CN114091581A - Vehicle operation behavior type identification method based on sparse track - Google Patents

Vehicle operation behavior type identification method based on sparse track Download PDF

Info

Publication number
CN114091581A
CN114091581A CN202111293639.XA CN202111293639A CN114091581A CN 114091581 A CN114091581 A CN 114091581A CN 202111293639 A CN202111293639 A CN 202111293639A CN 114091581 A CN114091581 A CN 114091581A
Authority
CN
China
Prior art keywords
track
data
vehicle
sparse
point
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
CN202111293639.XA
Other languages
Chinese (zh)
Inventor
巴继东
涂来
胡志华
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.)
WUHAN YANGTZE COMMUNICATIONS INDUSTRY GROUP CO LTD
Wuhan Yangtze Communications Zhilian Technology Co ltd
Original Assignee
WUHAN YANGTZE COMMUNICATIONS INDUSTRY GROUP CO LTD
Wuhan Yangtze Communications Zhilian Technology 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 WUHAN YANGTZE COMMUNICATIONS INDUSTRY GROUP CO LTD, Wuhan Yangtze Communications Zhilian Technology Co ltd filed Critical WUHAN YANGTZE COMMUNICATIONS INDUSTRY GROUP CO LTD
Priority to CN202111293639.XA priority Critical patent/CN114091581A/en
Publication of CN114091581A publication Critical patent/CN114091581A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G06F18/24137—Distances to cluster centroïds
    • G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/21—Design, administration or maintenance of databases
    • G06F16/215—Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/29—Geographical information databases
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/251—Fusion techniques of input or preprocessed data
    • 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
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Health & Medical Sciences (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Biology (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Quality & Reliability (AREA)
  • Remote Sensing (AREA)
  • Traffic Control Systems (AREA)

Abstract

The invention discloses a vehicle operation behavior type identification method based on a sparse track, which comprises the following steps: cleaning the track data of the vehicles of multiple types through data to remove invalid data, and obtaining the cleaned track data of the vehicles; performing data fusion on the cleaned vehicle track data and the semantic segmentation data to obtain fused data; sparse sampling is carried out on the track segment data based on monitoring points; training a travel type classification depth network; and inputting a sparse track sequence to be judged, and outputting the running category distribution of the track section. The invention has the beneficial effects that: the invention can integrate the floating car data with the private car and other multi-source data for analysis, provides more insight for the moving behavior mode of the car, and can better master the urban traffic state; the possibility of finding a vehicle movement pattern is provided, providing a new perspective for distinguishing vehicles for illegal operations.

