CN113554869B - Road closure detection method based on multi-feature fusion - Google Patents

Road closure detection method based on multi-feature fusion Download PDF

Info

Publication number
CN113554869B
CN113554869B CN202110747124.6A CN202110747124A CN113554869B CN 113554869 B CN113554869 B CN 113554869B CN 202110747124 A CN202110747124 A CN 202110747124A CN 113554869 B CN113554869 B CN 113554869B
Authority
CN
China
Prior art keywords
road
closed
grid
road section
track
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.)
Active
Application number
CN202110747124.6A
Other languages
Chinese (zh)
Other versions
CN113554869A (en
Inventor
毛嘉莉
蔡圣诚
周傲英
赵俐晟
金澈清
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
East China Normal University
Original Assignee
East China Normal University
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 East China Normal University filed Critical East China Normal University
Priority to CN202110747124.6A priority Critical patent/CN113554869B/en
Publication of CN113554869A publication Critical patent/CN113554869A/en
Application granted granted Critical
Publication of CN113554869B publication Critical patent/CN113554869B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0125—Traffic data processing
    • G08G1/0129—Traffic data processing for creating historical data or processing based on historical data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/047—Probabilistic or stochastic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/01—Detecting movement of traffic to be counted or controlled
    • G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
    • G08G1/0137—Measuring and analyzing of parameters relative to traffic conditions for specific applications
    • G—PHYSICS
    • G08—SIGNALLING
    • G08G—TRAFFIC CONTROL SYSTEMS
    • G08G1/00—Traffic control systems for road vehicles
    • G08G1/065—Traffic control systems for road vehicles by counting the vehicles in a section of the road or in a parking area, i.e. comparing incoming count with outgoing count

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Traffic Control Systems (AREA)

Abstract

The invention discloses a road closure detection method based on multi-feature fusion, which comprises the following steps: and an off-line stage, performing gridding processing on the area to be detected and establishing grid indexes for the track data and the road network data. Then, extracting the traffic correlation strength between the road sections; meanwhile, a grid traffic flow sequence, a strongly associated road steering flow sequence of each road section and a vehicle turning frequency sequence in a local area are obtained based on historical track data. And in the on-line detection stage, the traffic flow of each grid at the current moment is predicted by using a CNN and LSTM combined model, and candidate closed grids are screened. And predicting the steering quantity of the strongly-associated road of each road section and the turn-around frequency in a local area by adopting a GCN and LSTM combined model, and identifying the closed road section according to the sudden reduction of the steering quantity or the sharp increase of the turn-around frequency of the vehicle. And finally, further judging the road closure type by combining the driving yaw detection and the single-vehicle track data. The invention improves the detection efficiency to ensure the real-time performance.

