CN115731261B - Vehicle lane change behavior recognition method and system based on expressway radar data - Google Patents
Vehicle lane change behavior recognition method and system based on expressway radar data Download PDFInfo
- Publication number
- CN115731261B CN115731261B CN202110997952.5A CN202110997952A CN115731261B CN 115731261 B CN115731261 B CN 115731261B CN 202110997952 A CN202110997952 A CN 202110997952A CN 115731261 B CN115731261 B CN 115731261B
- Authority
- CN
- China
- Prior art keywords
- track
- vehicle
- sequence
- data
- initial
- 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
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The invention relates to a vehicle lane change behavior recognition method and system based on expressway radar data, belongs to the technical field of vehicle lane change recognition, and solves the problems of low reliability and poor recognition accuracy of a vehicle lane change recognition method based on radar data in the prior art. The method comprises the following steps: acquiring radar data of a highway, and generating an initial track of a vehicle based on the radar data; after cleaning the initial track of the vehicle, matching the initial track of the vehicle based on an improved Frechet distance track matching algorithm to obtain a track of the vehicle; after abnormality removal and smoothing are carried out on the vehicle track, extracting feature data, and adding a category label for the feature data to obtain sample data, wherein the category label comprises lane change or non-lane change; training the support vector machine model based on the sample data to obtain an optimal support vector machine model; and inputting the initial track characteristic data of the vehicle to be identified into an optimal support vector machine model to obtain the identification result of the lane change behavior of the vehicle.
Description
Technical Field
The invention relates to the technical field of vehicle lane change identification, in particular to a vehicle lane change behavior identification method and system based on expressway radar data.
Background
Due to the high running speed of the expressway, once traffic accidents occur, the loss and the mortality rate of the expressway are far higher than those of an ordinary road, and the expressway is a typical behavior mode for changing the road of the vehicle and is closely connected with the driving safety. The lane change behavior of the vehicles in the expressway is identified, so that the technical system of the vehicle early warning system is improved, the traffic informatization and refinement management level of China is improved, and auxiliary support is provided for guaranteeing the traffic safety of the roads and improving the traffic service level.
At present, with the development of intelligent traffic, radar detectors are widely distributed in highways by virtue of the characteristics of accuracy, low cost and the like, and become one of important detection means and methods gradually, and play an important role in speed detection, lane changing assistance, unmanned driving, multi-sensor data fusion and other applications. However, the radar data has a large number of abnormal values and the obtained track is discontinuous, so that the existing vehicle lane change identification method based on the radar data is low in reliability and poor in identification precision because the radar data cannot be fully utilized due to the lack of pre-processing of the radar data.
Disclosure of Invention
In view of the above analysis, the embodiment of the invention aims to provide a vehicle lane change behavior recognition method and system based on expressway radar data, which are used for solving the problems of low reliability and poor recognition accuracy of the existing vehicle lane change recognition method based on radar data.
In one aspect, the embodiment of the invention provides a vehicle lane change behavior recognition method based on expressway radar data, which comprises the following steps:
acquiring radar data of a highway, and generating an initial track of a vehicle based on the radar data;
after the initial track of the vehicle is cleaned, matching is carried out based on an improved Frechet distance track matching algorithm, and a vehicle track is obtained;
after abnormality removal and smoothing are carried out on the vehicle track, extracting feature data, and adding a category label for the feature data to obtain sample data, wherein the category label comprises lane change or non-lane change;
training a support vector machine model based on the sample data to obtain an optimal support vector machine model;
and inputting the initial track characteristic data of the vehicle to be identified into an optimal support vector machine model to obtain the identification result of the lane change behavior of the vehicle.
Further, the radar data includes acquisition time, target ID, vehicle lateral position, vehicle longitudinal position, lateral instantaneous speed, and longitudinal instantaneous speed.
Further, a vehicle initial trajectory is generated based on the radar data by:
traversing radar data in sequence according to acquisition time:
initializing a first piece of data which is not added to the track sequence in the radar data as a first track point of a j=1th track sequence, and sequentially finding a next track point in the track sequence in the radar data: if there is data which is the same as the target ID of the first track point in the track sequence and is not added to the track sequence, judging the piece of data according to the following formula:
wherein x is z And y z The lateral and longitudinal positions of the vehicle, x, respectively, representing the last track point in the track sequence un And y un A vehicle lateral position and a longitudinal position respectively representing data in the radar data which is the same as the target ID of the first track point in the track sequence and which is not added to the track sequence;
if so, the piece of data is added to the track sequence, otherwise the piece of track ends, and j=j+1.