Description

Vehicle operation behavior type identification method based on sparse track
Technical Field
The invention relates to the technical field of vehicle track operation, in particular to a vehicle operation behavior type identification method based on a sparse track.
Background
The development of internet big data and the popularization of intelligent acquisition equipment greatly simplify the collection of moving object trajectory data, and the data resources are increasingly abundant. One of the most common types of trajectories is generated by a vehicle equipped with an automobile data recorder such as a Beidou/GPS and the like, and the automobile data recorder records the position characteristics of the vehicle at certain time intervals to form a section of trajectory. The vehicle trajectory data contains abundant space-time dynamic information, not only depicts the travel mode and the activity rule of urban residents, but also helps people to know the movement behavior characteristics of different types of vehicles in different stages.
However, since vehicle trajectory data is derived from real life, the trajectory data that can be collected is generally limited due to technical limitations or privacy concerns; in general, only the running track of a commercial vehicle or a few private vehicles can be collected and recorded; for most vehicles, the traffic management department can only obtain the vehicle identification and the time when the vehicle passes through the position of the camera through the camera arranged on the road network, and the traffic management department can obtain the sporadic sparse tracks of the most vehicles.
1. For example, the chinese patent discloses a method and apparatus for determining a movement track based on sparse track point data (application number: CN201310108188.7), by using the density of track point data distribution on at least two historical tracks of an object, clustering the track point data to generate at least two regions, and further determining the region passed by each historical track according to the at least two historical tracks and the at least two regions, and determining at least one area track according to the area passed by each historical track, so that an area track can be determined as the moving track of the object according to the at least one area track and the high-frequency track point data, the problem that the moving track of the object cannot be accurately determined due to the fact that enough track point data cannot be acquired in the prior art can be solved, and the reliability of determining the moving track is improved.
2. A method (application number: CN202110842404.5) for finely dividing and identifying urban road traffic states aiming at sparse track data comprises the steps of 1, acquiring speed values, and calculating the distance between a taxi track point and the end point of the driving direction of each road section as a space relative position value; expanding a [ speed-space ] domain according to the spatial relative position value and the speed value of the track point, calculating the intersection area of the [ speed-space ] domains of the front vehicle and the rear vehicle, constructing a vehicle queue for the track point on the road section on the basis of the intersection area, and selecting an optimal queue according to the Thevenin bauxid index; 3: performing secondary processing on the track queue to obtain finely divided dividing points of the traffic states of all road sections; 4, setting the number of traffic state categories, and obtaining a division threshold value of each category of traffic state by combining a road traffic congestion degree evaluation method; the speed values of the vehicle queues in the finely divided local road sections are compared with the division threshold values of the traffic states of all classes to obtain the traffic states of all the road sections.
In the prior art, although vehicle track data can be analyzed and identified, the type of vehicle operation behavior cannot be identified.
Therefore, it is necessary to provide a sparse trajectory-based vehicle operation behavior type identification method for the above problems.
Disclosure of Invention
In view of the above-mentioned shortcomings in the prior art, the present invention provides a vehicle operation behavior type identification method based on sparse tracks, so as to solve the above-mentioned problems.
A vehicle operation behavior type identification method based on sparse tracks comprises the following steps:
step 1: the method comprises the steps that invalid data of the multi-type vehicle track data are removed through data cleaning, the cleaned vehicle track data are obtained, and the operation state data and the cleaned track data are aligned to form track segment data marked with the operation state;
step 2: performing data fusion on the cleaned vehicle track data and the semantic segmentation data to obtain fused data;
and step 3: sparse sampling is carried out on the track segment data based on monitoring points;
and 4, step 4: training a travel type classification depth network;
and 5: and (4) inputting the sparse track sequence to be judged by using the deep neural network obtained by training in the step (4), and outputting the running category distribution of the track section.
Wherein the further process of the step 1 is as follows:
(1) track data cleaning is carried out, and invalid data, redundant data and drift data are removed;
(2) map matching, namely matching the vehicle position to a corresponding position on a map road to eliminate a dynamic drift point;
(3) and calculating the basic characteristics of the Beidou/GPS track points.
The vehicle track data method in the step 2 comprises a commercial vehicle track data segment and a private vehicle track data travel segment.
The method for segmenting the commercial vehicle track data comprises the following steps:
(1) grouping all track data according to the vehicle ID, sorting the track data from small to large according to sampling time, and taking a track point set P of a single vehicle and order data D of the vehicle as input;