Description

Road closure detection method based on multi-feature fusion
Technical Field
The invention belongs to the technical field of track mining, and particularly relates to a road closure detection method based on multi-feature fusion.
Background
With the widespread use of GPS devices, the daily trips of residents increasingly rely on map navigation. In order to ensure the high precision of the electronic map, it is necessary to sense the dynamic changes of the road network to update the map. In recent years, much research has been devoted to the detection of road topology changes such as missing road discovery, intersection location and range recognition in a road network. In a road network, some road sections cannot pass due to factors such as traffic accidents, traffic control, road construction and the like. The untimely detection of the road closure event not only brings inconvenience to the travel of residents, but also causes huge economic loss. For example, in 2019, 10 months and 10 days, the bridge No. 1 national road in the tin-free city 312 collapses, so that the peripheral road section is in a closed state for a long time. However, because the map navigation software cannot detect the closed road section in time, many drivers still travel according to the navigation route, and large-area traffic paralysis occurs in the area near the collapsed bridge.
The traditional navigation system mainly finds the road closure event by providing a road segment closure real-time reporting function module for traffic participants. These methods are time-consuming and labor-consuming, have long update periods, and cannot ensure the accuracy of detection. With the rapid development of location-based technologies, the data platform accumulates a large amount of trajectory data, which contains rich real-time information of road segments. The track data can be used for detecting the closing condition of the road section, and the road network map is updated in real time. In view of the interaction between the dynamic change of the road network and the traffic conditions between adjacent roads and the inclined distribution characteristics of the trajectory data, only a few studies have been devoted to solving the problem of road closure detection, and they propose a threshold-based method and a statistical-based method, or learn the upper and lower bounds of traffic flow based on historical data, judge a road closure event by detecting an abnormality of a traffic flow sequence, or calculate the possibility of road closure by establishing a poisson process.
Considering that road closure usually causes traffic abnormal conditions such as traffic flow sudden drop, some scholars design traffic abnormal detection methods to find closed roads, including area-based traffic abnormal detection and track-based traffic abnormal detection. The former method of identifying abnormal regions according to abnormal flow of vehicles between regions includes Principal Component Analysis (PCA), wavelet transform-based method, Mahalanobis distance-based method, Likelihood Ratio Test (LRT) -based method, Support Vector Machine (SVM) -based method, relative divergence-based method, etc. The latter proposes an isolation-based anomaly detection method that identifies closed roads by comparing the degree of difference between the anomalous route and the historically normal route.
The method only considers the attribute of abnormal and sudden drop of the traffic flow and does not combine the space-time characteristics of the trajectory data, the traffic characteristic variation diversity of different road closure events cannot be covered, and the traffic jam is easily mistakenly judged as road closure. In addition, the existing method does not pay attention to the distinguishing of the road closure types, and cannot provide accurate support for navigation service.
Disclosure of Invention
The invention aims to solve the technical problem of providing a road closure detection method based on multi-feature fusion, which aims at overcoming the defects of the prior art, accurately positions a closed road by using track data, ensures high-efficiency execution and provides effective decision support for navigation service and path planning. The invention provides a road closure detection method based on multi-feature fusion, aiming at the phenomenon that a road section cannot pass through temporarily or for a long time due to various factors in daily life and an electronic map is not updated timely so that navigation is inaccurate.
In order to achieve the technical purpose, the technical scheme adopted by the invention is as follows:
a road closure detection method based on multi-feature fusion comprises two stages of off-line road closure feature modeling and on-line road closure detection, and specifically comprises the following steps:
s1, an off-line road closure characteristic modeling stage: and carrying out grid division on the area to be detected, and respectively establishing grid indexes for the road network data and the motor vehicle track data.
And S2, obtaining a road section sequence passed by the motor vehicle track data through map matching to extract the traffic correlation strength between the road sections.
S3: and extracting the traffic flow of each grid unit, the strong-correlation road steering flow sequence of each road section and the vehicle turning frequency sequence in a local area (the road section and the adjacent road) in each time window tf according to the historical track data.
S4: and (3) an online detection stage: and (4) providing a combined model of a Convolutional Neural Network (CNN) and a long-short term memory neural network (LSTM) according to the historical traffic flow sequence of each grid obtained in the step (S3) to obtain a predicted value of the traffic flow of each grid at the current moment, and screening out a closed candidate grid according to the fact that the difference between the predicted value and the true value of the traffic flow is greater than a set threshold value.
S5: for the candidate closed grids obtained in the step S4, a combined model of a graph convolution neural network (GCN) and a long-short term memory neural network (LSTM) is used to obtain the road turning flow strongly associated with each road segment and the predicted value of the turning frequency in a local area (i.e., each road segment itself and its adjacent roads) at the current moment, and the closed road segments are identified according to the sudden decrease of the turning flow or the sharp increase of the turning frequency.
And S6, judging the road closure type by combining motor vehicle track yaw detection and single vehicle track data based on the closed road section identified by the S5.
In order to optimize the technical scheme, the specific measures adopted further comprise:
the step S1 specifically includes:
in the off-line stage, because the closed behaviors of roads are not isolated, the traffic conditions of surrounding adjacent road sections can be influenced, in order to better capture the traffic behavior changes of the closed road sections and the adjacent road sections thereof and reduce the calculation expense of map matching, the areas to be detected are averagely divided into grids, and grid indexes are established for road network data and motor vehicle track data. Considering that the average length of the road section and the larger grid unit size neglect the fluctuation of the local traffic flow, the side length of the grid is set to be l, wherein l is 500m in the invention.
The step S2 specifically includes:
when a road is closed, the number of vehicles turning to the road section on the upstream road is reduced, that is, the traffic conditions of adjacent road sections in the road network are influenced mutually, and the influence degrees are different, so that the correlation strength between roads needs to be extracted to highlight the influence of the traffic behaviors between the correlated roads.
