CN110081890A - A kind of dynamic K arest neighbors map-matching method of combination depth network - Google Patents
A kind of dynamic K arest neighbors map-matching method of combination depth network Download PDFInfo
- Publication number
- CN110081890A CN110081890A CN201910438446.5A CN201910438446A CN110081890A CN 110081890 A CN110081890 A CN 110081890A CN 201910438446 A CN201910438446 A CN 201910438446A CN 110081890 A CN110081890 A CN 110081890A
- Authority
- CN
- China
- Prior art keywords
- error
- data
- value
- angle
- dynamic
- Prior art date
- Legal status (The legal status 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 status listed.)
- Granted
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
- G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
- G01C21/28—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network with correlation of data from several navigational instruments
- G01C21/30—Map- or contour-matching
Abstract
A kind of dynamic K arest neighbors map-matching method of combination depth network, comprising: Step 1: acquisition GPS data, carries out data cleansing to the noise occurred in GPS data collection process；Step 2: map match obtains the range error and deflection error in experimental data；Step 3: the input data to multi-layer perception (MLP) is normalized, using normalized longitude and latitude as the input of multiple perceptron model, training multiple perceptron model obtains dynamic k value；Step 4: each test data is combined according to the k value that training obtains with Euclidean distance, the Prediction distance error and prediction direction error of test data are obtained using k nearest-neighbor algorithm, then obtains the subpoint of corresponding test point；Step 5: obtaining the subpoint of test data according to the longitude of test data, latitude and Prediction distance error and prediction direction error.The present invention can improve the single k value situation of the overall situation existing for k nearest neighbor algorithm, get Optimal Error value.
Description
Technical field
The invention belongs to map matching technology fields, are related to a kind of dynamic K arest neighbors map match of combination depth network
Method.
Background technique
With the development of global-positioning technology and universal, more and more equipment are embedded in GPS (global position
System) global positioning function.These equipment can be collected into a large amount of mobile position data daily, contain in these data
Traffic information and user behavior abundant, can be by these data applications to path prediction, GPS track analysis and activity recognition etc.
In research.But since always in the presence of the error being difficult to avoid that, (such as satellite and its propagation are on the way for GPS system in data acquisition
Caused by diameter itself, it is unable to measure or cannot be inherently missed with the propagation delay error of calibration model calculating, receiver user
Difference), lead between the GPS positioning got point and physical location that there are deviations.The main purpose of map match is digitally
It is in figure that a series of tracing points are associated with road network, with correct as GPS positioning error and caused by positional shift.
There is the characteristic of similitude, which is primarily referred to as the GPS positioning of adjacent domain between GPS positioning point adjacent domain
The distance between point error and deflection error due to disturbing factor (environment, building etc.) proximity and there is similitude, because
This, which is combined with machine learning, so that there are class relationship between the GPS positioning point between adjacent domain, using going through
Error can be corrected by history data.Existing map-matching algorithm, as based on geometry matching algorithm, based on topological relation
Matching algorithm, the matching algorithm based on probability statistics and the matching algorithm based on advanced method, mainly according to before and after location data
It is limited to put the topological structure between the geometrical relationship constituted and road to be matched, seldom number is positioned using a large amount of history
Error correction is carried out according to characteristic, therefore, it may be considered that the characteristic that point tolerance has similitude is positioned according to adjacent domain, using one
The location data error calibration method that kind combines multi-layer perception (MLP) (MLP) with k nearest-neighbor (kNN) algorithm.
Summary of the invention
It is an object of the invention to be directed to above-mentioned the problems of the prior art, a kind of dynamic K of combination depth network is provided
Arest neighbors map-matching method, by training multiple perceptron model, the model after enabling training is raw for each location data
At a k value, dynamic k value is then formed, improves the single k value situation of the overall situation existing for k nearest neighbor algorithm, gets Optimal Error
Value.
