始定位坐标。可以提高初始定位坐标对真实位置坐标描述的准确度,进而保证了后续最终位置坐标的准确度,提升了重建轨迹的精度。
请参考图8,其示出了本实施例提供的另一种轨迹重建方法的流程图,该轨迹重建方法可以应用于上文所述的实施环境中的终端102中。在上述图2所示实施例的基础上,所述位移向量用于表示所述终端设备在所述时间点的运动方向及步长,上述步骤206,具体可以包括以下步骤:
步骤802,通过旋转向量传感器获取终端设备的旋转向量信号。
步骤804,根据旋转向量信号获取各时间点的运动方向,运动方向为终端设备朝向与地磁北极间的偏角。
步骤806,通过预设步长及各时间点的运动方向确定各时间点的位移向量。
在本申请的一个实施例中,终端设备保持平放状态,且终端设备的前端朝 向行走方向,旋转向量传感器为一种基于加速度计,陀螺仪,和磁力计的复合传感器,对其输出的四元数进行处理,可以得到每一时间点时终端设备朝向与地磁北极间的偏角,将其作为运动方向。根据预设的单位步长,可以得到各时间点的运动方向确定各时间点的位移向量。
通过本实施例得到的各时间点的位移向量,可以更加准确的反映每一时间点到下一时间点的运动方向及运动距离,进而保证了后续最终位置坐标的准确度,提升了重建轨迹的精度。
请参考图9,其示出了本实施例提供的另一种轨迹重建方法的流程图,该轨迹重建方法可以应用于上文所述的实施环境中的终端102中。在上述图2所示实施例的基础上,还可以包括以下步骤:
步骤902,通过线性加速度传感器获取终端设备相对重力方向的线性加速度信号。
步骤904,对线性加速度信号进行滑动平均处理。
步骤906,对经过滑动平均处理后的线性加速度信号进行局部峰值检测,得到多个时间点,各时间点为线性加速度信号处于局部峰值的时刻。
在本申请的一个实施例中,终端设备保持平放状态,且终端设备的前端朝向行走方向,线性加速度传感器为一种基于加速度计和陀螺仪的复合传感器,该线性加速度传感器输出的终端设备相对重力方向的线性加速度信号,对该线性加速度信号进行滑动平均处理,对经过滑动平均处理后的线性加速度信号进行局部峰值检测,得到多个时间点,各时间点为线性加速度信号处于局部峰值的时刻,该多个时间点为脚步落下的多个时刻。
通过本实施例得到的多个时间点,可以更加准确的反映脚步落下的多个时刻,进而保证了后续最终位置坐标的准确度,提升了重建轨迹的精度。
应该理解的是,虽然上述流程图中的各个步骤按照箭头的指示依次显示,但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明,这些步骤的执行并没有严格的顺序限制,这些步骤可以以其它的顺序执 行。而且,上述流程图中的至少一部分步骤可以包括多个子步骤或者多个阶段,这些子步骤或者阶段并不必然是在同一时刻执行完成,而是可以在不同的时刻执行,这些子步骤或者阶段的执行顺序也不必然是依次进行,而是可以与其它步骤或者其它步骤的子步骤或者阶段的至少一部分轮流或者交替地执行。
请参考图10,其示出了本申请实施例提供的一种轨迹重建装置1000的框图。如图10所示,所述轨迹重建装置1000可以包括:采集模块1001、无线保真网络定位模块1002、位移向量确定模块1003、模型定位模块1004及轨迹重建模块1005,其中:
所述采集模块1001,用于采集终端设备在多个时间点的惯性传感器数据及无线保真网络的信号强度数据。
所述无线保真网络定位模块1002,用于根据各所述时间点的无线保真网络的信号强度数据,生成各所述时间点对应的初始定位坐标。
所述位移向量确定模块1003,用于根据各所述时间点的惯性传感器数据,生成各所述时间点对应的位移向量。
所述模型定位模块1004,用于将各所述时间点对应的初始定位坐标及位移向量输入全局定位模型,得到各所述时间点的最终定位坐标。
所述轨迹重建模块1005,用于根据各最终定位坐标进行轨迹重建。
在本申请的一个实施例中,所述模型定位模块1004,具体用于:根据各所述时间点对应的初始定位坐标及位移向量对初始定位模型进行训练,得到局部定位模型;所述局部定位模型包括多个模型参数;将所述多个模型参数上传至中心节点,以使所述中心节点对多个所述局部定位模型的各个模型参数进行均衡操作,并得到均衡后的多个模型参数;接收所述中心节点返回的所述均衡后的多个模型参数;根据所述均衡后的多个模型参数建立所述全局定位模型;将各所述时间点对应的初始定位坐标及位移向量输入所述全局定位模型,得到各所述时间点的最终定位坐标。