The method comprises the steps of obtaining a road section sequence (road section upstream and downstream relation) passed by each track by adopting a map matching algorithm based on hidden Markov, and evaluating the association strength between roads according to the frequency of the upstream and downstream relation, wherein the higher the frequency is, the higher the association strength is. The most relevant road is called a strongly associated road, with a respective strongly associated road in each direction for a bi-directional road.
The step S3 specifically includes:
based on the trajectory data with the grid index established in S1, the total number of vehicles in each grid in each time window tf can be obtained through the trajectory data as the traffic flow.
Based on the link sequence through which each track passes and the correlation strength between the links obtained in S2, the strong-correlation road steering flow sequence of each link may be obtained according to the number of times that the road with the highest correlation strength appears as the strong-correlation road and the upstream and downstream sequence pair of 'the strong-correlation road-the current road' in the link sequence corresponding to the track.
For the extraction of the U-turn track, firstly traversing the track, and when the difference value between the track point and the preorder track point is larger than a preset threshold thangle1Judging the track point as a turning point, clustering all the turning points by adopting a DBSCAN algorithm, extracting a convex hull form of a clustering result, and finally judging whether the direction difference value of a track entry point and a track exit point in a convex hull range exceeds a preset threshold thangle2Identifying a turning track and recording the turning frequency of the vehicles in the local area in each time window tf (tf is 1h, th in the invention)angle1＝35°，thangle2＝150°)。
The step S4 specifically includes:
in the on-line detection stage, the prediction of a closed characteristic value is carried out by combining the time-space dependence of traffic data, wherein the closed characteristic value comprises flow, steering flow and turning frequency; considering that traffic data has certain periodicity and tendency (time correlation), such as early peak and late peak, and once a road closure event occurs, the road closure event affects a certain surrounding area (space correlation), only comparing historical data at the same time or comparing traffic conditions of adjacent road sections easily misjudges the closure event, and a method for comparing difference between a predicted value and a true value and combining space-time dependence of the traffic data is needed to detect the road section with the difference degree larger than a set threshold value as a closed road section. In order to reduce the time overhead of map matching, a method of screening candidate closed grids and then detecting closed road sections in the grids is adopted. Specifically, the hourly traffic flow of each grid is converted into a set of two-dimensional matrices as an input of a model, and since the grids are of a regular spatial structure (euclidean structure) with translational invariance, a multi-layer Convolutional Neural Network (CNN) is adopted to capture the inter-grid euclidean spatial features. The transition per layer can be defined as:
Figure BDA0003143344460000031
wherein
Figure BDA0003143344460000032
A traffic matrix representing the kth layer of the i-grid within the t time window,
Figure BDA0003143344460000033
flow matrix representing the k-1 th layer of i grid in t time window, is convolution operator, f is ReLU activation function, WkWeight parameter, b, that the model needs to learnkRefers to the bias parameter.
After spatial features are extracted through the convolutional neural network CNN, a feature matrix with spatial attributes output by the convolutional neural network CNN is processed through a fully-connected neural network layer and converted into a one-dimensional feature matrix
Figure BDA0003143344460000034
And thus input into a long-short term memory neural network (LSTM) to capture the temporal correlations, which refer to the periodicity and trend of traffic data.
The following formula is an architectural representation of LSTM:
Figure BDA0003143344460000041
Figure BDA0003143344460000042
Figure BDA0003143344460000043
Figure BDA0003143344460000044
Figure BDA0003143344460000045
Figure BDA0003143344460000046
w, U, b is the parameter of learning, I, F, O is the input layer, the hidden layer and the output layer respectively, which is used to control whether the information of the preamble network layer is reserved. Operator σ,
Figure BDA0003143344460000047
Respectively an activation function and a hadamard product,
Figure BDA0003143344460000048
representing the output under the i-grid t time window,
Figure BDA0003143344460000049
is a one-dimensional feature matrix with spatial properties,
Figure BDA00031433444600000410
the memory unit maintains the accumulation degree of the preamble information and the retention degree of the current information.
And finally, inputting the feature matrix combining the space attribute and the time attribute into a fully-connected neural network layer to obtain a flow predicted value of each grid at the current moment. According to the real-time track flow, the traffic flow of each grid at the current moment is counted, the predicted value is compared with the actual value, and if the difference value between the predicted value and the actual value exceeds a preset threshold thαThen the grid is determined as a candidate closed grid (in the present invention)
Figure BDA00031433444600000411
dgTo detect the trace density of the grid, avg (d)g) The average trace density for all grids).
The step S5 specifically includes:
based on the candidate closed grids obtained in S6, in view of the fact that the road-closing event may affect the traffic behavior of the area around the candidate closed grids and the information loss of the grid margin, each candidate closed grid and the adjacent grid are extracted to form a road-closing detection area. Because the closed road section has obvious traffic characteristic change: the turning flow of the strongly associated road is reduced and the frequency of turning around in the local area is increased, and the road section closing characteristics at the current moment are firstly obtained according to the method of S3. And then extracting the history sequence of the strongly-associated road turning flow and the turning frequency of each road section in the detection area based on the result obtained in the step S4. The road network is characterized in a non-Euclidean structure diagram form by adopting a combined prediction model of a graph convolution neural network (GCN) and a long-short term memory neural network (LSTM). And establishing an adjacency matrix A of the road section according to the upstream and downstream relations of the road section, forming a characteristic matrix Y by each node according to the steering flow of the strongly related road and the turning frequency in the local area, and inputting the characteristic matrix Y into a GCN graph convolution neural network layer. The propagation mode between the GCN graph convolution neural network layers is as follows:
Figure BDA00031433444600000412
wherein
Figure BDA00031433444600000413
I is a unit matrix of the image data,
Figure BDA00031433444600000414
is that
Figure BDA00031433444600000415
Degree matrix of (H)1Is the output of layer 1, W1All the weight parameters of the current layer are included and σ is a non-linear activation function.
Then, a two-layer GCN graph convolution neural network is constructed, and the activation functions are softmax and ReLu:
Figure BDA00031433444600000416
wherein
Figure BDA0003143344460000051
W(0)Is a weight matrix from input layer to hidden layer, W(1)Is a weight matrix from hidden layer to output layer.