(2) taking out the starting time StartTime and the ending time EndTime in the order, and judging the sampling time of each track point in the set P as follows, wherein if the time point Is between the StartTime and the EndTime, the current track point Is a passenger-carrying travel segment, and Is _ occ Is set to be 1, otherwise, the current track point Is a no-load travel segment, and Is _ occ Is set to be 0;
(3) and (5) circulating the steps 1-2 until all the track points of the vehicles are processed.
The process of the private car schedule segmentation comprises the following steps:
(1) and similarly, grouping all the positioning track data according to the vehicle ID, and sorting the positioning track data from small to large according to the sampling time. Collecting the track points P of a single vehicle;
(2) starting from the first track point of the set P, recording as the starting point of the first journey on the day;
(3) from the starting point, calculating the sampling time interval and the speed direction change between two adjacent track data points in a rolling manner;
(4) and setting the track point with the sampling time interval larger than 15 minutes and the large speed direction change as the starting point of the next section of stroke, wherein the last track point is the end point of the previous section of stroke.
Constructing a depth network model facing sparse track input in step 4:
(1) carrying out track coding on the multi-attribute embedded layer;
(2) a recursion module for modeling a trajectory order factor;
(3) and (4) carrying out track classification by using the information extracted by the previous module, and distributing labels for the tracks.
Compared with the prior art, the invention has the beneficial effects that: the vehicle operation behavior type identification method can integrate floating vehicle data and private vehicle and other multi-source data for analysis, provides more insights for vehicle movement behavior modes, can better master urban traffic states, has certain reference value for further research and policy making under urban traffic management, and can monitor illegal operation by people by using cameras and passing vehicle data at checkpoints; the recognition method provides the possibility of finding the moving mode of the vehicle, and provides a new view for distinguishing the vehicles for illegal operation; the sparse track is adopted to analyze and identify the vehicle behaviors, explore and mine the space-time characteristics of sparse data, and have important practical significance for traffic operation management work such as abnormal driving behavior early warning, illegal operation behavior discovery and the like.
Drawings
FIG. 1 is a flow chart of a sparse trajectory-based vehicle operation behavior classification model training process of the present invention;
FIG. 2 is a schematic diagram of a monitoring point-based sparse sampling method for track segment data according to the present invention;
FIG. 3 is a schematic diagram of a network structure of a LSTM-based sparse track fusion feature classification model according to the present invention;
FIG. 4 is a flowchart of a sparse trajectory-based vehicle operation behavior classification model decision process.
Detailed Description
It should be noted that the embodiments and features of the embodiments may be combined with each other without conflict.
The embodiments of the invention will be described in detail below with reference to the drawings, but the invention can be implemented in many different ways as defined and covered by the claims.
As shown in fig. 1 to 4, a sparse trajectory-based vehicle operation behavior type identification method includes the following steps:
step 1: the method comprises the steps that invalid data of the multi-type vehicle track data are removed through data cleaning, the cleaned vehicle track data are obtained, and the operation state data and the cleaned track data are aligned to form track segment data marked with the operation state; the vehicle track is a time sequence of the positions visited by the vehicle on the road, and Traj is set<p1p2p3…pi...pn>Wherein p isi(1. ltoreq. i. ltoreq. n) is a point on the trajectory, determined by the geographical position and the timestamp, and can be defined as pi＝(loni，lati，ti) Wherein t isiIs a time stamp, (lon)i，lati) Two-dimensional coordinates consisting of longitude and latitude.
Step 1.1 track data cleaning
The track data acquired by the Beidou/GPS equipment comprises a plurality of fields, and the fields which are meaningful for track characteristic analysis, such as the instant time, the longitude and latitude, the vehicle direction, the instant speed and the like, are usually reserved; meanwhile, vehicle data are observed in combination with the requirements of vehicle movement behavior analysis, the range of each field of the collected data is limited according to corresponding standards, and data filtering is carried out and mainly comprises two parts: determining the range of a recording space position field (longitude and latitude) according to the specific geographic position of the target area; for example, setting the range of the field Lat to [22.44,22.79], the range of the field Lon to [113.76,114.64 ]; the field Speed records the instantaneous Speed of the vehicle at a certain moment of time, and the Speed limit standard of the urban road needs to be complied with, so the range of the field Speed is set to [0,120 ].
In addition, when the vehicle is parked for a long time and the Beidou/GPS equipment still collects the position points with no obvious change of the longitude and latitude information, the records have no any significance for analyzing the moving behavior of the vehicle, can be regarded as redundant data and need to be removed; however, in order to avoid misidentification as a parking spot caused by traffic conditions (such as traffic congestion and traffic lights), the standard for identifying the sample point as a long-time parking spot is a plurality of points that are adjacent in time, have no change in longitude and latitude, and have a speed of zero, and these redundant points are deleted; therefore, meaningful vehicle track points can be obtained, and the storage space of data is reduced.