在本申请的一个实施例中,所述局部定位模型为局部变分高斯过程状态空间模型,所述多个模型参数包括第一模型参数、第二模型参数、第三模型参数 及第四模型参数,所述模型定位模块1004,还具体用于:利用高斯过程对各所述时间点对应的初始定位坐标与真实位置坐标间的映射关系进行建模,得到观测量子模型,所述观测量子模型包括所述第一模型参数及所述第二模型参数;利用高斯过程对各所述时间点对应的位移向量与真实位置坐标间的映射关系进行建模,得到状态演进子模型,所述状态演进子模型包括所述第三模型参数及所述第四模型参数;根据所述观测量子模型及所述状态演进子模型得到所述局部变分高斯过程状态空间模型。
在本申请的一个实施例中,所述多个模型参数包括第一模型参数、第二模型参数、第三模型参数及第四模型参数,所述模型定位模块1004,还具体用于:将所述第一模型参数、所述第二模型参数、所述第三模型参数及所述第四模型参数上传至中心节点,以使所述中心节点对多个所述局部变分高斯过程状态空间模型的各个模型参数进行均衡操作,并得到均衡后的所述第一模型参数、所述第二模型参数、所述第三模型参数及所述第四模型参数。
在本申请的一个实施例中,所述观测量子模型包括:
Figure PCTCN2020098555-appb-000025
g t=g(x t)
Figure PCTCN2020098555-appb-000026
式中,m g(x)表示高斯过程g(x)的均值函数,k g(x,x′)表示高斯过程g(x)的协方差函数,k g(x,x′)由所述第一模型参数θ g确定,g t表示概率分布y t|g t的均值,R表示概率分布y t|g t的协方差矩阵,R为所述第二模型参数。
在本申请的一个实施例中,所述模型定位模块1004,还具体用于:根据训练数据集{x 1:N,y 1:N},生成对应的观测量对数边际似然函数:
Figure PCTCN2020098555-appb-000027
式中,I N为单位矩阵,所述训练数据集{x 1:N,y 1:N}用于描述各所述时间点对应的初始定位坐标与真实位置间的映射关系;通过最大化所述观测量对数边际似然函数,得到所述第一模型参数及所述第二模型参数;根据所述第一模型参数及所述第二模型参数确定所述观测量子模型。
在本申请的一个实施例中,所述状态演进子模型包括:
Figure PCTCN2020098555-appb-000028
f t=f(x t-1,u t)
Figure PCTCN2020098555-appb-000029
式中,m f(x,u)表示高斯过程f(x,u)的均值函数,k f((x,u),(x′,u′))表示高斯过程f(x,u)的协方差函数,k f((x,u),(x′,u′))由所述第三模型参数θ f确定,f t表示概率分布x t|f t的均值,Q表示概率分布x t|f t的协方差矩阵,Q为所述第四模型参数。
在本申请的一个实施例中,所述模型定位模块1004,还具体用于:获取M个诱导点{z 1:M,v 1:M};基于所述M个诱导点{z 1:M,v 1:M}得到所述局部变分高斯过程状态空间模型的证据下界,所述证据下界为与后验近似q(x 0:T)、q(v 1:M)、所述第三模型参数θ f、所述第四模型参数Q、以及诱导输入z 1:M相关的泛函;固定{θ f,Q,z 1:M},对所述证据下界进行最大化处理,得到q(x 0:T)及q(v 1:M);固定q(x 0:T)和q(v 1:M),对所述证据下界进行最大化处理,得到所述第三模型参数θ f,所述第四模型参数Q;根据所述第三模型参数及所述第四模型参数确定所述状态演进子模型。
在本申请的一个实施例中,所述无线保真网络定位模块1002,具体用于:根据各所述时间点的无线保真网络的信号强度数据,建立相应的定位对数似然函数;通过最大化所述定位对数似然函数,得到各所述时间点对应的初始定位坐标。
在本申请的一个实施例中,所述位移向量确定模块1003,具体用于:通过旋转向量传感器获取所述终端设备的旋转向量信号;根据所述旋转向量信号获取各所述时间点的运动方向,所述运动方向为所述终端设备朝向与地磁北极间的偏角;通过预设步长及各所述时间点的运动方向确定各所述时间点的位移向量。
在本申请的一个实施例中,所述位移向量确定模块1003,还具体用于:通过线性加速度传感器获取所述终端设备相对重力方向的线性加速度信号;对所述线性加速度信号进行滑动平均处理;对经过滑动平均处理后的线性加速度信号进行局部峰值检测,得到所述多个时间点,各所述时间点为所述线性加速度 信号处于局部峰值的时刻。