Further, the vehicle initial track is cleaned through power constraint, specifically:
each track sequence in the initial track of the vehicle is subjected to power constraint:
calculating the velocity v of adjacent track points in the track sequence by k ：
Wherein x is k And y k A vehicle transverse position and a vehicle longitudinal position respectively representing a kth track point of the track sequence; x is x k+1 And y k+1 Respectively represent the (k+1) th track of the track sequenceThe vehicle lateral position and the vehicle longitudinal position of the point; t is t k And t k+1 Respectively representing the acquisition time of the kth track point and the kth+1th track point of the track sequence;
calculating the acceleration a of adjacent speed according to the speeds of adjacent track points in the track sequence k ：
If all track points in the track sequence meet v k ＜v 0 And a k ＜a 0 The initial track of the vehicle keeps the track sequence, otherwise, the track sequence is deleted; wherein v is 0 And a 0 A speed threshold and an acceleration threshold, respectively;
further, matching is carried out based on an improved Frechet distance track matching algorithm, and a vehicle track is obtained, specifically:
selecting a track sequence with the number of track points larger than or equal to a length threshold value in the initial track of the vehicle as an initial long track of the vehicle, otherwise, selecting a track sequence with the number of track points larger than or equal to a length threshold value as an initial short track of the vehicle;
calculating the Frechet distance between each track sequence in the initial short track of the vehicle and each track in the initial long track of the vehicle;
and if the Frechet distance is smaller than the set similarity threshold, averaging corresponding data in the two track sequences, and updating the track sequences.
Further, the fraiche distance F (a, B) of the track sequence a in the initial short track of the vehicle and the track sequence B in the initial long track of the vehicle is calculated by:
F(A,B)＝inf{max{|d(α(t),β(t))|}+kσ(d)}；
where d (α (t), β (t))=y At -y Bt +ΔT(v At -v Bt )；
Wherein d (α (t), β (t)) represents the distance between the track points α (t) and β (t) of the track sequence a and the track sequence B at the time t, y At And y Bt Respectively representing the longitudinal position of the vehicle at the time t, v, of the track sequence A and the track sequence B At And v Bt The longitudinal instantaneous speeds of the track sequence A and the track sequence B at the moment T are respectively represented, delta T represents the acquisition time interval of radar data, k represents the discrete weight coefficient, sigma (d) represents the standard deviation of the distances of the track sequence A and the track sequence B at all moments, and inf { } represents the maximum lower bound of the function value.
Further, the vehicle track is subjected to abnormality removal and smoothing, specifically:
if the number of track points of the track sequence in the vehicle track is greater than or equal to the length threshold value, the initial track of the vehicle keeps the track sequence, otherwise, the track sequence is deleted;
and carrying out track smoothing on the vehicle track after the abnormality is removed based on a Savitzky-Golay filter.
Further, the feature data is extracted by:
the method comprises the steps of respectively taking a first track point and a last track point, of which the transverse instantaneous speed is greater than a track changing threshold value, in a track sequence of a vehicle as a start track point and an end track point of a track changing section in the track sequence to obtain the track changing section of the track sequence;
the lane change interval based on the track sequence extracts the following characteristics:
offset angle:
average lane change transverse velocity v xca ：
If t e -t s Is not 0, then
Otherwise v xca ＝0；
Average lane change longitudinal speed v yca ：
If t e -t s Is not 0, then
Otherwise v yca ＝0；
Lateral offset Δx c ：
Track change time delta t c ：
Δt c ＝t e -t s ；
Wherein Δx and Δy represent the vehicle lateral position change amount and the vehicle longitudinal position change amount of the start track point and the end track point of the lane change section, respectively; t is t e And t s Respectively representing acquisition moments of a start track point and an end track point of a track change interval; v xi And v yi Respectively representing the transverse instantaneous speed and the longitudinal instantaneous speed of the track point at the acquisition time i;
and->
Respectively at the acquisition time t e And t s The vehicle lateral position of the locus point at that time;
and carrying out normalization processing on the extracted features to obtain feature data of a track sequence in the vehicle track.
