CN-122015854-A - Intelligent terminal tight combination method for multi-machine collaborative filtering enhancement

CN122015854ACN 122015854 ACN122015854 ACN 122015854ACN-122015854-A

Abstract

The invention relates to an intelligent terminal tight combination method based on multi-machine collaborative filtering enhancement, which combines the observables of a plurality of intelligent mobile phones, fuses a plurality of observations by parallel application of filtering and optimizing algorithms, and outputs an optimal estimated state quantity in real time by navigation. Firstly, modeling a GNSS/PDR tight combination model based on Kalman filtering for a single smart phone, wherein the model is used as the front end of an overall system, the state quantity calculated by the front end in real time is used as the initial value of further optimization of a back end factor graph, an EKF state constraint factor is built at the same time, then, a single factor graph model is modeled for each smart phone, wherein the single factor graph model comprises the EKF state constraint factor, the GNSS pseudo-range factor and a PDR step size factor built at the front end, then, the inter-machine distance constraint factor is built by utilizing distance observation among a plurality of smart phones, the single smart phone factor graph is expanded into a multi-machine collaborative overall factor graph, and finally, the integrated filtering solution is used as the optimal estimation of the factor graph, the EKF state constraint factor, the multi-machine PDR step size factor, the GNSS pseudo-range factor and the inter-machine distance factor, so that the robustness and the reliability of the combined navigation system are enhanced, and the GNSS pseudo-range factor is constrained by using a Huber robust kernel function in view of the rough GNSS pseudo-range observation.

Inventors

MENG JIAN
SHEN ZULIANG
LIU LING
Nan Zihan
LI SHENGYING
YU LIYAO

Assignees

东南大学

Dates

Publication Date: 20260512
Application Date: 20260205

Claims (10)