Due to the influences of hardware equipment, environmental factors and the like, the track data acquired by the Beidou/GPS cannot be completely correct; if the Beidou/GPS equipment is abnormal, an error signal is generated, and a track drifting phenomenon is formed; the drift phenomenon causes many problems, and the generated noise point error is too large to obtain useful information; Beidou/GPS drift mainly comprises static drift and dynamic drift, the vehicle stops running but the positioning coordinate still changes and is called as static drift, and before track data mining is carried out, the illogical static drift points need to be filtered. The Beidou/GPS drift is also generated when the vehicle moves, and is called as dynamic drift; for example, a vehicle running on a road has its next track point located off the road, and the track point deviates significantly from the entire track sequence; and matching the dynamic drifting points into the road network by a map matching method according to the characteristics of the dynamic drifting.
Step 1.2 map matching
Map matching is a key step for researching vehicle driving behaviors and identifying network traffic conditions from data, Beidou/GPS track data and road network information are fused by technologies such as a probability statistical method and a geometric method, and track point positions are matched to corresponding positions on an actual road, namely, a longitude and latitude sequence of a vehicle track is matched with a digital map network; map matching is essentially a pattern matching problem for sequences of planar line segments. According to the analysis of the current research situation, a leuven map matching module proposed by Meert and Verbeke is selected to perform map matching preprocessing operation on the sparse track.
Step 1.3 basic trajectory feature extraction
The vehicle-mounted Beidou/GPS equipment not only collects position information such as longitude, latitude and altitude, but also can collect instantaneous speed and direction of vehicle running, and the speed of a time segment can be distinguished through the longitude and latitude and timestamp information of a Beidou/GPS track point, so that the basic characteristics of vehicle motion are described;
step 2: carrying out data fusion on the cleaned vehicle track data and the semantic segmentation data to obtain fused data, wherein the stroke sections of different vehicles are divided differently; for subsequent analytical studies, the vehicle travel is segmented according to different methods:
step 2.1, segmenting the operating vehicle:
for a commercial vehicle, the order data fields shown in table 1 record information such as time and place for passengers to get on or off the vehicle, and passenger carrying information of the vehicle can be judged, so that travel sections are divided; firstly, a field Is _ occ Is added to the track data of the commercial vehicle, and Is a passenger-carrying travel section when the value Is IS _ occ Is 1, and Is an idle travel section when the value Is 0.
Table 1 order data section field description
Figure BDA0003335841620000071
Figure BDA0003335841620000081
The method for dividing the travel sections based on the order information comprises the following steps:
(1) and grouping all the track data according to the vehicle ID, and sorting the track data from small to large according to the sampling time. Taking a track point set P of a single vehicle and order data D of the vehicle as input;
(2) taking out the starting time StartTime and the ending time EndTime in the order, and judging the sampling time of each track point in the set P as follows, wherein if the time point Is between the StartTime and the EndTime, the current track point Is a passenger-carrying travel segment, and Is _ occ Is set to be 1, otherwise, the current track point Is a no-load travel segment, and Is _ occ Is set to be 0;
(3) and (5) circulating the steps 1-2 until all the track points of the vehicles are processed.
Step 2.2 segmentation of track data travel of private car
The data set of the private car has no order data, and only each track point of the track data records the acquisition time, the position information and the speed information of the car; selecting part of running characteristics in the track as reference conditions for searching for the stroke segmentation points; when the vehicle runs in the road, except at the crossroad or under the emergency, the vehicle can not change the running direction suddenly generally; the sectional point of the change of the moving mode is when the moving object rapidly changes in direction and speed and tends to be stable; the segmentation algorithm firstly calculates the change of the track midpoint and the previous point in the speed and direction, and if the change is larger than a set threshold value, the change is set as a segmentation point. However, this method produces more segmentation points at the red road lights or in the case of traffic congestion; thus, the trajectory segmentation is based on time intervals and speed and direction changes; considering the actual situation, setting the time interval to be 15 minutes, setting the difference between the sampling moments of two points on the track to be more than 15 minutes and setting the sampling moments to be the end point of the previous section of the stroke and the starting point of the next section of the stroke respectively when the speed and the direction change too much; the flow of the private car journey segmentation is as follows:
(1) similarly, grouping all the positioning track data according to the vehicle ID, sorting the positioning track data from small to large according to the sampling time, and collecting a track point set P of a single vehicle;
(2) starting from the first track point of the set P, recording as the starting point of the first journey on the day;