On the other hand, the embodiment of the invention also provides a vehicle channel change behavior recognition system based on the expressway radar data, which comprises the following steps:
the data acquisition module is used for acquiring radar data of the expressway and generating an initial track of the vehicle based on the radar data;
the vehicle track generation module is used for carrying out matching on the basis of an improved French distance track matching algorithm after the initial track of the vehicle is cleaned, so as to obtain a vehicle track;
the sample data generation module is used for extracting characteristic data after abnormality removal and smoothing of the vehicle track, and adding a category label for the characteristic data to obtain sample data, wherein the category label comprises lane change or non-lane change;
the model training module is used for training the support vector machine model based on the sample data to obtain an optimal support vector machine model;
the recognition module is used for inputting the initial track characteristic data of the vehicle to be recognized into the optimal support vector machine model to obtain the recognition result of the lane change behavior of the vehicle.
Further, the radar data includes acquisition time, target ID, vehicle lateral position, vehicle longitudinal position, lateral instantaneous speed, and longitudinal instantaneous speed.
Compared with the prior art, the invention has at least one of the following beneficial effects:
the invention provides a vehicle lane change behavior recognition method and system based on expressway radar data:
1. generating an initial track of a vehicle based on radar data, obtaining the track of the vehicle through cleaning and track matching repair, extracting characteristic data after abnormality removal and smoothing to obtain sample data, training a support vector machine model by using the sample data, and finally inputting the characteristic data to be identified into an optimal support vector machine model to obtain an identification result of a vehicle lane change behavior, thereby solving the problem that the existing method lacks of large error of the identification result caused by early processing of radar data, and improving the reliability and precision of the identification result;
2. the vehicle initial track is generated after the position of the data of the adjacent acquisition time of the same vehicle in the radar data is judged, namely the same track cannot cross the physical interval, so that the space constraint is carried out when the initial track is generated, and the accuracy of the obtained vehicle initial track is improved;
3. the vehicle initial track is cleaned through power constraint, abnormal tracks in the track can be removed, track matching is carried out based on an improved Frechet distance track matching algorithm, data in the track is repaired, and the accuracy of the obtained vehicle track is further improved;
4. after the abnormal removal and smoothing of the vehicle track are carried out, abnormal data in the track are further removed, basic data are provided for subsequent feature extraction, the accuracy of model training in the later period is improved, the model is trained by combining radar data and vehicle lane changing extraction of five related features, and the accuracy and reliability of model identification are improved.
In the invention, the technical schemes can be mutually combined to realize more preferable combination schemes. Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention may be realized and attained by the structure particularly pointed out in the written description and drawings.
Drawings
The drawings are only for purposes of illustrating particular embodiments and are not to be construed as limiting the invention, like reference numerals being used to refer to like parts throughout the several views.
Fig. 1 is a flowchart of a vehicle lane change behavior recognition method based on expressway radar data in embodiment 1 of the invention.
Detailed Description
Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form a part hereof, and together with the description serve to explain the principles of the invention, and are not intended to limit the scope of the invention.
Example 1
An embodiment 1 of the present invention discloses a vehicle lane change behavior recognition method based on highway radar data, as shown in fig. 1, comprising the following steps:
s1, acquiring radar data of a highway, and generating an initial track of a vehicle based on the radar data.
In practice, the radar data includes acquisition time, target ID, vehicle lateral position, vehicle longitudinal position, lateral instantaneous speed, and longitudinal instantaneous speed. Preferably, the radar data further includes vehicle type information.
Specifically, the acquisition time interval of radar data is fixed; the target ID is used to identify different vehicles; the vehicle transverse position being 0 represents the road center line, the positive and negative of the vehicle transverse position being used for representing the direction of travel of the vehicle; the vehicle type 1 represents a trolley, the vehicle type 2 represents a cart, more specifically, the cart with the length of more than 6 meters is classified as a cart, and the rest are classified as carts.
In practice, an initial trajectory of the vehicle is generated based on the radar data by:
traversing radar data in sequence according to acquisition time:
initializing a first piece of data which is not added to the track sequence in the radar data as a first track point of a j=1th track sequence, and sequentially finding a next track point in the track sequence in the radar data: if there is data which is the same as the target ID of the first track point in the track sequence and is not added to the track sequence, judging the piece of data according to the following formula:
wherein x is z And y z The lateral and longitudinal positions of the vehicle, x, respectively, representing the last track point in the track sequence un And y un A vehicle lateral position and a longitudinal position respectively representing data in the radar data which is the same as the target ID of the first track point in the track sequence and which is not added to the track sequence;
if so, the piece of data is added to the track sequence, otherwise the piece of track ends, and j=j+1.