To achieve the goals above, the technical solution adopted by the present invention the following steps are included:
Step 1: acquisition GPS data, carries out data cleansing to the noise occurred in GPS data collection process；
Step 2: data preparation；
2.1, map match is carried out to the GPS data after cleaning；
2.2, the distance between original point and match point in GPS data, direction difference are sought, is missed as range error and direction
Difference；
It is obtained Step 3: training multiple perceptron model carries out dynamic k value；
3.1, longitude, the latitude in GPS data after cleaning are normalized；
3.2, multiple perceptron model is built by Selecting All Parameters；
3.3, using after normalization longitude, latitude data as the input of multiple perceptron model, by range error with not
The error obtained with the nearest-neighbor algorithm under the conditions of K value is compared, and the smallest k value of error is obtained, as each GPS point
Label, using the smallest k value of error as the output of multiple perceptron model；
3.4, after stablizing multiple perceptron model training by training data, a phase is predicted for each test data
The k value answered, and then form dynamic k value correcting method；
Step 4: corresponding k value and Euclidean distance that test data is obtained according to multiple perceptron model training carry out
In conjunction with, use k nearest-neighbor algorithm obtain test point range error and deflection error；
Step 5: obtaining test according to the longitude of test data, latitude and Prediction distance error and prediction direction error
The subpoint of data, the point are the result passed through after the dynamic K arest neighbors method in conjunction with depth network is corrected.
Rejected in the step one in data cleansing be above range, vehicle within certain time GPS coordinate without
Variation, format error and abnormal data.
The calculation formula of range error and deflection error is as follows in the step 2.2:
In above formula, (lon1, lat1), (lon2, lat2) is respectively original point and match point, and dlon, dlat are respectively
Longitude, latitude are subjected to the difference after Circular measure conversion, distance is range error；
The angle of line and direct north between angle=original point and match point
Angle in formula is deflection error.
The calculation formula of normalized is as follows in the step 3.1:
In formula, MaxValue, MinValue are the maximum value and minimum value of sequence.
Setup parameter content is as follows when building multiple perceptron model in the step 3.2:
The MLP number of plies is 3, including an input layer, a hidden layer, an output layer；
Input unit number is 2, including longitude and latitude；
Hidden layer unit number is 100；
Activation primitive between hidden layer and output layer is
In formula, viFor the output of i-th of hidden unit, c is Hidden unit number；
Optimizer is AdamOptimizer, learning rate 0.0001；
20% is used as test data, and 80% is used as training data；
Loss function is mean square error；
Training method is the error that propagated forward obtains predicted value and label value, and backpropagation updates network parameter.
The value range of label classification is 1-10 in the step 3.4, and k value is divided into 10 classes and is predicted.
The range error of test point and the calculating formula of deflection error are as follows in the step four:
angle_errj=angle_northi
if|head_anglej-head_anglei| 45 ° of or of < | head_anglej-head_anglei- 360 | 45 ° of <
In formula, distanceiFor range error, head_anglejFor course angle, dis_err is the prediction of each test point
Range error, angle_northiIt is pre- with the angle of direct north to meet course angle constraint condition similitude correction front and back point
Survey deflection error.
Compared with prior art, the present invention is with following the utility model has the advantages that existing error calibration method is main according to fixed
It is limited before and after the data of position to put the topological structure between the geometrical relationship constituted and road to be matched, seldom using a large amount of
Historical location data characteristic carry out error correction, the present invention firstly, training multiple perceptron model, enable training after model
A k value enough is predicted for each location data, forms dynamic k value, improves the single k value of the overall situation existing for k nearest neighbor algorithm
Situation；Secondly, obtaining the range error and deflection error of each anchor point by the map-matching method based on geometry；Finally,
K arest neighbors (k-Nearest Neighbor, kNN) algorithm and Euclidean algorithm (Euclidean Algorithm) are combined
Study is obtained to the error of correction data is needed, to obtain the subpoint of the point.The present invention is missed according to adjacent domain anchor point
Difference has the characteristic of similitude, using a kind of positioning for combining multi-layer perception (MLP) (MLP) with k nearest-neighbor (kNN) algorithm
Data error correction method in experiment, has carried out sufficiently model proposed by the present invention using true taxi track data
Verifying.The result shows that the more existing error correction model of model proposed by the present invention has preferable improve.