关于轨迹重建装置的具体限定可以参见上文中对于轨迹重建方法的限定,在此不再赘述。上述轨迹重建装置中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中,也可以以软件形式存储于计算机设备中的存储器中,以便于处理器调用执行以上各个模块对应的操作。
在一个实施例中,提供了一种计算机设备,该计算机设备可以是终端,其内部结构图可以如图11所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口、显示屏和输入装置。其中,该计算机设备的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统和计算机程序。该内存储器为非易失性存储介质中的操作系统和计算机程序的运行提供环境。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机程序被处理器执行时以实现一种轨迹重建方法。该计算机设备的显示屏可以是液晶显示屏或者电子墨水显示屏,该计算机设备的输入装置可以是显示屏上覆盖的触摸层,也可以是计算机设备外壳上设置的按键、轨迹球或触控板,还可以是外接的键盘、触控板或鼠标等。
本领域技术人员可以理解,图11中示出的结构,仅仅是与本申请方案相关的部分结构的框图,并不构成对本申请方案所应用于其上的计算机设备的限定,具体的计算机设备可以包括比图中所示更多或更少的部件,或者组合某些部件,或者具有不同的部件布置。
在一个实施例中,提供了一种计算机设备,包括存储器和处理器,存储器中存储有计算机程序,该处理器执行计算机程序时实现以下步骤:
采集终端设备在多个时间点的惯性传感器数据及无线保真网络的信号强度数据;
根据各所述时间点的无线保真网络的信号强度数据,生成各所述时间点对应的初始定位坐标;
根据各所述时间点的惯性传感器数据,生成各所述时间点对应的位移向量;
将各所述时间点对应的初始定位坐标及位移向量输入全局定位模型,得到各所述时间点的最终定位坐标;
根据各最终定位坐标进行轨迹重建。
在一个实施例中,提供了一种计算机可读存储介质,其上存储有计算机程序,计算机程序被处理器执行时实现以下步骤:
采集终端设备在多个时间点的惯性传感器数据及无线保真网络的信号强度数据;
根据各所述时间点的无线保真网络的信号强度数据,生成各所述时间点对应的初始定位坐标;
根据各所述时间点的惯性传感器数据,生成各所述时间点对应的位移向量;
将各所述时间点对应的初始定位坐标及位移向量输入全局定位模型,得到各所述时间点的最终定位坐标;
根据各最终定位坐标进行轨迹重建。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程,是可以通过计算机程序来指令相关的硬件来完成,所述的计算机程序可存储于一非易失性计算机可读取存储介质中,该计算机程序在执行时,可包括如上述各方法的实施例的流程。其中,本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用,均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器(ROM)、可编程ROM(PROM)、电可编程ROM(EPROM)、电可擦除可编程ROM(EEPROM)或闪存。易失性存储器可包括随机存取存储器(RAM)或者外部高速缓冲存储器。作为说明而非局限,RAM以多种形式可得,诸如静态RAM(SRAM)、动态RAM(DRAM)、同步DRAM(SDRAM)、双数据率SDRAM(DDRSDRAM)、增强型SDRAM(ESDRAM)、同步链路(Synchlink)DRAM(SLDRAM)、存储器总线(Rambus)直接RAM(RDRAM)、直接存储器总线动态RAM(DRDRAM)、以及存储器总线动态RAM(RDRAM)等。
以上实施例的各技术特征可以进行任意的组合,为使描述简洁,未对上述实施例中的各个技术特征所有可能的组合都进行描述,然而,只要这些技术特 征的组合不存在矛盾,都应当认为是本说明书记载的范围。
以上所述实施例仅表达了本申请的几种实施方式,其描述较为具体和详细,但并不能因此而理解为对发明专利范围的限制。应当指出的是,对于本领域的普通技术人员来说,在不脱离本申请构思的前提下,还可以做出若干变形和改进,这些都属于本申请的保护范围。因此,本申请专利的保护范围应以所附权利要求为准。