It can be understood that the vehicle initial track is generated by performing position determination on the data of the adjacent acquisition time of the same vehicle in the radar data, so that the space constraint is performed when the initial track is generated, namely, the generated vehicle initial track does not have two running directions in the same track or can cross a physical interval, and the generated track is accurate and is more suitable for vehicle lane change identification.
S2, after the initial track of the vehicle is cleaned, the initial track of the vehicle is matched based on an improved Frechet distance track matching algorithm, and the track of the vehicle is obtained.
During implementation, the initial track of the vehicle is cleaned through power constraint, specifically:
each track sequence in the initial track of the vehicle is subjected to power constraint:
calculating the velocity v of adjacent track points in the track sequence by k ：
Wherein x is k And y k A vehicle transverse position and a vehicle longitudinal position respectively representing a kth track point of the track sequence; x is x k+1 And y k+1 The vehicle lateral position and the vehicle longitudinal position of the (k+1) th track point of the track sequence are respectively represented; t is t k And t k+1 The acquisition times of the kth and the kth+1th track points of the track sequence are respectively represented.
Calculating the acceleration a of adjacent speed according to the speeds of adjacent track points in the track sequence k ：
If all track points in the track sequence meet v k ＜v 0 And a k ＜a 0 The initial track of the vehicle keeps the track sequence, otherwise, the track sequence is deleted; wherein v is 0 And a 0 A speed threshold and an acceleration threshold, respectively. It should be noted that it is the initial trajectory of the vehicle that retains the trajectory sequence when the velocity of each adjacent trajectory point in the trajectory sequence and the acceleration of the adjacent velocity satisfy the conditions, otherwise the trajectory sequence is deleted.
Specifically, the velocity threshold v 0 According to the road segment speed limit regulation, the speed limit is 110m/s, the speed threshold value can be 150m/s, and the acceleration threshold value a 0 According to the vehicle dynamic performance standard, the acceleration threshold value is 0.6m/s 2 。
It can be understood that the power constraint refers to that the matching of the track points in the track sequence should meet the actual conditions of speed and acceleration, the speed is obtained according to the positions of two adjacent track points, the acceleration is obtained according to three track points, and whether the speed and the acceleration are smaller than the maximum power condition of the vehicle in the expressway environment is checked, so that the initial track of the vehicle is cleaned through the power constraint, the abnormal track in the track can be removed, and the accuracy of the track is improved.
When the method is implemented, the vehicle track is obtained by matching based on an improved Frechet distance track matching algorithm, and the method specifically comprises the following steps:
and selecting a track sequence with the track points of which the number is greater than or equal to a length threshold value in the initial track of the vehicle as an initial long track of the vehicle, otherwise, selecting an initial short track of the vehicle.
And calculating the Frechet distance between each track sequence in the initial short track of the vehicle and each track in the initial long track of the vehicle.
Specifically, the friendship distance F (a, B) of the track sequence a in the initial short track of the vehicle and the track sequence B in the initial long track of the vehicle is calculated by:
F(A,B)＝inf{max{|d(α(t),β(t))|}+kσ(d)}；
where d (α (t), β (t))=y At -y Bt +ΔT(v At -v Bt )；
Wherein d (α (t), β (t)) represents the distance between the track points α (t) and β (t) of the track sequence a and the track sequence B at the time t, y At And y Bt Respectively representing the longitudinal position of the vehicle at the time t, v, of the track sequence A and the track sequence B At And v Bt The longitudinal instantaneous speeds of the track sequence A and the track sequence B at the moment T are respectively represented, delta T represents the acquisition time interval of radar data, k represents the discrete weight coefficient, sigma (d) represents the standard deviation of the distances of the track sequence A and the track sequence B at all moments, and inf { } represents the maximum lower bound of the function value. It can be appreciated that the longitudinal instantaneous speed is added in the French distance, so that the matching method is more suitable for matching judgment of radar data tracks, and the matching accuracy is higher.
And if the Frechet distance is smaller than the set similarity threshold, averaging corresponding data in the two track sequences, and updating the track sequences. Specifically, vehicle position information and speed information in corresponding data in the track sequence are averaged for repairing the data in the track sequence. Preferably, the vehicle type information can be updated, and if the proportion of the vehicle type in each track point in the initial long track sequence of the vehicle is high, the vehicle type in the corresponding data is changed into the large vehicle.