Detailed description of the invention
Error Graph between the original anchor point of Fig. 1 and matching result；
Fig. 2 is adjacent domain error similarity graph: (a) GPS positioning point range error figure；(b) GPS positioning point direction is missed
Difference；
Fig. 3 is k Distribution value histogram；
Fig. 4 is multi-layer perception (MLP) loss function curve graph.
Specific embodiment
Present invention will be described in further detail below with reference to the accompanying drawings.
1, defining track sets is T, and T is a series of continuous GPS positioning point p1,p2,...,pn, each anchor point includes
Longitude (pi.lat), latitude (pi.lon), time (pi.time), speed (pi.velocity), course angle (pi.path-angle)
Etc. features.2, range error is defined: assuming that RiFor anchor point piMatching result, then by RiWith piThe distance between be known as distance
ErrorAs shown in Figure 1.3, deflection error is defined: assuming that RiFor anchor point piMatching result, then by Ri、piBetween line with
Direct north is known as deflection error along clockwise angleAs shown in Figure 1.4, definition of proximity domain error similarity:
GPS positioning point piWith consecutive points pi-n,...,pi-1And pi+1,...,pm+1(wherein m, n ∈ N*)WithBetween have
Similitude.Referring to fig. 2, it can be seen that the range error of adjacent GPS positioning point is the most close, from Fig. 2 (b) from Fig. 2 (a)
It can be seen that the distance between adjacent GPS positioning point error is the most close.
The dynamic K arest neighbors map-matching method of present invention combination depth network specifically includes the following steps:
Step 1: data scrubbing；
Due to having noise in GPS data collection process, so needing to carry out data cleansing to it.For exceeding Xi'an
Range, the vehicle GPS coordinate within certain time is unchanged, the data of various format errors and exception are rejected.
Step 2: data preparation；
2.1, the process for carrying out data cleansing to GPS data only rejects abnormal data, and the positioning of GPS data misses
Difference still exists, therefore needs to carry out map match to GPS data before being trained model.
2.2, carry out map match after it is available arrive match point, it is poor by the distance between GPS original point and match point,
For direction difference as range error and deflection error, the expression formula of range error and deflection error is as follows:
In formula, (lon1, lat1), (lon2, lat2) is respectively original point and match point, and dlon, dlat are respectively will be through
Degree, latitude carry out the difference after Circular measure conversion, and distance is range error；
The angle of line and direct north between angle=original point and match point
Angle is deflection error in formula；
Step 3: training multiple perceptron model carries out dynamic k value and obtains；
For deep learning network, the format of input data will affect the training effect of model.Data are returned
One changes, and can not only accelerate the speed that gradient decline solves optimal solution, can also improve precision.The present invention is defeated to multi-layer perception (MLP)
Enter that the mathematic(al) representation being normalized is as follows, wherein, maximum, the minimum value that MaxValue, MinValue are sequence.
Multi-layer perception (MLP) (MLP) mainly has preferably effect in the identification of handwritten numeral, can be very good logarithm
Word 0-9 is differentiated.MLP is applied to field of traffic from image domains and made improvements by the principle for using for reference handwriting recognition,
The parameter setting of MLP is illustrated in table 1.Firstly, multi-layer perception (MLP) is built according to the parameter setting in table 1, secondly, will normalization
Input of the longitude and latitude afterwards as multi-layer perception (MLP) obtains actual error and the application condition acquired of KNN under the conditions of different K values
Label of the most similar k value as each anchor point, wherein label classification by as shown in figure 3, value range be 1-10, then will
K value is divided into 10 classes and is predicted.It is pre- for each test data after being stablized multiple perceptron model training by training data
A corresponding k value is surveyed to form dynamic k value to overcome the shortcomings of the single k value of the original overall situation and prepare for error correction.