Claims (14)

  1. 一种轨迹重建方法,其特征在于,所述方法包括:
    采集终端设备在多个时间点的惯性传感器数据及无线保真网络的信号强度数据;
    根据各所述时间点的无线保真网络的信号强度数据,生成各所述时间点对应的初始定位坐标;
    根据各所述时间点的惯性传感器数据,生成各所述时间点对应的位移向量;
    将各所述时间点对应的初始定位坐标及位移向量输入全局定位模型,得到各所述时间点的最终定位坐标;
    根据各最终定位坐标进行轨迹重建。
  2. 根据权利要求1所述的方法,其特征在于,将各所述时间点对应的初始定位坐标及位移向量输入全局定位模型,得到各所述时间点的最终定位坐标,包括:
    根据各所述时间点对应的初始定位坐标及位移向量对初始定位模型进行训练,得到局部定位模型;所述局部定位模型包括多个模型参数;
    将所述多个模型参数上传至中心节点,以使所述中心节点对多个所述局部定位模型的各个模型参数进行均衡操作,并得到均衡后的多个模型参数;
    接收所述中心节点返回的所述均衡后的多个模型参数;
    根据所述均衡后的多个模型参数建立所述全局定位模型;
    将各所述时间点对应的初始定位坐标及位移向量输入所述全局定位模型,得到各所述时间点的最终定位坐标。
  3. 根据权利要求2所述的方法,其特征在于,所述局部定位模型为局部变分高斯过程状态空间模型,所述多个模型参数包括第一模型参数、第二模型参数、第三模型参数及第四模型参数,所述根据各所述时间点对应的初始定位坐标及位移向量对初始定位模型进行训练,得到局部定位模型,包括:
    利用高斯过程对各所述时间点对应的初始定位坐标与真实位置坐标间的映射关系进行建模,得到观测量子模型,所述观测量子模型包括所述第一模型参数及所述第二模型参数;
    利用高斯过程对各所述时间点对应的位移向量与真实位置坐标间的映射关 系进行建模,得到状态演进子模型,所述状态演进子模型包括所述第三模型参数及所述第四模型参数;
    根据所述观测量子模型及所述状态演进子模型得到所述局部变分高斯过程状态空间模型。
  4. 根据权利要求3所述的方法,其特征在于,所述将所述多个模型参数上传至中心节点,以使所述中心节点对多个所述局部定位模型的各个模型参数进行均衡操作,并得到均衡后的多个模型参数,包括:
    将所述第一模型参数、所述第二模型参数、所述第三模型参数及所述第四模型参数上传至中心节点,以使所述中心节点对多个所述局部变分高斯过程状态空间模型的各个模型参数进行均衡操作,并得到均衡后的所述第一模型参数、所述第二模型参数、所述第三模型参数及所述第四模型参数。
  5. 根据权利要求3所述的方法,其特征在于,所述观测量子模型包括:
    Figure PCTCN2020098555-appb-100001
    g t=g(x t)
    Figure PCTCN2020098555-appb-100002
    式中,m g(x)表示高斯过程g(x)的均值函数,k g(x,x′)表示高斯过程g(x)的协方差函数,k g(x,x′)由所述第一模型参数θ g确定,g t表示概率分布y t|g t的均值,R表示概率分布y t|g t的协方差矩阵,R为所述第二模型参数。
  6. 根据权利要求5所述的方法,其特征在于,所述利用高斯过程对各所述时间点对应的初始定位坐标与真实位置间的映射关系进行建模,得到观测量子模型,包括:
    根据训练数据集{x 1:N,y 1:N},生成对应的观测量对数边际似然函数:
    Figure PCTCN2020098555-appb-100003
    式中,I N为单位矩阵,所述训练数据集{x 1:N,y 1:N}用于描述各所述时间点对应的初始定位坐标与真实位置间的映射关系;
    通过最大化所述观测量对数边际似然函数,得到所述第一模型参数及所述第二模型参数;
    根据所述第一模型参数及所述第二模型参数确定所述观测量子模型。
  7. 根据权利要求3所述的方法,其特征在于,所述状态演进子模型包括:
    Figure PCTCN2020098555-appb-100004
    f t=f(x t-1,u t)
    Figure PCTCN2020098555-appb-100005