1. The intelligent terminal tightly-combining method based on multi-machine collaborative filtering enhancement is characterized by integrating the observation of a plurality of intelligent mobile phones and applying a filtering and optimizing method in parallel, and comprises the following steps: A filtering model and an optimizing model are built for each single intelligent mobile phone: Step 1, respectively establishing a state transfer equation and a measurement update equation according to PDR track recursion and GNSS pseudo-range observation; Step 2, calculating prior estimated state quantity of the system based on the extended Kalman filtering principle Priori error covariance matrix ; Step 3, calculating an innovation vector based on an extended Kalman filtering principle Covariance matrix corresponding to innovation vector And construct standardized information test statistics ; Step 4, using Huber equivalent weight function to divide the standardized innovation test statistic into two parts, namely effective measurement and suspicious measurement, and respectively calculating the corresponding weight of the two parts, namely the effective measurement and the suspicious measurement, wherein the weight is 100%, the suspicious measurement is subjected to weight reduction processing by using the weight factor, and after the Huber function is weighted, the PDR predicted quantity and the GNSS pseudo-range observed quantity output state estimation are fused Covariance matrix ; Step 5, constructing an EKF state constraint factor according to the EKF output state quantity, wherein the factor residual error corresponds to the difference between the EKF output state and the state to be optimized; Step 6, constructing a PDR step length factor according to a PDR positioning principle, wherein the factor residual error corresponds to the difference between the step length obtained by a PDR algorithm and the positioning interval of a corresponding adjacent epoch; Step 7, constructing a GNSS pseudo-range factor according to the GNSS observation model, wherein the factor residual error corresponds to the difference between the predicted pseudo-range and the GNSS observation pseudo-range; step 8, constructing a distance constraint factor between the smart phones according to the distance between the smart phones, wherein the factor residual error corresponds to the difference between the distance between the smart phones and the predicted distance; and 9, taking the factors corresponding to each mobile phone and the EKF state constraint factors into an overall factor graph, taking the Kalman filtering output state quantity constructed in the steps 1-4 as an iteration initial value of factor graph optimization, and optimizing and outputting the navigation state in real time.
2. The intelligent terminal tight combination method based on multi-machine collaborative filtering enhancement according to claim 1, wherein step 1 comprises the following procedures: Respectively establishing a state transfer equation and a measurement update equation according to the PDR track recursion and the GNSS pseudo-range observation, In the formula, Is a 5-dimensional state vector, and comprises a PDR north error, an east error, a step error, a course angle error and a receiver clock error respectively; representing a state transition matrix from epoch k-1 to k; Is the 5-dimensional state vector of the last epoch; n-dimensional observation vectors are represented, the N-dimensional observation vectors respectively correspond to differences between PDR reverse prediction pseudo-ranges and GNSS observation pseudo-ranges, and N represents the number of visible satellites; Representing an observation matrix; Representing the process noise at k-1 epoch, The observation noise under the k epoch is represented, and the process noise and the observation noise are modeled as mutually independent gaussian white noise, namely: 。
3. the intelligent terminal tight combination method based on multi-machine collaborative filtering enhancement according to claim 2, wherein the step 2 includes the following procedures: The prior estimation state quantity and the prior error covariance matrix of the system are calculated based on the extended Kalman filtering principle respectively: In the formula, Representing a priori state estimation vectors; Representing a priori state error covariance matrix; representing a prior epoch posterior state error covariance matrix; A covariance matrix of system noise; the step 3 comprises the following steps: calculating an innovation vector and a covariance matrix thereof based on a Kalman filtering principle: In the formula, Representing an innovation vector; representing covariance matrix corresponding to the innovation vector, wherein the innovation vector satisfies multidimensional state distribution when no fault is observed, namely Calculating normalized innovation from the innovation vector and covariance matrix as test statistic: In the formula, In order to normalize the innovation vector, The representation takes diagonal elements.
4. The intelligent terminal tight combination method based on multi-machine collaborative filtering enhancement according to claim 3, wherein step 4 comprises the following processes: the standardized innovation test statistic is divided into two parts, namely effective measurement and suspicious measurement by using a Huber equivalent weight function, and the corresponding weight is calculated respectively, wherein the effective measurement is 100% weighted, the suspicious measurement is subjected to weight reduction by using a weight factor pair: In the formula, Representing the observation weight obtained after fault monitoring; Respectively representing Huber equivalence weight function test thresholds, ; Kalman filter measurement update, calculating Kalman gain Posterior state quantity Posterior error covariance matrix Wherein Observation weights obtained for Huber weight functions Diagonal matrix of components: 。
5. the intelligent terminal tight combination method based on multi-machine collaborative filtering enhancement according to claim 4, wherein step 5 comprises the following procedures: according to the EKF output state quantity, an EKF state constraint factor is constructed: In the formula, Representing the state quantity to be estimated of the ith mobile phone under the kth epoch, Representing the state quantity of EKF fusion output of the ith mobile phone under the kth epoch, And (3) with And respectively representing an error function and a covariance matrix of the EKF state constraint factor of the ith smart phone under k epochs.
6. The intelligent terminal tight combination method based on multi-machine collaborative filtering enhancement according to claim 5, wherein step 6 comprises the following processes: according to the PDR positioning principle, constructing a PDR step factor: In the formula, And (3) with Respectively representing an error function and a covariance matrix of the PDR step length constraint factor of the ith smart phone under k epochs; And (3) with (J=k-1, k) respectively represents the east position and the north position of the ith smart phone to be optimally solved under the jth epoch.
7. The intelligent terminal tight combination method based on multi-machine collaborative filtering enhancement according to claim 6, wherein step 7 comprises the following procedures: constructing GNSS pseudo-range factors according to the GNSS observation model: In the formula, And (3) with Representing an error function and a covariance matrix of a GNSS pseudo-range factor corresponding to the ith smart phone under the kth epoch; Indicating the nth satellite observed pseudo range of the ith smart phone under the k epoch, Representing the position, satellite position and receiver clock error to be solved by the ith smart phone under the kth epoch In view of multipath and non-line-of-sight effects of the GNSS signals, to suppress the influence of the GNSS observation coarse difference, the obtained predicted pseudo-range is reversely deduced, and the pseudo-range factor of the GNSS is constrained by using a Huber kernel function, namely: In the formula, 。
8. The intelligent terminal tight combination method based on multi-machine collaborative filtering enhancement according to claim 7, wherein step 8 includes the following procedures: According to the distance constraint between smart phones, an inter-machine distance factor is constructed: In the formula, And (3) with Respectively representing an error function and a covariance matrix corresponding to the distance factor of the smart phone; Representing the distance between the two smart phones at the kth epoch; representing a prediction distance between positioning pushers to be solved by two smart phones; 、 respectively representing the east and north positioning of the first smart phone under the kth epoch; 、 respectively representing the east and north positioning of the second smart phone under the kth epoch.
9. The intelligent terminal tight combination method based on multi-machine collaborative filtering enhancement according to claim 8, wherein step 9 includes the following procedures: expanding a factor graph established based on a single smart phone into a multi-phone collaborative integral factor graph through inter-machine distance constraint factors, and taking the Kalman filtering output state quantity constructed in the steps 1-4 as an iteration initial value of factor graph optimization to optimize and output a navigation state in real time, namely: In the formula, The representation takes the minimum value of the value, 、 Respectively representing EKF constraint factor expressions of the first mobile phone and the second mobile phone under the m-th epoch, And (3) with Respectively representing covariance matrixes corresponding to EKF constraint factors of the first mobile phone and the second mobile phone under the mth epoch, 、 The PDR step length constraint factor expressions of the first mobile phone and the second mobile phone under the mth epoch are respectively represented, And (3) with Respectively representing covariance matrixes corresponding to PDR step length constraint factors of the first mobile phone and the second mobile phone under the mth epoch, 、 Respectively representing pseudo-range constraint factor expressions between the first mobile phone and the second mobile phone and the ith satellite under the mth epoch, And (3) with Respectively representing covariance matrixes corresponding to pseudo-range constraint factors between the first mobile phone and the second mobile phone and the ith satellite under the mth epoch, 、 And (5) representing a distance constraint factor expression and a covariance matrix between the two mobile phones under the mth epoch.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements a smart terminal tight-combining method based on multi-machine collaborative filtering enhancement according to any of the preceding claims 1-9 when executing the program.