After the spatial topological correlation of the traffic features is extracted through GCN, the spatial topological correlation is input into the LSTM long-short term memory neural network to extract the time correlation in the same way as S4. Obtaining the steering flow of each road section in the candidate grid and the predicted value of the turn-around frequency in the local area, respectively comparing the steering flow and the predicted value with the actual value, and if the difference value between the predicted value and the actual value of the steering flow exceeds the preset threshold thβOr the difference between the real value and the predicted value of the turning frequency exceeds a preset threshold thγThen the road segment is determined to be a closed road segment. (in the invention)
Figure BDA0003143344460000052
Wherein avg (T) is the historical average value of the road steering flow strongly correlated to the current road section, avg (T)all) The historical average value of the steering flow of the strongly correlated roads of all the road sections, avg (U), is the historical average value of the turn-around frequency in the local area of the current road section, avg (U)all) As a historical average of the frequency of U-turns in the local area of all road segments).
The step S6 specifically includes:
further observation is carried out on the closed road obtained based on S5 by combining the single vehicle track data, and the closed type is divided into: the closed type motor vehicle is closed singly, and is closed in parallel and is not closed in parallel. The discrimination method is as follows:
1. totally closed and motor vehicle singly seals: and extracting the single-vehicle track data of the grid where the closed road section is located, matching the single-vehicle track data with the road sections in the grid by adopting a map matching algorithm based on hidden Markov, judging the closed road section as a totally closed road section if the closed road section is not successfully matched with any single-vehicle track, and judging the closed road section as a motor vehicle single-vehicle closed road section if the closed road section is successfully matched with the single-vehicle track.
2. Non-merging and semi-closing: based on the result obtained at S2, the bidirectional road has a strongly associated road in each direction link. In the prediction of the strong-correlation road steering flow of the target road segment at S5, if the predicted value of the strong-correlation road steering flow of one direction road segment of the bidirectional road is greater than the true value, but the predicted value of the strong-correlation road steering flow of the other direction road segment is less than the true value, the bidirectional road segment is judged to be non-merging-road semi-closed.
3. Merging and semi-sealing: one of the obvious characteristics of the merging road semi-closed is that the driving route of the vehicle at the closed position has an obvious yaw, and the original double-lane driving is changed into a single-lane driving. Firstly, extracting a road center line of a closed road, then calculating the offset of the historical track and the current track of the road section relative to the road center line, and then calculating the Wasserstein distance to obtain the difference degree of the historical track offset distribution and the current track offset distribution, if the difference degree exceeds a threshold thθJudging that the vehicle driving route on the road section has yaw, and finally judging the road section as a merging road semi-closed road (th in the invention)θ＝1.25* avg(distW) Wherein avg (dist)W) Is the historical average of the distance between the current road segment's track and its predecessor track Wasserstein).
The invention firstly adopts a method of establishing grid indexes to grid the track data and the road network data for region level detection and road section level detection, thereby improving the detection efficiency. And modeling the road closure characteristics based on a DBSCAN clustering algorithm and a hidden Markov map matching algorithm, and extracting the correlation strength, steering flow and turn-around frequency between road segments. And then, by combining the time-space correlation of the trajectory data, predicting the closed characteristic by adopting a characteristic learning neural network which captures two dependencies of time and space, detecting a closed road according to the difference degree of a predicted value and a true value, finally detecting the yaw degree of the road by detecting the distribution difference degree of the trajectory points on the road, and further identifying the closed types by combining the flow of the single vehicle.
Where the predictive model may be replaced depending on the features considered. For example, only considering the time series relationship, the prediction can be performed by using traditional mathematical models such as ARIMA, EWMA and the like, or by using a recurrent neural network or a long-short term memory neural network alone.
The invention has the following beneficial effects:
1. the method and the device consider the detection of the closed road by combining a plurality of characteristics of the grid motor vehicle flow, the road section steering flow and the road section turning frequency, avoid the misjudgment of traffic jam and detect various types of closed roads except the totally closed road compared with the prior art.
2. The invention further identifies the type of the detected closed road by combining track yaw and road section single traffic flow, and compared with the prior art, the invention classifies the closed road and provides better support for navigation service.
3. The invention considers the real-time performance of detection, provides a two-step detection method by combining the detection of the regional level and the detection of the road section level, and improves the detection efficiency while ensuring the detection precision compared with the prior art.
4. Based on motor vehicle track data, the method for detecting the closed road combines a plurality of characteristics such as traffic flow, steering flow of a strongly-associated road, turn-around frequency in a local area and the like for the first time, and is different from the prior art that only single characteristic of the traffic flow is considered, and a closed road section cannot be accurately positioned.
5. Considering that different types of closure events have different influence degrees and influence ranges on traffic states, the method accurately classifies the road closure types by combining track abnormal yaw and single-vehicle track data for the first time so as to further serve the reasonable planning of a travel route, is different from the prior art that only the road is detected in a totally closed mode, and cannot cover the identification of multiple closure types under daily conditions.
6. The method is different from the prior art that the search of the regional level or the road section level is directly carried out, and the detection timeliness is improved while the detection precision is ensured by combining the two steps.
Drawings
Fig. 1 is a schematic flow chart of a road closure detection method with multi-feature fusion according to the present invention.
FIG. 2 is a model diagram of an online road closure identification stage according to the present invention.
Detailed Description
The invention is further described in detail with reference to the following specific examples and the accompanying drawings. The procedures, conditions, experimental methods and the like for carrying out the present invention are general knowledge and common general knowledge in the art except for the contents specifically mentioned below, and the present invention is not particularly limited. The technical solution of the present invention is further described in detail below with reference to the accompanying drawings.