1. multi-layer perception (MLP) parameter setting of table
Step 4: the Prediction distance error and deflection error of k nearest-neighbor acquisition test data；
According to matching result carry out error analysis it is found that the distance between adjacent GPS positioning point error, deflection error have
There is similitude, therefore utilize this feature on the basis of dynamic k value, is with Euclidean distance (Euclidean Distance)
Similarity metric, the similarity between the smaller then two o'clock of Euclidean distance between two anchor points are higher, then it is assumed that
More there is similitude between error vector between two o'clock.In the k value and Europe that each test data is obtained according to model training are several
It obtains distance to be combined using k nearest-neighbor algorithm, obtains range error and deflection error.Range error and deflection error are as follows
Shown in formula:
angle_errj=angle_northi
if|head_anglej-head_anglei| 45 ° of or of < | head_anglej-head_anglei- 360 | 45 ° of <
In formula, distanceiFor range error, head_anglejFor course angle, dis_err is the prediction of each test point
Range error, angle_northiIt is pre- with the angle of direct north to meet course angle constraint condition similitude correction front and back point
Survey deflection error.
Step 5: test data is obtained according to the longitude of test data, latitude, Prediction distance error and prediction direction error
Subpoint, which is the result passed through after the dynamic K arest neighbors method in conjunction with depth network is corrected.
Present invention proposition uses the network model for combining multi-layer perception (MLP) (MLP) with k nearest-neighbor (kNN) algorithm,
It is used for correction site error.By the network model, when can effectively improve the progress error correction of k nearest-neighbor algorithm
Global constant k value makes GPS positioning point fail to get Optimal Error value, influences to correct result.
Mixed model proposed by the present invention is compared with perseverance defining K value kNN, MLP and SVR.
Referring to fig. 4, the loss of network training and test tends to reduce, and restrains after 20 iteration, shows that model is correct
It trains and there is no over-fitting.Using MLP obtain K value after, seek correction error using KNN method, with permanent defining K value kNN,
MLP and SVR comparing result is as shown in table 2.Analyzing from table 2 can obtain, and be directly input with longitude and latitude regardless of MLP or SVR
The method for obtaining correction error has no positive effect.Using the KNN method of optimum k value, there is biggish improvement as a result, the present invention mentions
Mixed model out, the method for more permanent defining K value has further improvement on the basis of dynamic k value.
The different model error correction Comparative results of table 2.
The above is only presently preferred embodiments of the present invention, not to do restriction in any form to the present invention,
It will be apparent to a skilled person that the present invention can also carry out under the premise of not departing from spirit of that invention and principle
Several simple modifications and replacement, these modifications and replacement also can fall into the patent that the present invention delimited by submitted claim
Protection scope.
Claims (7)
1. a kind of dynamic K arest neighbors map-matching method of combination depth network, which comprises the following steps:
Step 1: acquisition GPS data, carries out data cleansing to the noise occurred in GPS data collection process；
Step 2: data preparation；
2.1, map match is carried out to the GPS data after cleaning；
2.2, the distance between original point and match point in GPS data, direction difference are sought, as range error and deflection error；
It is obtained Step 3: training multiple perceptron model carries out dynamic k value；
3.1, longitude, the latitude in GPS data after cleaning are normalized；
3.2, multiple perceptron model is built by Selecting All Parameters；
3.3, using after normalization longitude, latitude data as the input of multiple perceptron model, by range error and different K values
Under the conditions of the obtained error of nearest-neighbor algorithm be compared, obtain the smallest k value of error, as the label of each GPS point,
Using the smallest k value of error as the output of multiple perceptron model；
3.4, corresponding for each test data prediction one after being stablized multiple perceptron model training by training data
K value, and then form dynamic k value correcting method；
Step 4: test data is combined according to the corresponding k value that multiple perceptron model training obtains with Euclidean distance,
The range error and deflection error of test point are obtained using k nearest-neighbor algorithm；
Step 5: obtaining test data according to the longitude of test data, latitude and Prediction distance error and prediction direction error
Subpoint, which is the result passed through after the dynamic K arest neighbors method in conjunction with depth network is corrected.