    式中,m f(x,u)表示高斯过程f(x,u)的均值函数,k f((x,u),(x′,u′))表示高斯过程f(x,u)的协方差函数,k f((x,u),(x′,u′))由所述第三模型参数θ f确定,f t表示概率分布x t|f t的均值,Q表示概率分布x t|f t的协方差矩阵,Q为所述第四模型参数。
  8. 根据权利要求7所述的方法,其特征在于,所述利用高斯过程对各所述时间点对应的位移向量与真实位置间的映射关系进行建模,得到状态演进子模型,包括:
    获取M个诱导点{z 1:M,v 1:M};
    基于所述M个诱导点{z 1:M,v 1:M}得到所述局部变分高斯过程状态空间模型的证据下界,所述证据下界为与后验近似q(x 0:T)、q(v 1:M)、所述第三模型参数θ f、所述第四模型参数Q、以及诱导输入z 1:M相关的泛函;
    固定{θ f,Q,z 1:M},对所述证据下界进行最大化处理,得到q(x 0:T)及q(v 1:M);
    固定q(x 0:T)和q(v 1:M),对所述证据下界进行最大化处理,得到所述第三模型参数θ f,所述第四模型参数Q;
    根据所述第三模型参数及所述第四模型参数确定所述状态演进子模型。
  9. 根据权利要求1所述的方法,其特征在于,所述根据各所述时间点的无线保真网络的信号强度数据,生成各所述时间点对应的初始定位坐标,包括:
    根据各所述时间点的无线保真网络的信号强度数据,建立相应的定位对数似然函数;
    通过最大化所述定位对数似然函数,得到各所述时间点对应的初始定位坐标。
  10. 根据权利要求1所述的方法,其特征在于,所述位移向量用于表示所 述终端设备在所述时间点的运动方向及步长;所述根据各所述时间点的惯性传感器数据,生成各所述时间点对应的位移向量,包括:
    通过旋转向量传感器获取所述终端设备的旋转向量信号;
    根据所述旋转向量信号获取各所述时间点的运动方向,所述运动方向为所述终端设备朝向与地磁北极间的偏角;
    通过预设步长及各所述时间点的运动方向确定各所述时间点的位移向量。
  11. 根据权利要求1所述的方法,其特征在于,所述方法还包括:
    通过线性加速度传感器获取所述终端设备相对重力方向的线性加速度信号;
    对所述线性加速度信号进行滑动平均处理;
    对经过滑动平均处理后的线性加速度信号进行局部峰值检测,得到所述多个时间点,各所述时间点为所述线性加速度信号处于局部峰值的时刻。
  12. 一种轨迹重建装置,其特征在于,所述装置包括:
    采集模块,用于采集终端设备在多个时间点的惯性传感器数据及无线保真网络的信号强度数据;
    无线保真网络定位模块,用于根据各所述时间点的无线保真网络的信号强度数据,生成各所述时间点对应的初始定位坐标;
    位移向量确定模块,用于根据各所述时间点的惯性传感器数据,生成各所述时间点对应的位移向量;
    模型定位模块,用于将各所述时间点对应的初始定位坐标及位移向量输入全局定位模型,得到各所述时间点的最终定位坐标;
    轨迹重建模块,用于根据各最终定位坐标进行轨迹重建。
  13. 一种计算机设备,包括存储器和处理器,所述存储器存储有计算机程序,其特征在于,所述处理器执行所述计算机程序时实现权利要求1至11中任一项所述方法的步骤。
  14. 一种计算机可读存储介质,其上存储有计算机程序,其特征在于,所述计算机程序被处理器执行时实现权利要求1至11中任一项所述的方法的步骤。
PCT/CN2020/098555 2019-07-22 2020-06-28 轨迹重建方法、装置、计算机设备和存储介质 WO2021012879A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
ZA2022/01092A ZA202201092B (en) 2019-07-22 2022-01-24 Trajectory reconstruction method and apparatus, computer device and storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910660363.0A CN110493710B (zh) 2019-07-22 2019-07-22 轨迹重建方法、装置、计算机设备和存储介质
CN201910660363.0 2019-07-22