(3) from the starting point, calculating the sampling time interval and the speed direction change between two adjacent track data points in a rolling manner;
(4) and setting the track point with the sampling time interval larger than 15 minutes and the large speed direction change as the starting point of the next section of stroke, wherein the last track point is the end point of the previous section of stroke.
Step 2.3 track segment State labeling
After the vehicle travel sections are divided, attaching two state labels of no-load travel and trip travel to each track; as for the types of vehicles, the types of the vehicles comprise a taxi, a net appointment passenger vehicle, a net appointment freight vehicle and a private car, the taxi and the net appointment vehicle have two travel sections, and the private car has only one travel section; therefore, the method can be finally divided into 5 vehicle behaviors, namely the no-load behavior and the travel behavior of the taxi, the no-load behavior and the travel behavior of the net appointment car and the travel behavior of the private car.
And step 3: sparse sampling of track segment data based on monitoring points
After completing the track sequence segmentation and labeling of step 2, the data of each track segment has the following form: t (traj) ═ c wherein traj ═ c<p1，p2，p3，...，pi，...pN>，pi(1. ltoreq. i. ltoreq. n) is a point on the trajectory, determined by the geographical position and the timestamp, and can be defined as pi＝(loni，lati，ti) Wherein t isiFor time stamping, loni，latiTwo-dimensional coordinates consisting of longitude and latitude, and C is a travel behavior category;
c belongs to { taxi is unloaded, taxi trip, net car appointment is unloaded, net car appointment trip, private car trip }
In the step, sparse sampling is carried out on the track segment data based on the road network monitoring points. Set effective monitoring points of road network as li＝(loni，lati，diri) E.g. L, where (lon)i，lati) Is the latitude and longitude coordinate, dir, of the i-th monitoring point cameraiThe shooting direction of the monitoring camera is taken; based on the traj track and the monitoring point set L, the sparse track traj is generated according to the following process stepss: from the first point p of the track segment1At the beginning, the monitor points traversed by the successive strokes, i.e. pi←p1。
Looking for distance track segments in L, as in FIG. 2
Figure BDA0003335841620000101
Proximal (Dis)k≤LWkWherein LWkThe distance is expressed as a similar threshold value, and a monitoring point l can be selected by taking the valuekWidth of the road) and the image pickup direction faces the monitoring point (| Dir) of the track proceeding directionk-Dir(pi+1，pi)|≤θThWherein thetaThThe value can be selected from 15 degrees for the direction relative threshold value, if a monitoring point l meeting the condition is foundkThen will (l)k，ti) Storing in sparse track sequences trajs；pi←pi+1After the step (2) is repeated to finish the steps, a sparse track sequence traj is obtaineds＝<(l1，t1)，(l2，t2)，...，(ln，tn)>And a corresponding trip type T (traj)s)＝c。
And 4, step 4: and training a travel type classification deep network.
Constructing a depth network model facing to sparse track input, inputting sparse track trajs and a type c corresponding to the sparse track trajs based on an LSTM network structure, and training network parameters; the LSTM for sparse trajectory classification includes three parts: (1) carrying out track coding on the multi-attribute embedded layer; (2) a recursion module for modeling a trajectory order factor; (3) and (3) carrying out track classification by using the information extracted by the previous module, and distributing labels for the tracks, wherein the track classification mainly comprises three components as shown in figure 3.
The track coding module is used for coding track point characteristics, the longitude and latitude position characteristics of the monitoring points are coded by using a geohash code, other semantic characteristics around the monitoring points, such as surrounding POI, are coded by using a one-hot code, other quantization characteristics, such as vehicle section speed, lane number, driving direction and the like, are directly used as an extended dimension vector and are aggregated with a coding vector, and the time characteristics of the monitoring points are input by using a mode of day date in the week + day time (such as Tuesday, 8 hours, 30 minutes and 00 seconds);
multiplying each attribute by their respective embedding matrix to extract their respective embedded representation, applying an aggregation function input to the LSTM unit; the LSTM unit captures patterns in the sequence over time intervals of variable length, learning features that include the first and last points of the trajectory.
The output features of the LSTM unit are the hidden features of the trajectory classification module, each hidden feature output is input to a fully connected layer with the goal of mapping the learned knowledge to the corresponding label. The goal of training the model is to minimize the cross-entropy loss of the classification module, as shown by the following equation:
Figure BDA0003335841620000111
wherein DtrainThe method comprises the steps of training a trajectory section, wherein L is a classified trajectory label, C is a trip type, T is time, and P is a random probability.
Then, highlighting the difference between the labels by using a softmax function, and further outputting the probability distribution of all possible data labels; in order to avoid model overfitting, dropout and regularization techniques are used; the Dropout layer is used in the whole model, so the units are randomly discarded in the training process; in addition, the weights and biases of the LSTM cells are regularized using the L1 regularization method.
And 5: and (4) inputting the sparse track sequence to be judged by using the deep neural network obtained by training in the step (4), and outputting the running category distribution of the track section.
The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.