The invention discloses a closed road detection method based on multi-feature fusion, and as shown in figure 1, the method comprises an off-line stage and an on-line stage. In the off-line road closed characteristic modeling stage, firstly, gridding is carried out on the area to be detected, then grid indexes are built on the track data and the road network data, then the correlation strength between roads is extracted, the traffic flow of each grid is extracted based on historical data, the strong correlation road steering flow of each road section and the turn-around frequency in a local area are extracted. In the stage of on-line road closure detection, as shown in fig. 2, the invention provides two strategies of candidate closed grid screening and road closure detection. In consideration of the space-time dependence of the road closure features, the invention adopts a deep learning technology to predict the value of the closure features of each grid (or road) in the current time period. A closed candidate mesh (or road closure) is identified by calculating the difference between the predicted and true values for each mesh (or road) and comparing them to a predefined threshold. Finally, the type of the closed road is further determined to better support the navigation service.
As shown in fig. 1, the invention relates to a road closure detection method based on multi-feature fusion, which comprises the following steps:
s1, an off-line road closure characteristic modeling stage: and gridding the detection area and establishing a grid index for the road network data and the track data.
In an embodiment, step S1 specifically includes:
in the off-line stage, because the closed behaviors of roads are not isolated, the traffic conditions of surrounding road sections can be influenced, in order to better capture the traffic behavior changes of the closed road sections and the adjacent road sections thereof and reduce the map matching overhead, the areas to be detected are averagely divided into grids, and grid indexes are established for road network data and track data. Considering that the average length of the road section and an excessively large grid may ignore the fluctuation of the local traffic volume, the side length of the grid is set to l, and l is 500 m.
S2: and obtaining a road section sequence passed by the motor vehicle track data based on map matching so as to extract the correlation strength among the road sections in the road network.
In an embodiment, step S2 specifically includes:
when a road is closed, the number of vehicles turning to the road segment on the upstream road will be reduced, that is, adjacent road segments in the road network will influence traffic behaviors mutually, and the influence degrees are different, and the correlation strength between each road needs to be extracted to focus on the influence of the traffic behaviors between the strongly correlated roads. The method comprises the steps of obtaining a road section sequence (road section upstream and downstream relation) passed by each track by adopting a map matching algorithm based on hidden Markov, and evaluating the association strength between roads according to the frequency of the upstream and downstream relation, wherein the higher the frequency is, the higher the association strength is. The most relevant road is called a strongly associated road, and the bidirectional road has a respective strongly associated road in each direction.
S3: and extracting the traffic flow of each grid unit in each time window, the strong associated road steering flow of each road section and the turn-around frequency in a local area (self and adjacent roads) according to historical data.
In an embodiment, step S3 specifically includes:
based on the track data with the grid indexes established and obtained in S1, the vehicle number in the track data is used for obtaining the vehicle of each grid under each time window tfAnd (4) flow rate. And then, based on the link sequence passed by each track and the correlation strength among the links obtained in the step S2, a strong correlation road steering flow sequence of each link is obtained. Finally, a U-turn track needs to be extracted, the track is traversed firstly, and when the difference value between the track point direction and the previous track point direction is larger than a threshold thangle1Judging the track point as a turning point, clustering all the turning points by a DBSCAN algorithm, extracting a convex hull form of a clustering result, and finally judging whether the direction difference value of a track entry point and a track exit point in a convex hull range is greater than a threshold thangle2To determine the turning track and record the turning frequency of each road section per time window (tf is 1h, th in the invention)angle1＝35°，thangle2＝150°)。
S4: and (3) an online detection stage: and (4) obtaining a predicted value of the traffic flow of each grid at the current moment based on a Convolutional Neural Network (CNN) and a long-short term memory neural network (LSTM) model from the historical traffic flow sequence of each grid obtained in the step (S3), and screening out a closed candidate grid according to the characteristic that the predicted value is obviously greater than a true value.
In an embodiment, step S4 specifically includes:
in the detection stage, the traffic data has certain periodicity and tendency (time correlation), such as early peak and late peak, and once a road closure event occurs, the traffic data can affect a certain surrounding area (space correlation), so that misjudgment is easy to occur if the historical data at the same time are compared in a single longitudinal direction or the surrounding conditions are compared in a transverse direction, so that in the road closure detection stage, the time-space dependence of the traffic data is considered, a method for comparing a predicted value and a real value is adopted, and if the difference value is too large, the traffic data is judged to be closed. Meanwhile, in order to reduce the time overhead of subsequent map matching, candidate closed grids are screened out first, and then the closed condition of the road sections in the grids is further detected. Specifically, the hourly traffic flow of each grid is converted into a set of two-dimensional matrices as an input of a model, and since the grids are regular spatial structures (euclidean structures) with translational invariance, a Convolutional Neural Network (CNN) with multiple layers is adopted to capture the inter-grid euclidean spatial features. The transition per layer can be defined as:
Figure BDA0003143344460000081
wherein
Figure BDA0003143344460000082
The flow matrix of the k-th layer under the time window of i grid t is represented by a convolution operator, f is a ReLU activation function, and WkWeight parameter, b, that the model needs to learnkRefers to the bias parameter.
After the spatial features are extracted through the CNN, the feature matrix with the spatial attributes output by the CNN is processed through a full-connection neural network layer and converted into a one-dimensional feature matrix
Figure BDA0003143344460000083
And thus input into a long-short term memory neural network (LSTM) to capture temporal correlations. The following formula is an architectural representation of LSTM:
Figure BDA0003143344460000084
Figure BDA0003143344460000085
Figure BDA0003143344460000086
Figure BDA0003143344460000087
Figure BDA0003143344460000088
Figure BDA0003143344460000089
w, U, b is the parameter of learning, I, F, O is the input layer, the hidden layer and the output layer respectively, which is used to control whether the information of the preamble network layer is reserved. Operator σ,
Figure BDA00031433444600000810
Respectively an activation function and a hadamard product,
Figure BDA00031433444600000811
representing the output under the i-grid t time window.