Specifically, the set similarity threshold is determined according to the vehicle lengths of different vehicle types.
It can be understood that radar data can easily identify the same vehicle track as different tracks, so that short tracks appear, meanwhile, the accuracy of normal tracks is affected, data in the tracks are updated through matching of the short tracks and the normal tracks, and the accuracy of the tracks is improved, namely, the similarity of the tracks is judged through improved French distance, the smaller the value of the French distance is, the higher the similarity between the two tracks is, the data in the tracks meeting the conditions are updated, the data of track sequences are restored, and the accuracy of the data in the obtained vehicle tracks are further improved.
And S3, after the vehicle track is subjected to abnormality removal and smoothing, extracting characteristic data, and adding a category label for the characteristic data to obtain sample data, wherein the category label comprises lane change or non-lane change.
In implementation, the abnormal removal and smoothing of the vehicle track is carried out, specifically:
if the number of track points of the track sequence in the vehicle track is greater than or equal to the length threshold value, the initial track of the vehicle keeps the track sequence, otherwise, the track sequence is deleted; the process is used for eliminating abnormal short tracks, and eliminating short tracks can reduce the influence of track reliability.
And carrying out track smoothing on the vehicle track after the abnormality is removed based on a Savitzky-Golay filter.
Specifically, the vehicle track after the abnormality removal is smoothed for three times at five points, namely, the transverse position and the longitudinal position of the vehicle for each track point in each track sequence are respectively taken from front to back, two adjacent track points are respectively taken, and the vehicle track is smoothed by adopting a cubic polynomial to approach.
It can be understood that by removing and smoothing the anomaly of the vehicle track, the anomaly data in the track is further removed, so that basic data is provided for subsequent feature extraction, and the accuracy of the later model training is improved.
Specifically, the length threshold is determined according to the statistical result of the initial track of the vehicle, and the taken length threshold can divide the initial track of the vehicle into an initial long track and an initial short track of the vehicle, wherein the initial long track and the initial short track have the same track number.
In practice, the feature data is extracted by:
and respectively taking the first track point and the last track point of the track sequence with the transverse instantaneous speed greater than the track changing threshold value in the track of the vehicle as a start track point and an end track point of the track changing section in the track sequence to obtain the track changing section of the track sequence. Wherein the lane change threshold is determined based on empirical values. It can be understood that the channel change interval is extracted, so that the subsequent channel change characteristic data extraction is more obvious, and the interference of straight-line driving sections with different lengths in the track is prevented.
The lane change interval based on the track sequence extracts the following characteristics:
offset angle:
average lane change transverse velocity v xca ：
If t e -t s Is not 0, then
Otherwise v xca ＝0；
Average lane change longitudinal speed v yca ：
If t e -t s Is not 0, then
Otherwise v yca ＝0；
Lateral offset Δx c ：
Track change time delta t c ：
Δt c ＝t e -t s ；
Wherein Δx and Δy represent the vehicle lateral position change amount and the vehicle longitudinal position change amount of the start track point and the end track point of the lane change section, respectively; t is t e And t s Respectively representing acquisition moments of a start track point and an end track point of a track change interval; v xi And v yi Respectively representing the transverse instantaneous speed and the longitudinal instantaneous speed of the track point at the acquisition time i;
and->
Respectively at the acquisition time t e And t s The vehicle transverse position of the trajectory point at that time.
And carrying out normalization processing on the extracted features to obtain feature data of a track sequence in the vehicle track.
Specifically, the normalization processing of the characteristic data adopts a min-max method, taking the lane change duration as an example, and the expression is as follows:
wherein Δt is cn Represents normalized lane change duration data, max (Δt c ) Maximum value, min (Δt c ) Representing the minimum value of the lane change duration in the vehicle trajectory.
And S4, training the support vector machine model based on the sample data to obtain an optimal support vector machine model.
Specifically, sample data is divided into a training set and a testing set, and an RBF function is selected as a kernel function by a support vector machine model, and parameters are determined through network searching and cross validation, specifically:
firstly, a candidate set of parameters of a support vector machine is given out, grid search is carried out, a plurality of candidate parameter combinations are given out according to priori knowledge, each parameter combination is tested, and a group of parameters with the best training effect is taken as optimal parameters; secondly, cross verification is carried out, and different training sets are adopted to train on the model with set parameters, so that result contingency caused by division of the training sets and the testing sets is reduced; the method comprises the steps of randomly extracting data in sample data into training sets and test sets of different combinations to obtain different training sets for cross-validation.