2. combining the dynamic K arest neighbors map-matching method of depth network according to claim 1, it is characterised in that: in number
That rejects when according to cleaning is above unchanged range, the vehicle GPS coordinate within certain time, format error and abnormal data.
3. combining the dynamic K arest neighbors map-matching method of depth network according to claim 1, which is characterized in that described
Step 2.2 in the calculation formula of range error and deflection error it is as follows:
In above formula, (lon1, lat1), (lon2, lat2) is respectively original point and match point, and dlon, dlat are respectively will be through
Degree, latitude carry out the difference after Circular measure conversion, and distance is range error；
The angle of line and direct north between angle=original point and match point
Angle in formula is deflection error.
4. combining the dynamic K arest neighbors map-matching method of depth network according to claim 1, which is characterized in that described
Step 3.1 in normalized calculation formula it is as follows:
In formula, MaxValue, MinValue are the maximum value and minimum value of sequence.
5. combining the dynamic K arest neighbors map-matching method of depth network according to claim 1, which is characterized in that described
Step 3.2 in when building multiple perceptron model setup parameter content it is as follows:
The MLP number of plies is 3, including an input layer, a hidden layer, an output layer；
Input unit number is 2, including longitude and latitude；
Hidden layer unit number is 100；
Activation primitive between hidden layer and output layer is
In formula, viFor the output of i-th of hidden unit, c is Hidden unit number；
Optimizer is AdamOptimizer, learning rate 0.0001；
20% is used as test data, and 80% is used as training data；
Loss function is mean square error；
Training method is the error that propagated forward obtains predicted value and label value, and backpropagation updates network parameter.
6. combining the dynamic K arest neighbors map-matching method of depth network according to claim 1, it is characterised in that: described
Step 3.4 in label classification value range be 1-10, k value is divided into 10 classes and is predicted.
7. combining the dynamic K arest neighbors map-matching method of depth network according to claim 1, which is characterized in that described
The step of four in test point range error and deflection error calculating formula it is as follows:
angle_errj=angle_northi
if|head_anglej-head_anglei| 45 ° of or of < | head_anglej-head_anglei- 360 | 45 ° of <
In formula, distanceiFor range error, head_anglejFor course angle, dis_err is the Prediction distance of each test point
Error, angle_northiTo meet prediction side of the course angle constraint condition similitude correction front and back point with the angle of direct north
To error.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910438446.5A CN110081890B (en) | 2019-05-24 | 2019-05-24 | Dynamic k nearest neighbor map matching method combined with deep network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910438446.5A CN110081890B (en) | 2019-05-24 | 2019-05-24 | Dynamic k nearest neighbor map matching method combined with deep network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110081890A true CN110081890A (en) | 2019-08-02 |
| CN110081890B CN110081890B (en) | 2023-02-03 |
Family
ID=67421688
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910438446.5A Active CN110081890B (en) | 2019-05-24 | 2019-05-24 | Dynamic k nearest neighbor map matching method combined with deep network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110081890B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111879305A (en) * | 2020-06-16 | 2020-11-03 | 华中科技大学 | Multi-mode perception positioning model and system for high-risk production environment |
| CN112967296A (en) * | 2021-03-10 | 2021-06-15 | 重庆理工大学 | Point cloud dynamic region graph convolution method, classification method and segmentation method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5774831A (en) * | 1996-12-06 | 1998-06-30 | Gupta; Surender Kumar | System for improving average accuracy of signals from global positioning system by using a neural network to obtain signal correction values |
| CN103983263A (en) * | 2014-05-30 | 2014-08-13 | 东南大学 | Inertia/visual integrated navigation method adopting iterated extended Kalman filter and neural network |