Claims (6)

1. A vehicle operation behavior type identification method based on sparse tracks is characterized by comprising the following steps: the method comprises the following steps:
step 1: the method comprises the steps that invalid data of the multi-type vehicle track data are removed through data cleaning, the cleaned vehicle track data are obtained, and the operation state data and the cleaned track data are aligned to form track segment data marked with the operation state;
step 2: performing data fusion on the cleaned vehicle track data and the semantic segmentation data to obtain fused data;
and step 3: sparse sampling is carried out on the track segment data based on monitoring points;
and 4, step 4: training a travel type classification depth network;
and 5: and (4) inputting the sparse track sequence to be judged by using the deep neural network obtained by training in the step (4), and outputting the running category distribution of the track section.
2. The sparse trajectory-based vehicle operation behavior type identification method as claimed in claim 1, wherein:
wherein the further process of the step 1 is as follows:
(1) track data cleaning is carried out, and invalid data, redundant data and drift data are removed;
(2) map matching, namely matching the vehicle position to a corresponding position on a map road to eliminate a dynamic drift point;
(3) and calculating the basic characteristics of the Beidou/GPS track points.
3. The sparse trajectory-based vehicle operation behavior type identification method as claimed in claim 1, wherein: the vehicle track data method in the step 2 comprises a commercial vehicle track data segment and a private vehicle track data travel segment.
4. The sparse trajectory-based vehicle operation behavior type identification method as claimed in claim 3, wherein: the method for segmenting the commercial vehicle track data comprises the following steps:
(1) grouping all the track data according to the vehicle ID, and sequencing the track data from small to large according to the sampling time; taking a track point set P of a single vehicle and order data D of the vehicle as input;
(2) taking out the starting time StartTime and the ending time EndTime in the order, and judging the sampling time of each track point in the set P as follows, wherein if the time point Is between the StartTime and the EndTime, the current track point Is a passenger-carrying travel segment, and Is _ occ Is set to be 1, otherwise, the current track point Is a no-load travel segment, and Is _ occ Is set to be 0;
(3) and (5) circulating the steps 1-2 until all the track points of the vehicles are processed.
5. The sparse trajectory-based vehicle operation behavior type identification method as claimed in claim 3, wherein:
the process of the private car schedule segmentation comprises the following steps:
(1) all the positioning track data are grouped according to the vehicle ID, and are sorted from small to large according to the sampling time; collecting the track points P of a single vehicle;
(2) starting from the first track point of the set P, recording as the starting point of the first journey on the day;
(3) from the starting point, calculating the sampling time interval and the speed direction change between two adjacent track data points in a rolling manner;
(4) and setting the track point with the sampling time interval larger than 15 minutes and the large speed direction change as the starting point of the next section of stroke, wherein the last track point is the end point of the previous section of stroke.
6. The sparse trajectory-based vehicle operation behavior type identification method as claimed in claim 1, wherein:
constructing a depth network model facing sparse track input in step 4:
(1) carrying out track coding on the multi-attribute embedded layer;
(2) a recursion module for modeling a trajectory order factor;
(3) and (4) carrying out track classification by using the information extracted by the previous module, and distributing labels for the tracks.
CN202111293639.XA 2021-11-03 2021-11-03 Vehicle operation behavior type identification method based on sparse track Pending CN114091581A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111293639.XA CN114091581A (en) 2021-11-03 2021-11-03 Vehicle operation behavior type identification method based on sparse track