Description

Intelligent terminal tight combination method for multi-machine collaborative filtering enhancement Technical Field The invention relates to a multi-machine collaborative filtering enhanced intelligent terminal tight combination method, belongs to a multi-source information fusion technology, and is particularly suitable for the field of intelligent terminal multi-sensor fusion robust real-time navigation positioning. Background With the continuous development of smart phone technology, smart phone-based location services (Location Based Service, LBS) have become a basic service requirement for people in daily life and traveling. Currently, smartphones have built in a variety of sensors that can be used for direct positioning or assisted positioning, such as GNSS chips, tri-axial accelerometers, gyroscopes, magnetic sensors, cameras, bluetooth, etc. The user can determine the current position, speed, gesture and other information through the data of the sensors, so as to acquire the corresponding positioning service. Among sensors embedded in smart phones, GNSS receivers are the only navigation devices that can provide all-weather three-dimensional coordinates and speed and time information for users, and can provide meter-level and even centimeter-level high-precision positioning services in open environments. However, GNSS signals in challenging environments such as tunnels, urban canyons, etc. are very susceptible to non-line-of-sight and multipath effects, resulting in reduced accuracy and reliability of the positioning results. The pedestrian track estimation algorithm (PEDESTRAIN DEAD Reckoning, PDR) achieves recursive navigation solutions by fusing sensor data to estimate the step size and orientation during movement. Therefore, the PDR is independent of external information and is an autonomous navigation system, and meanwhile, unlike an inertial measurement unit (Inertial Measurement Unit, IMU), the PDR solves the integral of output data, so that the inertial navigation defect is overcome to a certain extent. But the PDR algorithm can realize better performance in linear motion, but still generates larger errors in an environment with more turns. The single navigation positioning method cannot be used for complex and changeable actual situations, and obvious advantage complementation exists among multiple sensors. The GNSS can perform navigation initialization for the PDR and correct track drift, and the PDR can improve the problem of discontinuous positioning caused by low GNSS resolving frequency. Currently, the combined GNSS/PDR navigation method has become one of the methods widely adopted for pedestrian navigation. Meanwhile, in the multi-source information fusion, a filtering technology and an optimizing technology are two main methods for realizing the optimal estimation of autonomous navigation. The GNSS/PDR integrated navigation method based on Extended Kalman filtering (Extended KALMAN FILTER, EKF) is mature in theory, clear in mechanism, simple in implementation, and has very mature theoretical support in the aspects of front-end fault detection and rear-end confidence evaluation, but has the limitations of inflexible structure, strong nonlinear constraint, poor reconfigurability and the like. The GNSS/PDR integrated navigation method based on Factor graph optimization (Graph Optimization, FGO) converts the multisource fusion problem into the equivalent nonlinear optimization problem through maximum posterior probability density estimation to realize the optimal estimation of the state variable, can effectively solve the problems of heterogeneous and nonlinear states of the observation sensor in the integrated navigation system, has the advantages of plug and play, data smoothing, simple fault isolation, topology expansion and the like, and effectively improves the robustness of the positioning result in complex challenge scenes. However, the factor graph optimization model also has the theoretical and practical problems of high calculation complexity, complex fault propagation mechanism, lack of posterior confidence quantification and the like. Meanwhile, GNSS/PDR integrated navigation is extremely easy to degrade into a single navigation system under an extremely challenging environment, so that a positioning result is rapidly diverged. Therefore, in order to further improve the positioning robustness and reliability of the intelligent terminal pedestrian navigation system in the challenging environment, a more efficient and effective information fusion method is urgently required to be developed. Disclosure of Invention The invention aims to solve the problems of multi-sensor fusion degradation, positioning reliability reduction, algorithm operation efficiency low and the like of a pedestrian navigation system based on a smart phone under an extremely challenging environment. The invention provides a multi-machine collaborative filtering enhanced intelligent terminal tight combination