And finally, inputting the characteristic matrix combining the space and time attributes into a fully-connected neural network layer to obtain a flow predicted value of each grid at the current moment. According to the real-time track flow, the traffic flow of each grid at the current moment is counted, the predicted value is compared with the actual value, and if the actual value is smaller than the predicted value and exceeds the threshold thαThen the grid is determined as a candidate closed grid (in the present invention)
Figure BDA0003143344460000091
dgTo detect the trace density of the grid, avg (d)g) The average trace density for all grids).
S5: and (4) obtaining predicted values of the steering flow of the road with strong association of each road section and the turn-around frequency in the local area at the current moment by using a graph convolutional neural network (GCN) and a long-short term memory neural network (LSTM) model based on the grid possibly with road closure obtained in the S4, and screening the closed road section according to the sharp characteristic of sudden reduction or turn-around frequency of the steering flow.
In an embodiment, step S5 specifically includes:
based on the candidate closed grids obtained in S6, first, considering that a closed circuit event generally affects traffic behaviors in an area near the closed circuit event and information loss of grid margins, each candidate closed grid and its neighboring grids are extracted to form a closed circuit detection area. Because of the two obvious traffic characteristic changes of the closed road section: the turning flow of the strongly associated road is reduced and the frequency of turning around in the local area is increased, and the road section closing characteristics at the current moment are firstly obtained according to the method of S3. And then extracting the history sequence of the strongly-associated road turning flow and the turning frequency of each road section in the detection area based on the result obtained in the step S4. The method adopts a combined prediction model of graph convolutional neural network (GCN) and LSTM, and the road network is characterized in a graph (non-Euclidean structure) form. And establishing an adjacency matrix A of the road section according to the upstream and downstream relations of the road section, forming a characteristic matrix Y by each node according to the steering flow of the strongly related road and the turning frequency in the local area, and inputting the characteristic matrix Y into the GCN network layer. The transmission mode between GCN layers is as follows:
Figure BDA0003143344460000092
wherein
Figure BDA0003143344460000093
I is a unit matrix of the image data,
Figure BDA0003143344460000094
is like
Figure BDA0003143344460000095
Degree matrix of (H)1Is the output of layer 1, WlAll the weight parameters of the current layer are included and σ is a non-linear activation function.
Then, a two-layer GCN is constructed, and the activation functions are softmax and ReLu:
Figure BDA0003143344460000096
wherein
Figure BDA0003143344460000097
W(0)Is a weight matrix from input layer to hidden layer, W(1)Is a weight matrix from hidden layer to output layer.
Space for extracting traffic characteristics through GCNAfter the topological correlation, the time correlation is extracted by inputting the extracted time correlation into the LSTM in the same manner as S4. Obtaining the steering flow of the strongly-associated road of each road section in the candidate grid and the predicted value of the turn-around frequency in the local area, respectively comparing the steering flow and the predicted value with the actual value, and if the actual value of the steering flow is smaller than the predicted value and exceeds the threshold thβOr the real value of the turn-around frequency is larger than the predicted value and exceeds the threshold thγThen the road segment is judged as a closed road segment (in the invention)
Figure BDA00031433444600000910
Figure BDA0003143344460000099
Wherein avg (T) is the historical average value of the steering flow of the strongly-associated road of the current road section, avg (T)all) The historical average value of the steering flow of the strongly-associated roads of all the road sections, avg (U), is the historical average value of the turn-around frequency in the local area of the current road section, avg (U)all) As a historical average of the frequency of U-turns in the local area of all road segments).
And S6, obtaining a closed road section based on S5, and judging the closed type by combining motor vehicle track yaw detection and bicycle track data.
In an embodiment, step S6 specifically includes:
further observation is carried out on the closed road obtained based on S5 by combining the single vehicle track data, and the closed type is divided into: the closed type motor vehicle is closed singly, and is closed in parallel and is not closed in parallel. The discrimination method is as follows:
1. totally closed and motor vehicle singly seals: and extracting the single-vehicle track data of the grid where the closed road section is located, matching the single-vehicle track data with the road sections in the grid by adopting a map matching algorithm based on hidden Markov, judging the closed road section as a totally closed road section if the closed road section is not successfully matched with any single-vehicle track, and judging the closed road section as a motor vehicle single-vehicle closed road section if the closed road section is successfully matched with the single-vehicle track.
2. Non-merging and semi-closing: based on the result obtained at S2, the bidirectional road has a strongly associated road in each direction link. In the prediction of the steering flow of the strongly-associated road at the target road section in the S5 step, if the predicted value of the steering flow of the strongly-associated road at one direction road section of the bidirectional road is greater than the true value, but the predicted value of the steering flow of the strongly-associated road at the other direction road section of the bidirectional road is less than the true value, the bidirectional road is judged to be non-merging and semi-closed.
3. Merging and semi-sealing: one of the obvious characteristics of the merging road semi-closed is that the driving route of the vehicle at the closed position has an obvious yaw, and the original double-lane driving is changed into a single-lane driving. Firstly, extracting a road center line of a closed road, then calculating the offset of the historical track and the current track of the road section relative to the road center line, and then calculating the Wasserstein distance to obtain the difference degree of the historical track offset distribution and the current track offset distribution, if the difference degree exceeds a threshold thθJudging that the vehicle driving route on the road section has yaw, and finally judging the road section as a merging road semi-closed road (th in the invention)θ＝1.25* avg(distW) Wherein avg (dist)W) Is the historical average of the distance between the current road segment's track and its predecessor track Wasserstein).
The method comprehensively adopts the technologies of data grid indexing, DBSCAN clustering technology, map matching based on hidden Markov, convolutional neural network, long-short term memory neural network, graph convolutional neural network, Wasserstein distance and the like, forms a road closure detection method based on multi-feature fusion, can accurately position the closure and the closure type of the road through track data mining, and thus provides effective help for navigation broadcasting and path planning.
The above is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above-mentioned examples, and any technical solutions that fall under the idea of the present invention fall within the protection scope of the present invention. It will be appreciated by those skilled in the art that changes may be made in these embodiments without departing from the principles of the invention, and that such changes and advantages can be understood by those skilled in the art as is encompassed within the appended claims.

Claims (5)

1. A road closure detection method based on multi-feature fusion is characterized by comprising the following steps:
s1: an off-line road closed characteristic modeling stage: gridding the area to be detected and respectively establishing grid indexes for the road network data and the track data;
s2: obtaining a road section sequence through which motor vehicle track data pass by map matching so as to extract the traffic correlation strength between road sections in a road network; the step S2 specifically includes:
when the roads are closed, extracting the traffic association strength between the roads, and determining the influence degree of the traffic behaviors between the strong associated roads; the strongly associated road means: obtaining a road section sequence passed by each track by adopting a map matching algorithm based on hidden Markov, namely obtaining an upstream-downstream relation of the road sections, and evaluating the association strength between roads according to the frequency of the upstream-downstream relation, wherein the higher the frequency is, the higher the association strength is, and the strongest correlation road is, namely, the strong correlation road; for a bidirectional road, there are strong associated roads in different directions;
s3: based on historical data, extracting the traffic flow of each grid unit in each time window tf, the strong correlation road steering flow sequence of each road section and the vehicle turning frequency sequence in a local area; the step S3 specifically includes:
the local area refers to each road section and adjacent roads;
the traffic flow of each grid cell in each time window tf is obtained by the sum of the number of vehicles in the trajectory data in the grid index established in S1;
the strongly associated road steering flow sequence of each road section is obtained by taking the road with the maximum association strength as the strongly associated road and the number of times of the upstream and downstream sequence pairs of the strongly associated road-the current road' in the road section sequence corresponding to the track according to the association strength between the road section sequence through which each piece of motor vehicle track data passes and the road sections obtained in the step S2;
the vehicle turning frequency sequence in the local area firstly traverses the track, and when a track point is equal to the previous track pointThe difference in direction between is greater than a preset threshold thangle1Judging the track point as a turning point, clustering all the turning points by adopting a DBSCAN algorithm, simultaneously extracting a convex hull form of a clustering result, and finally judging whether the direction difference value of a track entry point and a track exit point in a convex hull range is greater than a set threshold th according to whetherangle2Recognizing a turning track; recording the turning frequency of the vehicles in the local area in each time window tf;
the tf is 1h, thangle1＝35°，thangle2＝150°；
S4: and (3) an online detection stage: obtaining a predicted value of the traffic flow of each grid at the current moment by adopting a combined model of a convolutional neural network and a long-short term memory neural network according to the historical traffic flow sequence of each grid obtained in the step S3, and taking the grid with the difference between the predicted value of the traffic flow and the true value exceeding a threshold value as a closed candidate grid;
s5: for the closed candidate grids obtained in the S4, obtaining the steering flow of the road strongly associated with each road section and the predicted value of the turning frequency in the local area at the current moment by using a combined model of a graph convolution neural network and a long-short term memory neural network, and screening the closed road sections according to the sudden reduction of the steering flow or the sharp increase characteristic of the turning frequency;
s6: and obtaining a closed road section based on S5, and judging the type of road closure by combining motor vehicle track yaw detection and single vehicle track data.
2. The method for detecting road closure based on multi-feature fusion as claimed in claim 1, wherein the step S1 specifically includes:
in the off-line stage, in order to capture traffic behavior changes of a closed road section and an adjacent road section thereof and reduce the calculation overhead in the map matching process, an area to be detected is averagely divided into grids, grid indexes are established for road network data and track data, the side length of each grid is set to be l, and the l is 500 m.
3. The method for detecting road closure based on multi-feature fusion as claimed in claim 1, wherein the step S4 specifically includes:
in the on-line detection stage, the prediction of a closed characteristic value is carried out by combining the time-space dependence of traffic data, wherein the closed characteristic value comprises flow, steering flow and turning frequency; detecting the road sections with the difference degrees larger than a set threshold value as closed road sections by comparing the difference between the predicted value and the true value; in the detection process, in order to reduce the time overhead of map matching, a method of screening candidate closed grids and then detecting closed road sections in the grids is adopted; specifically, the small-level traffic flow of each grid is converted into a group of two-dimensional matrixes to serve as input of a model, and because the grids are of a regular space structure with translation invariance, namely a Euclidean structure, multi-layer convolutional neural networks are adopted to capture Euclidean space characteristics among the grids; the transition for each layer is defined as:
Figure FDA0003489614330000021
wherein
Figure FDA0003489614330000022
Representing the traffic matrix at the kth layer of the i-grid under the t time window,
Figure FDA0003489614330000023
flow matrix representing the k-1 th layer of i grid in t time window, is convolution operator, f is ReLU activation function, WkWeight parameter, b, that the model needs to learnkRefers to a bias parameter;
after spatial features are extracted through the convolutional neural network, a feature matrix with spatial attributes output by the convolutional neural network is processed through a fully-connected neural network layer and converted into a one-dimensional feature matrix
Figure FDA0003489614330000024
And inputting the data into a long-short term memory neural network to capture the time correlation, wherein the time correlation refers to the periodicity and trend of traffic data, and the following formula is an LSTM architecture tableThe following steps:
Figure FDA0003489614330000025
Figure FDA0003489614330000026
Figure FDA0003489614330000027
Figure FDA0003489614330000028
Figure FDA0003489614330000029
Figure FDA00034896143300000210
w, U, b is a learning parameter, I, F, O is an input layer, a hidden layer and an output layer respectively, and is used for controlling whether information of a preamble network layer is reserved or not; operator σ,
Figure FDA00034896143300000211
Respectively an activation function and a hadamard product,
Figure FDA00034896143300000212
representing the output of the i grid within the t time window;
finally, inputting the feature matrix combining the space and time attributes into a fully-connected neural network layer to obtain a flow predicted value of each grid at the current moment; according to the real-time track flow, the traffic flow of each grid at the current moment is countedComparing the predicted value with the real value, if the difference between the real value and the predicted value exceeds the preset threshold thαThen the grid is determined as a candidate closed grid, said
Figure FDA0003489614330000031
dgTo detect the trace density of the grid, avg (d)g) Is the average trace density of all grids.
4. The method for detecting road closure based on multi-feature fusion as claimed in claim 1, wherein the step S5 specifically includes:
extracting each candidate closed grid and adjacent grids thereof to form a closed circuit detection area based on the candidate closed grids obtained in the step S6; because of the two obvious traffic characteristic changes of the closed road section: reducing the steering flow of the strongly associated road and increasing the turning frequency in the local area, and firstly obtaining the road section closing characteristics at the current moment according to the method of S3; then extracting a historical sequence of strongly-associated road steering flow and turning frequency of each road section in the detection area based on the result obtained in the step S4; a combined prediction model of a graph convolution neural network and a long-short term memory neural network is adopted, and a road network is represented in a non-Euclidean structure diagram form; establishing an adjacency matrix A of a road section according to the upstream and downstream relation between road sections, forming a feature matrix Y by each node according to the steering flow of the strongly-related road and the turning frequency in a local area, and inputting the feature matrix Y into a graph convolution neural network layer; the propagation mode between the layers of the graph convolution neural network is as follows:
Figure FDA0003489614330000032
wherein
Figure FDA0003489614330000033
I is a unit matrix of the image data,
Figure FDA0003489614330000034
is that
Figure FDA0003489614330000035
Degree matrix of (H)lIs the output of the l-th layer, WlAll weight parameters of the current layer are contained, and sigma is a nonlinear activation function;
then, a two-layer graph convolution neural network is constructed, and the activation functions are softmax and ReLu:
Figure FDA0003489614330000036
wherein
Figure FDA0003489614330000037
W(0)Is a weight matrix from input layer to hidden layer, W(1)Is a weight matrix from hidden layer to output layer;
after the spatial topological correlation of the traffic characteristics is extracted through the graph convolution neural network, the spatial topological correlation is input into the long-term and short-term memory neural network in the same way as the S4 to extract the time correlation; obtaining the steering flow of each road section in the candidate grid and the predicted value of the turn-around frequency in the local area, respectively comparing the steering flow with the real value, and if the difference value between the real value and the predicted value of the steering flow exceeds a preset threshold thβOr the difference between the real value and the predicted value of the turning frequency exceeds a preset threshold thγJudging the road section as a closed road section; the above-mentioned
Figure FDA0003489614330000038
Wherein avg (T) is the historical average value of the road steering flow strongly correlated to the current road section, avg (T)all) The historical average value of the steering flow of the strongly correlated roads of all the road sections, avg (U), is the historical average value of the turn-around frequency in the local area of the current road section, avg (U)all) And the historical average value of the U-turn frequency in the local area of all the road sections is obtained.
5. The method for detecting road closure based on multi-feature fusion as claimed in claim 1, wherein the step S6 specifically includes:
further observation is carried out on the closed road obtained based on S5 by combining with the single-vehicle track data, and road closure types are divided into five types, namely closed type, single-vehicle closure, merging road semi-closure and non-merging road semi-closure; the discrimination method is as follows:
(1) totally closed and motor vehicle singly seals: extracting the single-vehicle track data of a grid where a closed road section is located, matching the single-vehicle track data with road sections in the grid by adopting a map matching algorithm based on hidden Markov, judging the closed road section as a fully closed road section if the closed road section has no matched single-vehicle track, and judging the closed road section as a single-vehicle closed road section if the closed road section has the matched single-vehicle track;
(2) non-merging and semi-closing: based on the result obtained at S2, the bidirectional road has a strongly associated road in each direction section; in the prediction of the strong-correlation road steering flow of the target road section at S5, if the predicted value of the strong-correlation road steering flow of one direction road section of the bidirectional road is greater than the true value, but the predicted value of the strong-correlation road steering flow of the other direction road section is less than the true value, judging that the bidirectional road is a non-merging semi-closed event;
(3) merging and semi-sealing: one remarkable characteristic of the merging road semi-closed is that the running track of the vehicle has obvious yaw, which is caused by changing the double-lane running into the single-lane running; firstly extracting a road center line of a closed road, then calculating the offset of the historical track and the current track of the road section relative to the road center line, then calculating the Wasserstein distance to obtain the difference degree of the historical track offset distribution and the current track offset distribution, and when the difference degree exceeds a preset threshold thθIf yes, judging that the vehicle driving route on the road section has yaw, and judging the road section as a merging road semi-closed road, thθ＝1.25*avg(distW) Wherein avg (dist)W) Is the historical average value of the distance between the track of the current road section and the preamble track thereof Wasserstein.
CN202110747124.6A 2021-07-01 2021-07-01 Road closure detection method based on multi-feature fusion Active CN113554869B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110747124.6A CN113554869B (en) 2021-07-01 2021-07-01 Road closure detection method based on multi-feature fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110747124.6A CN113554869B (en) 2021-07-01 2021-07-01 Road closure detection method based on multi-feature fusion

Publications (2)

Publication Number Publication Date
CN113554869A CN113554869A (en) 2021-10-26
CN113554869B true CN113554869B (en) 2022-04-05

Family

ID=78131246

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110747124.6A Active CN113554869B (en) 2021-07-01 2021-07-01 Road closure detection method based on multi-feature fusion

Country Status (1)

Country Link
CN (1) CN113554869B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114267173B (en) * 2021-12-13 2023-08-18 河北雄安京德高速公路有限公司 Multisource data fusion method, device and equipment for space-time characteristics of expressway

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101281038A (en) * 2008-05-27 2008-10-08 凯立德欣技术（深圳）有限公司 Periphery searching device and method on closed road
CN101571993A (en) * 2008-04-30 2009-11-04 奥城同立科技开发（北京）有限公司 Traffic control method for plane exits of closed side roads
CN108442198A (en) * 2018-05-21 2018-08-24 江苏苏科建设项目管理有限公司 A kind of automated driving system based on blocked road
US10497256B1 (en) * 2018-07-26 2019-12-03 Here Global B.V. Method, apparatus, and system for automatic evaluation of road closure reports

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103235933B (en) * 2013-04-15 2016-08-03 东南大学 A kind of vehicle abnormality behavioral value method based on HMM
CN105139682B (en) * 2015-07-27 2017-11-10 福建工程学院 The confirmation method and system of a kind of blocked road
CN105206075B (en) * 2015-08-12 2018-01-12 合肥工业大学 A kind of traffic information prompt system in urban road temporary enclosed section
CN105389978B (en) * 2015-09-15 2017-12-05 安徽省综合交通研究院股份有限公司 Close through street monitoring system and monitoring data processing method
GB201711408D0 (en) * 2016-12-30 2017-08-30 Maxu Tech Inc Early entry
CN108319895B (en) * 2017-12-29 2021-09-17 百度在线网络技术（北京）有限公司 Method and device for identifying intersection in electronic map
CN109598930B (en) * 2018-11-27 2021-05-14 上海炬宏信息技术有限公司 Automatic detect overhead closed system
CN110689719B (en) * 2019-05-31 2021-01-19 北京嘀嘀无限科技发展有限公司 System and method for identifying closed road sections
CN110444013B (en) * 2019-06-28 2021-01-05 安徽四创电子股份有限公司 Traffic monitoring system and method for closed road section
CN111931905B (en) * 2020-07-13 2024-05-14 江苏大学 Graph convolution neural network model and vehicle track prediction method using same

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101571993A (en) * 2008-04-30 2009-11-04 奥城同立科技开发（北京）有限公司 Traffic control method for plane exits of closed side roads
CN101281038A (en) * 2008-05-27 2008-10-08 凯立德欣技术（深圳）有限公司 Periphery searching device and method on closed road
CN108442198A (en) * 2018-05-21 2018-08-24 江苏苏科建设项目管理有限公司 A kind of automated driving system based on blocked road
US10497256B1 (en) * 2018-07-26 2019-12-03 Here Global B.V. Method, apparatus, and system for automatic evaluation of road closure reports

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
Efficient anomaly monitoring over moving object trajectory streams;Bu YY, Chen L, Fu AWC, Liu DW;《In: Proc. of the SIGKDD》;20091231;全文 *

Also Published As

Publication number Publication date
CN113554869A (en) 2021-10-26

Similar Documents

Publication Publication Date Title
CN110164128B (en) City-level intelligent traffic simulation system
CN109754597B (en) Urban road regional congestion regulation and control strategy recommendation system and method
CN110147321B (en) Software network-based method for identifying defect high-risk module
Qi et al. A Hidden Markov Model for short term prediction of traffic conditions on freeways
CN111310583A (en) Vehicle abnormal behavior identification method based on improved long-term and short-term memory network
CN111882869B (en) Deep learning traffic flow prediction method considering adverse weather
CN108345666A (en) A kind of vehicle abnormality track-detecting method based on time-space isolated point
CN113704956A (en) Urban road online microscopic simulation method and system based on digital twin technology
CN102081846A (en) Expressway charge data track matching based traffic state recognition method
CN111210621B (en) Signal green wave coordination route optimization control method and system based on real-time road condition
CN111179601B (en) Tunnel traffic operation control method
CN113379099B (en) Machine learning and copula model-based highway traffic flow self-adaptive prediction method
CN112614373B (en) BiLSTM-based weekly vehicle lane change intention prediction method
Salamanis et al. Identifying patterns under both normal and abnormal traffic conditions for short-term traffic prediction
CN114267173B (en) Multisource data fusion method, device and equipment for space-time characteristics of expressway
CN113554869B (en) Road closure detection method based on multi-feature fusion
Luo et al. Grouped travel time estimation in signalized arterials using point-to-point detectors
CN117671992A (en) Intelligent bus dispatching method and system
Yuan et al. Spatiotemporal attention mechanism-based deep network for critical parameters prediction in chemical process
CN110619340A (en) Method for generating lane change rule of automatic driving automobile
CN116758762A (en) Control method based on big data
Lu et al. Shape-based vessel trajectory similarity computing and clustering: A brief review
CN116978259B (en) Ship navigation track prediction method, device and storage medium
Wang et al. PSTN: periodic spatial-temporal deep neural network for traffic condition prediction
Katariya et al. A pov-based highway vehicle trajectory dataset and prediction architecture

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