Publications (1)

Publication Number Publication Date
WO2021012879A1 true WO2021012879A1 (zh) 2021-01-28

Family

ID=68547551

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/098555 WO2021012879A1 (zh) 2019-07-22 2020-06-28 轨迹重建方法、装置、计算机设备和存储介质

Country Status (3)

Country Link
CN (1) CN110493710B (zh)
WO (1) WO2021012879A1 (zh)
ZA (1) ZA202201092B (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117292085A (zh) * 2023-11-27 2023-12-26 浙江大学 一种支持三维建模的实体交互控制方法及其装置

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110493710B (zh) * 2019-07-22 2021-01-26 香港中文大学(深圳) 轨迹重建方法、装置、计算机设备和存储介质
CN111028347B (zh) * 2019-12-24 2021-06-22 贝壳找房(北京)科技有限公司 用于重建物理工作空间的三维模型的方法和系统
CN113157844B (zh) * 2021-04-26 2022-09-20 上海德衡数据科技有限公司 基于北斗定位模块的农业物联网方法、系统及装置
CN113820658A (zh) * 2021-08-18 2021-12-21 上海闻泰电子科技有限公司 无线定位方法、装置、电子设备及存储介质
CN113766634B (zh) * 2021-08-31 2023-08-04 深圳Tcl新技术有限公司 基于5g的定位方法、装置、计算机设备和存储介质
CN115470913A (zh) * 2022-09-29 2022-12-13 南京师范大学 基于量子游走的pir传感器网络行为轨迹的重构方法及装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103776447A (zh) * 2014-01-28 2014-05-07 无锡智感星际科技有限公司 一种近距离移动智能设备间定位方法
CN105157706A (zh) * 2015-08-25 2015-12-16 武汉易得路位置科技有限公司 一种基于多传感器信息的WiFi热点位置测量方法
CN105554879A (zh) * 2015-05-07 2016-05-04 中国科学院信息工程研究所 一种室内区域定位优化方法及系统
CN107302754A (zh) * 2017-05-10 2017-10-27 广东工业大学 一种基于WiFi与PDR的室内定位简易方法
CN110493710A (zh) * 2019-07-22 2019-11-22 香港中文大学(深圳) 轨迹重建方法、装置、计算机设备和存储介质