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111293639.XA CN114091581A (en) 2021-11-03 2021-11-03 Vehicle operation behavior type identification method based on sparse track

Publications (1)

Publication Number Publication Date
CN114091581A true CN114091581A (en) 2022-02-25

Family

ID=80298708

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111293639.XA Pending CN114091581A (en) 2021-11-03 2021-11-03 Vehicle operation behavior type identification method based on sparse track

Country Status (1)

Country Link
CN (1) CN114091581A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117077042A (en) * 2023-10-17 2023-11-17 北京鑫贝诚科技有限公司 Rural level crossing safety early warning method and system
CN117392850A (en) * 2023-11-29 2024-01-12 哈尔滨航天恒星数据系统科技有限公司 SMO-based traffic congestion real-time prediction and release method, electronic equipment and storage medium

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117077042A (en) * 2023-10-17 2023-11-17 北京鑫贝诚科技有限公司 Rural level crossing safety early warning method and system
CN117077042B (en) * 2023-10-17 2024-01-09 北京鑫贝诚科技有限公司 Rural level crossing safety early warning method and system
CN117392850A (en) * 2023-11-29 2024-01-12 哈尔滨航天恒星数据系统科技有限公司 SMO-based traffic congestion real-time prediction and release method, electronic equipment and storage medium
CN117392850B (en) * 2023-11-29 2024-05-28 哈尔滨航天恒星数据系统科技有限公司 SMO-based traffic congestion real-time prediction and release method, electronic equipment and storage medium

Similar Documents

Publication Publication Date Title
CN111243277B (en) Commuting vehicle space-time trajectory reconstruction method and system based on license plate recognition data
CN108629978B (en) Traffic track prediction method based on high-dimensional road network and recurrent neural network
CN109410586B (en) Traffic state detection method based on multi-metadata fusion
CN110176139B (en) DBSCAN + -based road congestion identification visualization method
CN109871876B (en) Expressway road condition identification and prediction method based on floating car data
CN110717433A (en) Deep learning-based traffic violation analysis method and device
CN108848460B (en) Man-vehicle association method based on RFID and GPS data
CN109686090B (en) Virtual traffic flow calculation method based on multi-source data fusion
WO2021082464A1 (en) Method and device for predicting destination of vehicle
CN108062857B (en) Prediction technique for cab-getter&#39;s trip purpose
CN107590999B (en) Traffic state discrimination method based on checkpoint data
CN114091581A (en) Vehicle operation behavior type identification method based on sparse track
CN111768619A (en) Express way vehicle OD point determining method based on checkpoint data
CN110836675A (en) Decision tree-based automatic driving search decision method
CN114170580A (en) Highway-oriented abnormal event detection method
CN115995149A (en) Multi-source data-based parking supply and demand characteristic dynamic evaluation method and system
CN114882069A (en) Taxi track abnormity detection method based on LSTM network and attention mechanism
CN114973659A (en) Method, device and system for detecting indirect event of expressway
CN115761920A (en) ETC portal data-based vehicle track credible extraction method and system
CN109979194B (en) Heavy vehicle ramp crawling behavior identification method based on Markov chain
CN116092037B (en) Vehicle type identification method integrating track space-semantic features
Chen et al. Detecting and analyzing unlicensed taxis: A case study of Chongqing City
CN116109986A (en) Vehicle track extraction method based on laser radar and video technology complementation
Golze et al. Impact analysis of accidents on the traffic flow based on massive floating car data
Hou Applications of big data technology in intelligent transportation system

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