| CN106441316A (en) * | 2016-09-08 | 2017-02-22 | 复旦大学 | Single-point road network matching method based on historical data |
| WO2018094375A1 (en) * | 2016-11-21 | 2018-05-24 | Nio Usa, Inc. | Method and system for vehicle data optimization |
| CN109313850A (en) * | 2016-06-07 | 2019-02-05 | 罗伯特·博世有限公司 | Method, apparatus and system for driver's identification of driving in the wrong direction |
-
2019
- 2019-05-24 CN CN201910438446.5A patent/CN110081890B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5774831A (en) * | 1996-12-06 | 1998-06-30 | Gupta; Surender Kumar | System for improving average accuracy of signals from global positioning system by using a neural network to obtain signal correction values |
| CN103983263A (en) * | 2014-05-30 | 2014-08-13 | 东南大学 | Inertia/visual integrated navigation method adopting iterated extended Kalman filter and neural network |
| CN109313850A (en) * | 2016-06-07 | 2019-02-05 | 罗伯特·博世有限公司 | Method, apparatus and system for driver's identification of driving in the wrong direction |
| CN106441316A (en) * | 2016-09-08 | 2017-02-22 | 复旦大学 | Single-point road network matching method based on historical data |
| WO2018094375A1 (en) * | 2016-11-21 | 2018-05-24 | Nio Usa, Inc. | Method and system for vehicle data optimization |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111879305A (en) * | 2020-06-16 | 2020-11-03 | 华中科技大学 | Multi-mode perception positioning model and system for high-risk production environment |
| CN111879305B (en) * | 2020-06-16 | 2022-03-18 | 华中科技大学 | Multi-mode perception positioning model and system for high-risk production environment |
| CN112967296A (en) * | 2021-03-10 | 2021-06-15 | 重庆理工大学 | Point cloud dynamic region graph convolution method, classification method and segmentation method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110081890B (en) | 2023-02-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11562168B2 (en) | Clustering for K-anonymity in location trajectory data | |
| CN105628033B (en) | A kind of map-matching method based on path connected relationship | |
| CN106776482B (en) | A kind of track similarity calculating method | |
| CN109405839B (en) | Traffic network off-line map matching algorithm based on multiple paths | |
| CN105424030B (en) | Fusion navigation device and method based on wireless fingerprint and MEMS sensor | |
| Lv et al. | The discovery of personally semantic places based on trajectory data mining | |
| US10546043B1 (en) | Triangulation for K-anonymity in location trajectory data | |
| Lian et al. | DeepWindow: Sliding window based on deep learning for road extraction from remote sensing images | |
| CN104197945B (en) | Global voting map matching method based on low-sampling-rate floating vehicle data | |
| Markos et al. | Unsupervised deep learning for GPS-based transportation mode identification | |
| Wei et al. | Map matching by Fréchet distance and global weight optimization | |
| CN106441316A (en) | Single-point road network matching method based on historical data | |
| CN104394588A (en) | Indoor positioning method based on Wi-Fi fingerprints and multi-dimensional scaling analysis | |
| WO2019134549A1 (en) | Positioning method and apparatus based on deep learning, computer device, and storage medium | |
| CN110081890A (en) | A kind of dynamic K arest neighbors map-matching method of combination depth network | |
| CN110389995A (en) | Lane information detection method, device, equipment and medium | |
| CN108362289A (en) | A kind of mobile intelligent terminal PDR localization methods based on Multi-sensor Fusion | |
| CN113932821B (en) | Track map matching method based on continuous window average direction characteristics | |
| Guo et al. | Transportation mode recognition with deep forest based on GPS data | |
| Hashemi | Automatic inference of road and pedestrian networks from spatial-temporal trajectories | |
| CN105890600A (en) | Subway passenger position inferring method based on mobile phone sensors | |
| CN111722306A (en) | Typhoon landing intensity prediction method and system based on TCN network model | |
| Lin et al. | Noise filtering, trajectory compression and trajectory segmentation on GPS data | |
| Huang et al. | Frequent pattern-based map-matching on low sampling rate trajectories | |
| Chen et al. | Tailored hidden Markov model: A tailored hidden Markov model optimized for cellular-based map matching |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |