CN114511080A - Model construction method and device and abnormal track point real-time detection method - Google Patents

Model construction method and device and abnormal track point real-time detection method Download PDF

Info

Publication number
CN114511080A
CN114511080A CN202111643789.9A CN202111643789A CN114511080A CN 114511080 A CN114511080 A CN 114511080A CN 202111643789 A CN202111643789 A CN 202111643789A CN 114511080 A CN114511080 A CN 114511080A
Authority
CN
China
Prior art keywords
track
abnormal
point
vector
data
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.)
Pending
Application number
CN202111643789.9A
Other languages
Chinese (zh)
Inventor
漆梦梦
尹玉成
施忠继
阮双双
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.)
Heading Data Intelligence Co Ltd
Original Assignee
Heading Data Intelligence Co Ltd
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 Heading Data Intelligence Co Ltd filed Critical Heading Data Intelligence Co Ltd
Priority to CN202111643789.9A priority Critical patent/CN114511080A/en
Publication of CN114511080A publication Critical patent/CN114511080A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S19/00—Satellite radio beacon positioning systems; Determining position, velocity or attitude using signals transmitted by such systems
    • G01S19/01—Satellite radio beacon positioning systems transmitting time-stamped messages, e.g. GPS [Global Positioning System], GLONASS [Global Orbiting Navigation Satellite System] or GALILEO
    • G01S19/13—Receivers
    • G01S19/35—Constructional details or hardware or software details of the signal processing chain
    • G01S19/37—Hardware or software details of the signal processing chain
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Remote Sensing (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Signal Processing (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Biophysics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Biology (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Traffic Control Systems (AREA)

Abstract

The invention relates to a model construction method and device and a real-time detection method of abnormal track points. The method comprises the steps of extracting abnormal tracks in historical track data and constructing model training data by acquiring the historical track data and preprocessing the historical track data; then extracting the characteristic vector of each track point in the abnormal track, and carrying out relative position coding on each track point; and finally, taking the feature vector and the relative position code of each track point as input, and training the transform network model to obtain an abnormal track point real-time detection model. The invention adopts a neural network model based on a transformer module to model a time sequence track sequence and provides abnormal and non-abnormal labels for each track point. The accuracy and efficiency of abnormal track point detection are improved.

Description

Model construction method and device and real-time detection method of abnormal track point
Technical Field
The invention relates to the technical field of data processing, in particular to a method and a device for constructing a real-time detection model of abnormal track points in a vehicle track sequence and a real-time detection method of the abnormal track points in the vehicle track sequence.
Background
The reasons for generating the track abnormal points comprise the following aspects: GPS receiver device and network status. The device is dropped, resulting in a large time difference and distance (jumping points) between two adjacent points. Device anomalies, resulting in a sudden large deviation (drift point) in the GPS point location under successive time stamps. The device is abnormal, and a plurality of pieces of data (repeated data) are returned under the same timestamp. 2. Multipath effects in urban canyons. It is also possible that the GPS signals are received after being reflected off different obstacles, resulting in an inaccurate calculated position. And 3, the GPS signal is shielded, and the GPS signal in the tunnel is weak under the viaduct and the like.
At present, a plurality of vehicle track data preprocessing methods are available, wherein track segment abnormity judgment is similar. The method is mainly used for data analysis of travel companies and excavation of abnormal information such as detours.
The method for discovering the abnormal track points in the current track segment comprises the following steps: a method of anti-differential analytical method test based on statistical analysis and a method of manual logic verification. Among them, the method based on statistical analysis has the problem of inaccurate estimation. The method for manual logic verification needs feature modeling calculation and manual judgment, and is high in accuracy but low in efficiency.
Disclosure of Invention
The invention provides a model construction method, a model construction device and an abnormal track point real-time detection method aiming at the technical problems in the prior art. The accuracy and efficiency of abnormal track point detection are improved.
The technical scheme for solving the technical problems is as follows:
on one hand, the invention provides a method for constructing a real-time detection model of abnormal track points in a vehicle track sequence, which comprises the following steps:
acquiring historical track data, preprocessing the historical track data, extracting abnormal tracks in the historical track data, and constructing model training data;
extracting a characteristic vector of each track point in the abnormal track, and carrying out relative position coding on each track point;
taking the feature vector and the relative position code of each track point as input, and training a transform network model to obtain an abnormal track point real-time detection model; the loss function during training is shown as follows:
Figure BDA0003444481840000021
in the formula piRepresenting the probability that the ith track point in the abnormal track is predicted as an abnormal point, yiLabel the label value for the human, the outlier is 1 and the outlier is 0.
Further, the preprocessing the historical track data includes:
according to the road network grids, dividing historical track data into a plurality of track segment groups, wherein one grid corresponds to one track segment group, and one track segment group comprises a plurality of track segments;
matching the track segment with a road vector or a lane group vector in a high-precision map according to the geographic coordinates of the road network grids;
measuring the similarity between the track segment and a road vector or a lane group vector matched with the track segment by using a Frechet distance;
dividing the track segments into normal track segments and abnormal track segments according to a similarity threshold; and carrying out manual quality inspection on the abnormal track segment, and generating model training data after marking track points in the abnormal track segment.
Further, the segmenting the historical track data into a plurality of track segments according to the road network grid includes:
projecting longitude and latitude coordinates of the historical track data into plane coordinates according to a 3-degree band UTM;
and performing grid division on the road network, and segmenting the historical track data into a plurality of track segment groups by using the road network grid according to the plane coordinates.
Further, the matching the track segment with the road vector or the lane group vector in the high-precision map according to the geographic coordinates of the road network grid includes:
acquiring road attributes in a high-precision map according to the geographic coordinates of the grid, extracting road vectors on road sections without lane increase and decrease, and extracting lane vector groups on intersections or road sections with lane increase and decrease;
if the road vector is extracted according to the grid geographic coordinates, directly matching the track segment group with the road vector;
if the lane vector group is extracted according to the grid geographic coordinates, clustering is respectively carried out on the lane vector group and the track segments in the track segment group, and the lane vectors in the lane vector group are matched with the corresponding sub-track groups in the track segment group according to the direction consistency.
Further, after the track segment is matched with a road vector or a lane group vector in a high-precision map according to the geographic coordinates of the road network grid, resampling track points in the track segment according to the shape point interval of the road or the lane in the high-precision map.
Further, the dimension of the feature vector of each track point in the abnormal track is N, and at least includes: whether the distance is a standing point, an adjacent track point time interval, an adjacent track point course difference value, an adjacent track point distance and an adjacent track point distance difference value.
Further, the encoding of the relative position of each trace point includes:
adopting a fixed step delta (an overlapping area) and acquiring track points in the track fragment in a segmented manner by using a sliding window with the size of m so as to process the condition that the track points in the track fragment are different; for example, a track sequence is processed sequentially from the first sampling of a track point sequence with the serial number from 1 to m, and from the second sampling of a track sequence with the serial number from m-delta to 2 m-delta.
And (3) carrying out relative position coding on each track point, wherein the position coding function is as follows:
Figure BDA0003444481840000031
n is the characteristic vector dimension of the track point, and pos is the serial number of the track point in the sliding window;
and activating the relative position code value by utilizing a trigonometric function, wherein when pos is an even number, the relative position code value is activated by adopting a sine function, and when pos is an odd number, the relative position code value is activated by adopting a cosine function.
On the other hand, the invention also provides a device for constructing the abnormal track point real-time detection model in the vehicle track sequence, which comprises the following steps:
the preprocessing module is used for acquiring historical track data, preprocessing the historical track data, extracting abnormal tracks in the historical track data and constructing model training data;
the characteristic extraction module is used for extracting a characteristic vector of each track point in the abnormal track and carrying out relative position coding on each track point;
the training module is used for taking the feature vector and the relative position code of each track point as input, training the transformer network model and obtaining an abnormal track point real-time detection model; the loss function during training is shown as follows:
Figure BDA0003444481840000041
in the formula piRepresenting the probability that the ith track point in the abnormal track is predicted as an abnormal point, yiLabel the label value for the human, the outlier is 1 and the outlier is 0.
Based on the above, the invention also provides a real-time detection method for the abnormal track points in the vehicle track sequence, which comprises the following steps:
acquiring track data to be predicted, extracting a feature vector of each track point in the track data, and performing relative position coding on each track point;
and inputting the characteristic vector and the relative position code into a trained detection model, and performing forward reasoning to obtain the binary attribute of whether the track point is an abnormal track point, wherein the detection model is constructed by the construction method of the real-time detection model of the abnormal track point in the vehicle track sequence.
The invention has the beneficial effects that: compared with the conventional method based on statistical analysis and manual logic quality inspection, the method has the advantage that the detection rate and accuracy of abnormal points in the track sequence are remarkably improved. And the acquisition of the track abnormal point truth value is extracted in a semi-automatic manual interaction mode through an algorithm, so that the efficiency is high. Compared with the traditional statistical algorithm, the iterative optimization process is simple and fast. The phenomenon that the gourd is pressed and lifted in the traditional algorithm can not occur. The method is mainly benefited from the modeling capability of the neural network on the complex system, and has good generalization performance. At the initial stage with less data volume, when the model has errors in some scenes, only the trace marking data of failure scenes are basically required to be added, and the model is trained and updated. And the network model can realize real-time reasoning effect on a terminal computing platform, and the speed is high.
The invention can be used in a real-time system of vehicle-mounted positioning to give out warning signals of track abnormity in real time. This provides reference information to the assisted autopilot safety system to prompt the user to take over in time.
In addition, the invention can also be used for off-line track data preprocessing, thereby improving the utilization rate of the track data. For example, in urban high-precision map making and updating, a crowd-sourced bag takes map precision into consideration and adopts a direct abandoning mode for abnormal track segments. The method can position the abnormal points in the track segment for interception, thereby improving the utilization rate of data and saving crowdsourcing cost.
The method can also be used for integrating the robustness analysis of the positioning algorithm and providing support for iterative optimization of a failure scene of the positioning algorithm.
Drawings
Fig. 1 is a schematic flow chart of a method for constructing a real-time detection model of an abnormal track point in a vehicle track sequence according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a pre-processing procedure provided in an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a device for constructing a real-time detection model of an abnormal track point in a vehicle track sequence according to an embodiment of the present invention;
fig. 4 is a schematic flow chart of a method for detecting an abnormal track point in a vehicle track sequence in real time according to an embodiment of the present invention.
Detailed Description
The principles and features of this invention are described below in conjunction with the following drawings, which are set forth by way of illustration only and are not intended to limit the scope of the invention.
Fig. 1 is a schematic flow chart of a method for constructing a real-time detection model of an abnormal track point in a vehicle track sequence according to an embodiment of the present invention. As shown in fig. 1, the method comprises the steps of:
and S1, acquiring historical track data, preprocessing the historical track data, extracting abnormal tracks in the historical track data, and constructing model training data.
The specific steps, as shown in fig. 2, include the following:
s1.1, according to the road network grids, dividing historical track data into a plurality of track segment groups, wherein one grid corresponds to one track segment group, and one track segment group comprises a plurality of track segments.
Projecting longitude and latitude coordinates of the historical track data into plane coordinates according to a 3-degree band UTM;
the road network is divided into grids, and a certain overlapping area exists between two adjacent grids. And according to the plane coordinates, segmenting the historical track data into a plurality of track segment groups by utilizing the road network grids.
S1.2, matching the track segment with a road vector or a lane group vector in a high-precision map according to the geographic coordinates of the road network grids.
Obtaining road attributes in a high-precision map according to the geographic coordinates of the gridExtracting road vector roadvectors from road sections without lane increase and decrease, and extracting lane vector group road vectors from intersections or road sections with lane increase and decrease; group Lane vectors ═ laneVector1，laneVector2，…,laneVectorm}。
If the road vector is extracted according to the grid geographic coordinates, directly matching the track segment group with the road vector;
if the lane vector group is extracted according to the grid geographic coordinates, clustering is respectively carried out on the lane vector group and the track segments in the track segment group, and the lane vectors in the lane vector group are matched with the corresponding sub-track groups in the track segment group according to the direction consistency.
Here, the track segments are grouped for Clustering purposes using a Density-Based Clustering algorithm DBCAN (Density-Based Spatial Clustering of Applications with Noise). DBSCAN hyper-parameter determination: and obtaining the minimum number min _ sample of the subclass track points obtained by clustering and the clustering radius eps by a range searching mode. The number m of target classes of the clustering algorithm is known, i.e. the number of lane vectors in the groupLaneVectors. And determining the value of the hyper-parameter through the known prior knowledge of the target class.
S1.3, measuring the similarity between the track segment and the matched road vector or lane group vector by using a Frechet distance.
S1.4, dividing the track segment into a normal track segment and an abnormal track segment according to a similarity threshold; the track segments with large distances are abnormal track segments, and the track segments with small distances are normal track segments. And carrying out manual quality inspection on the abnormal track segment, and generating model training data after marking track points in the abnormal track segment.
And manually marking the abnormal track points in the track segment in a visual mode. The flag bit of the abnormal track point is 1, and the flag bit of the normal track point is 0. An abnormal track segment corresponds to an 0/1 tag sequence, and the sequence length is equal to the track sequence length.
And S2, extracting the feature vector of each track point in the abnormal track, and carrying out relative position coding on each track point.
Each track point comprises a plurality of dimensional information, longitude and latitude, elevation, three-axis acceleration, three-axis angular velocity and three-axis velocity information. On the basis of these basic sensor data information the following dimensions of information are calculated: whether the point is a standing point (a point with the speed of 0), the time interval of adjacent track points, the course difference value of the adjacent track points, the distance between the adjacent track points, the distance difference value of the adjacent track points and the like. And setting the dimensionality of the feature vector of each track point as N.
The relative position coding is carried out on each track point, and the method comprises the following steps:
adopting a sliding window with fixed step length and the size of m to process the condition that track points in the track section are different;
since each point can encode the absolute position of the trace points if the time stamp is directly used as the feature input, it is difficult to learn the relative position between the trace points. Therefore, each trace point needs to be encoded with a relative position, and the position encoding function is as follows:
Figure BDA0003444481840000071
n is the characteristic vector dimension of the track point, and pos is the serial number of the track point in the sliding window;
the relative position code value is activated by utilizing the trigonometric function, so that the relative position code value can be limited to a certain small range, and the model training is easier to converge. And when pos is an even number, activating the relative position code value by adopting a sine function, and when pos is an odd number, activating the relative position code value by adopting a cosine function.
And S3, taking the feature vector and the relative position code of each track point as input, and training the transform network model to obtain the abnormal track point real-time detection model.
Since the positive and negative samples in the sample data are unevenly distributed, the loss function adopts weighted loss function Cross loss Entropy (BCE), and the loss function is shown as the following formula:
Figure BDA0003444481840000081
in the formula piRepresenting the probability that the ith track point in the abnormal track is predicted as an abnormal point, yiLabel the label value for the human, the outlier is 1 and the outlier is 0.
And then performing online reasoning test on the detection model. Only the feature vectors and the relative position codes of the track sequence data are input into a pre-trained network model, and the network model can output classification information and probability values corresponding to classification to each track point. The probability value may be used as a reference for the confidence level.
Preferably, in some embodiments, after step S1.2, the method further comprises resampling the trajectory segment, wherein the resampling can reduce the amount of computation, and the sampling interval is the same as the road vector interval. Specifically, resampling is carried out on track points in the track segment according to shape point intervals of roads or lanes in a high-precision map. For example, resampling is performed on a track segment corresponding to the roadVector, and the track segment is separated by about 5 meters. Resampling is carried out at intervals of 2 meters on track segments corresponding to group LaneVectors.
Compared with the conventional method based on statistical analysis and artificial logic quality inspection, the method provided by the invention has the advantages that the detection rate and accuracy of abnormal points in the track sequence are remarkably improved. And the acquisition of the track abnormal point truth value is extracted in a semi-automatic manual interaction mode through an algorithm, so that the efficiency is high.
Compared with the traditional statistical algorithm, the iterative optimization process is simple and fast. The phenomenon that the gourd is pressed and lifted in the traditional algorithm can not occur. The method is mainly benefited from the modeling capability of the neural network on the complex system, and has good generalization performance. In the initial stage with less data volume, when the model has errors in some scenes, only the trace marking data of the failure scenes are added, and the model is trained and updated.
And the network model can realize real-time reasoning effect on a terminal computing platform, and the speed is high.
Fig. 3 is a schematic structural diagram of a device for constructing a real-time detection model of an abnormal track point in a vehicle track sequence according to an embodiment of the present invention. As shown in fig. 3, the apparatus includes:
the preprocessing module is used for acquiring historical track data, preprocessing the historical track data, extracting abnormal tracks in the historical track data and constructing model training data;
the characteristic extraction module is used for extracting a characteristic vector of each track point in the abnormal track and carrying out relative position coding on each track point;
the training module is used for taking the feature vector and the relative position code of each track point as input, training the transformer network model and obtaining an abnormal track point real-time detection model; the loss function during training is shown as follows:
Figure BDA0003444481840000091
in the formula piRepresenting the probability that the ith track point in the abnormal track is predicted as an abnormal point, yiLabel the label value for the human, the outlier is 1 and the outlier is 0.
Fig. 4 is a schematic flow chart of a method for detecting an abnormal track point in a vehicle track sequence in real time according to an embodiment of the present invention. As shown in fig. 4, the real-time detection method includes:
acquiring track data to be predicted, extracting a feature vector of each track point in the track data, and performing relative position coding on each track point;
and inputting the characteristic vector and the relative position code into a trained detection model, and performing forward reasoning to obtain the binary attribute of whether the track point is an abnormal track point, wherein the detection model is constructed by the construction method of the real-time detection model of the abnormal track point in the vehicle track sequence.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.

Claims (9)

1. A method for constructing a real-time detection model of abnormal track points in a vehicle track sequence is characterized by comprising the following steps:
acquiring historical track data, preprocessing the historical track data, extracting abnormal tracks in the historical track data, and constructing model training data;
extracting a characteristic vector of each track point in the abnormal track, and carrying out relative position coding on each track point;
taking the feature vector and the relative position code of each track point as input, and training a transform network model to obtain an abnormal track point real-time detection model; the loss function during training is shown as follows:
Figure FDA0003444481830000011
in the formula piRepresenting the probability that the ith track point in the abnormal track is predicted as an abnormal point, yiLabel the label value for the human, the outlier is 1 and the outlier is 0.
2. The method of claim 1, wherein pre-processing the historical trajectory data comprises:
according to the road network grids, dividing historical track data into a plurality of track segment groups, wherein one grid corresponds to one track segment group, and one track segment group comprises a plurality of track segments;
matching the track segment with a road vector or a lane group vector in a high-precision map according to the geographic coordinates of the road network grids;
measuring the similarity between the track segment and a road vector or a lane group vector matched with the track segment by using a Frechet distance;
dividing the track segments into normal track segments and abnormal track segments according to a similarity threshold; and carrying out manual quality inspection on the abnormal track segment, and generating model training data after marking track points in the abnormal track segment.
3. The method of claim 2, wherein said segmenting historical track data into a plurality of track segments according to a road network grid comprises:
projecting longitude and latitude coordinates of the historical track data into plane coordinates according to a 3-degree band UTM;
and performing grid division on the road network, and segmenting the historical track data into a plurality of track segment groups by using the road network grid according to the plane coordinates.
4. The method of claim 2, wherein matching the track segment with a road vector or a lane group vector in a high precision map based on geographic coordinates of a road network grid comprises:
acquiring road attributes in a high-precision map according to the geographic coordinates of the grid, extracting road vectors on road sections without lane increase and decrease, and extracting lane vector groups on intersections or road sections with lane increase and decrease;
if the road vector is extracted according to the grid geographic coordinates, directly matching the track segment group with the road vector;
if the lane vector group is extracted according to the grid geographic coordinates, clustering is respectively carried out on the lane vector group and the track segments in the track segment group, and the lane vectors in the lane vector group are matched with the corresponding sub-track groups in the track segment group according to the direction consistency.
5. The method of claim 2, wherein after matching the track segment with the road vector or lane group vector in the high-precision map according to the geographic coordinates of the road network grid, the method further comprises resampling track points in the track segment according to the shape point interval of the road or lane in the high-precision map.
6. The method according to claim 1, wherein the dimension of the feature vector of each track point in the abnormal track is N, and the method at least comprises: whether the distance is a standing point, an adjacent track point time interval, an adjacent track point course difference value, an adjacent track point distance and an adjacent track point distance difference value.
7. The method of claim 1, wherein said encoding the relative position of each trace point comprises:
acquiring track points in the track segment in a segmented manner by adopting a sliding window with a fixed step length and the size of m;
and (3) carrying out relative position coding on each track point, wherein the position coding function is as follows:
Figure FDA0003444481830000021
n is the characteristic vector dimension of the track point, and pos is the serial number of the track point in the sliding window;
and activating the relative position code value by utilizing a trigonometric function, wherein when pos is an even number, the relative position code value is activated by adopting a sine function, and when pos is an odd number, the relative position code value is activated by adopting a cosine function.
8. The utility model provides a construction equipment of unusual track point real-time detection model in vehicle orbit sequence which characterized in that includes:
the preprocessing module is used for acquiring historical track data, preprocessing the historical track data, extracting abnormal tracks in the historical track data and constructing model training data;
the characteristic extraction module is used for extracting a characteristic vector of each track point in the abnormal track and carrying out relative position coding on each track point;
the training module is used for taking the feature vector and the relative position code of each track point as input, training the transformer network model and obtaining an abnormal track point real-time detection model; the loss function during training is shown as follows:
Figure FDA0003444481830000031
in the formula piRepresenting the probability that the ith track point in the abnormal track is predicted as an abnormal point, yiLabel the label value for the human, the outlier is 1 and the outlier is 0.
9. A real-time detection method for abnormal track points in a vehicle track sequence is characterized by comprising the following steps:
acquiring track data to be predicted, extracting a feature vector of each track point in the track data, and performing relative position coding on each track point;
inputting the feature vectors and the relative position codes into a trained detection model, and performing forward reasoning to obtain the binary attribute of whether the track points are abnormal track points, wherein the detection model is constructed by the construction method of the real-time detection model of the abnormal track points in the vehicle track sequence according to any one of claims 1 to 7.
CN202111643789.9A 2021-12-29 2021-12-29 Model construction method and device and abnormal track point real-time detection method Pending CN114511080A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111643789.9A CN114511080A (en) 2021-12-29 2021-12-29 Model construction method and device and abnormal track point real-time detection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111643789.9A CN114511080A (en) 2021-12-29 2021-12-29 Model construction method and device and abnormal track point real-time detection method

Publications (1)

Publication Number Publication Date
CN114511080A true CN114511080A (en) 2022-05-17

Family

ID=81548235

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111643789.9A Pending CN114511080A (en) 2021-12-29 2021-12-29 Model construction method and device and abnormal track point real-time detection method

Country Status (1)

Country Link
CN (1) CN114511080A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114898342A (en) * 2022-07-15 2022-08-12 深圳市城市交通规划设计研究中心股份有限公司 Method for detecting call receiving and making of non-motor vehicle driver in driving
CN117523382A (en) * 2023-07-19 2024-02-06 石河子大学 Abnormal track detection method based on improved GRU neural network

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114898342A (en) * 2022-07-15 2022-08-12 深圳市城市交通规划设计研究中心股份有限公司 Method for detecting call receiving and making of non-motor vehicle driver in driving
CN114898342B (en) * 2022-07-15 2022-11-25 深圳市城市交通规划设计研究中心股份有限公司 Method for detecting call receiving and making of non-motor vehicle driver in driving
CN117523382A (en) * 2023-07-19 2024-02-06 石河子大学 Abnormal track detection method based on improved GRU neural network
CN117523382B (en) * 2023-07-19 2024-06-04 石河子大学 Abnormal track detection method based on improved GRU neural network

Similar Documents

Publication Publication Date Title
CN110095127B (en) Hidden Markov model map matching method based on segmentation
US9384394B2 (en) Method for generating accurate lane level maps
Davidson et al. Application of particle filters to a map-matching algorithm
CN108961758B (en) Road junction widening lane detection method based on gradient lifting decision tree
CN114511080A (en) Model construction method and device and abnormal track point real-time detection method
JPWO2018008082A1 (en) Driving lane estimation system
CN112013862B (en) Pedestrian network extraction and updating method based on crowdsourcing trajectory
CN111949896B (en) Track abnormality detection method based on track big data
JP2007333385A (en) Immobile object position recording device
CN110427993B (en) High-speed train navigation blind area positioning method based on meteorological parameters
US20170227365A1 (en) Information processing apparatus and trajectory information correcting method
CN114440900A (en) Improved hidden Markov model map matching method and device
CN111275969A (en) Vehicle track filling method based on intelligent identification of road environment
Kumar et al. Citywide reconstruction of cross-sectional traffic flow from moving camera videos
CN109979194B (en) Heavy vehicle ramp crawling behavior identification method based on Markov chain
CN115311617A (en) Method and system for acquiring passenger flow information of urban rail station area
Boucher et al. Automatic detection of topological changes for digital road map updating
CN110647877B (en) Three-dimensional traffic facility positioning and deviation rectifying method and device based on neural network
Bonnifait et al. Multi-hypothesis map-matching using particle filtering
CN115148031B (en) Multi-sensor high-precision positioning method for parking lot inspection vehicle
CN110927765A (en) Laser radar and satellite navigation fused target online positioning method
Wang et al. Intersection and stop bar position extraction from crowdsourced gps trajectories
CN111832365A (en) Lane mark determination method and device
Qing et al. Using feature interaction among GPS Data for road intersection detection
Chen et al. An integrated map-match algorithm with position feedback and shape-based mismatch detection and correction

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