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101650431A (zh) * 2009-05-20 2010-02-17 北京派瑞根科技开发有限公司 基于网络计算获取物体自身运动轨迹方法
DE102009030076A1 (de) * 2009-06-23 2010-12-30 Symeo Gmbh Abbildungsverfahren mittels synthetischer Apertur, Verfahren zur Bestimmung einer Relativgeschwindigkeit zwischen einem wellenbasierten Sensor und einem Objekt bzw. Vorrichtung zur Durchführung der Verfahren
CN102087109A (zh) * 2009-12-04 2011-06-08 财团法人资讯工业策进会 位置估测系统、装置及其估测方法
CN202904027U (zh) * 2012-04-11 2013-04-24 四川中软科技有限公司 一种自主室内定位系统
CN106162555B (zh) * 2016-09-26 2019-09-10 湘潭大学 室内定位方法及系统
CN109470238B (zh) * 2017-09-08 2023-09-01 中兴通讯股份有限公司 一种定位方法、装置和移动终端
CN108668249B (zh) * 2018-07-10 2021-01-22 北京物资学院 一种移动终端室内定位方法及装置
CN108761514A (zh) * 2018-08-03 2018-11-06 北斗国信智能科技(北京)有限公司 一种融合北斗或gps和传感器的定位系统及定位方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103776447A (zh) * 2014-01-28 2014-05-07 无锡智感星际科技有限公司 一种近距离移动智能设备间定位方法
CN105554879A (zh) * 2015-05-07 2016-05-04 中国科学院信息工程研究所 一种室内区域定位优化方法及系统
CN105157706A (zh) * 2015-08-25 2015-12-16 武汉易得路位置科技有限公司 一种基于多传感器信息的WiFi热点位置测量方法
CN107302754A (zh) * 2017-05-10 2017-10-27 广东工业大学 一种基于WiFi与PDR的室内定位简易方法
CN110493710A (zh) * 2019-07-22 2019-11-22 香港中文大学(深圳) 轨迹重建方法、装置、计算机设备和存储介质

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117292085A (zh) * 2023-11-27 2023-12-26 浙江大学 一种支持三维建模的实体交互控制方法及其装置
CN117292085B (zh) * 2023-11-27 2024-02-09 浙江大学 一种支持三维建模的实体交互控制方法及其装置

Also Published As

Publication number Publication date
CN110493710B (zh) 2021-01-26
ZA202201092B (en) 2023-11-29
CN110493710A (zh) 2019-11-22

Similar Documents

Publication Publication Date Title
WO2021012879A1 (zh) 轨迹重建方法、装置、计算机设备和存储介质
US11295472B2 (en) Positioning method, positioning apparatus, positioning system, storage medium, and method for constructing offline map database
CN107635204B (zh) 一种运动行为辅助的室内融合定位方法及装置、存储介质
Zhao et al. Crowdsourcing and multisource fusion-based fingerprint sensing in smartphone localization
US10422849B2 (en) Method and system for signal-based localization
US20160371394A1 (en) Indoor localization using crowdsourced data
Sunny et al. Applications and challenges of human activity recognition using sensors in a smart environment
Vedadi et al. Automatic visual fingerprinting for indoor image-based localization applications
KR20140015594A (ko) 다수의 디바이스 상태들을 분류하는 방법 및 장치
CN113763549B (zh) 融合激光雷达和imu的同时定位建图方法、装置和存储介质
WO2023082797A1 (zh) 定位方法、定位装置、存储介质与电子设备
Zhao et al. Vivid: Augmenting vision-based indoor navigation system with edge computing
Batstone et al. Towards real-time time-of-arrival self-calibration using ultra-wideband anchors
Manos et al. Walking direction estimation using smartphone sensors: A deep network-based framework
Chen et al. A wifi indoor localization method based on dilated cnn and support vector regression
CA2894863A1 (en) Indoor localization using crowdsourced data
CN117352126A (zh) 肌肉受力可视化方法、装置、计算机设备以及存储介质
Xu et al. Smartphone-based crowdsourcing for panoramic virtual tour construction
CN108322889A (zh) 一种多源数据融合室内定位的方法、存储介质及智能终端
Huang et al. VariFi: Variational Inference for Indoor Pedestrian Localization and Tracking Using IMU and WiFi RSS
WO2022143711A1 (zh) 一种数据处理方法及装置
Chen et al. Deep-learning-based inertial odometry for pedestrian tracking using attention mechanism and res2net module
Li et al. An enhanced transition model for unsupervised localization
Zhang et al. Human motion capture system based on distributed wearable sensing technology
KR102669839B1 (ko) 3차원 가상모델 생성을 위한 전처리 방법 및 그를 위한 컴퓨팅 장치

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: 20844935

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: 20844935

Country of ref document: EP

Kind code of ref document: A1