S5, inputting the initial track characteristic data of the vehicle to be identified into an optimal support vector machine model to obtain the identification result of the lane change behavior of the vehicle. Preferably, the feature data is extracted after the initial track of the vehicle to be identified is smoothed.
Compared with the prior art, the vehicle lane change behavior recognition method based on the expressway radar data provided by the invention has the advantages that the vehicle initial track is generated based on the radar data, the vehicle track is obtained through cleaning and track matching restoration, the characteristic data is extracted after abnormality removal and smoothing to obtain sample data, the sample data is used for training the support vector machine model, the characteristic data to be recognized is finally input into the optimal support vector machine model to obtain the recognition result of the vehicle lane change behavior, the problem that the recognition result error is large due to the lack of the prior process of the radar data in the existing method is solved, and the reliability and the precision of the recognition result are improved.
Example 2
Another embodiment 2 of the present invention discloses a vehicle lane change behavior recognition system based on expressway radar data, including:
the data acquisition module is used for acquiring radar data of the expressway and generating an initial track of the vehicle based on the radar data.
And the vehicle track generation module is used for carrying out matching on the basis of an improved French distance track matching algorithm after the initial track of the vehicle is cleaned, so as to obtain the vehicle track.
And the sample data generation module is used for extracting characteristic data after the abnormal removal and smoothing of the vehicle track, and adding a category label for the characteristic data to obtain sample data, wherein the category label comprises lane change or non-lane change.
And the model training module is used for training the support vector machine model based on the sample data to obtain an optimal support vector machine model.
The recognition module is used for inputting the initial track characteristic data of the vehicle to be recognized into the optimal support vector machine model to obtain the recognition result of the lane change behavior of the vehicle.
It should be noted that, since the relevant portions of the vehicle lane change behavior recognition system and the foregoing vehicle lane change behavior recognition method in this embodiment may be referred to each other, repeated descriptions are repeated here, so that the description is omitted here. The system embodiment has the same principle as the method embodiment, so the system also has the corresponding technical effects of the method embodiment.
Example 3
To verify the effectiveness of the present invention in examples 1 and 2, a specific example 3 of the present invention uses Tang harbor high speed radar monitoring data, as shown in the following table, wherein X represents the lateral position of the vehicle, Y represents the longitudinal position of the vehicle, and V x Representing transverse instantaneous velocity and V y Representing a longitudinal instantaneous speed; the radar detector is fixed on the portal frame at the road side end in the same way as other equipment, can detect the vehicle operation data of two-way six lanes and emergent lane, detects the highway section total length 268.92m.
| Time stamp | Target ID | X (Rice) | Y (Rice) | V x (m/s) | V y (m/s) | Vehicle model |
| 2019-12-18_11-12-48.0925 | 4 | -6.98 | 232.84 | 0.85 | -28 | 1 |
| 2019-12-18_11-12-48.0925 | 5 | 10.26 | 53.27 | -0.42 | 17.8 | 2 |
| 2019-12-18_11-12-48.1675 | 4 | -7.47 | 230.64 | 0.13 | -28.16 | 1 |
| 2019-12-18_11-12-48.1675 | 5 | 10.3 | 54.08 | -0.37 | 17.76 | 2 |
| 2019-12-18_11-12-48.2445 | 3 | -7.13 | 169.74 | 0.46 | -32.28 | 1 |
| 2019-12-18_11-12-48.2445 | 4 | -7.39 | 228.6 | -0.15 | -28.23 | 1 |
| 2019-12-18_11-12-48.2445 | 5 | 10.29 | 55.58 | -0.3 | 17.84 | 2 |
| 2019-12-18_11-12-48.2445 | 6 | -4.35 | 241.08 | 0.44 | -31.01 | 1 |
| 2019-12-18_11-12-48.2445 | 8 | 12.79 | 49.47 | -0.25 | 17.94 | 1 |
| 2019-12-18_11-12-48.3185 | 2 | -10.47 | 150.44 | -0.21 | -21.78 | 2 |
From the radar data, the number of obtained vehicle tracks is 1009, the number of lane change tracks is 34, and the number of non-lane change tracks is 975. Dividing the training set and the testing set according to the proportion of 7:3, wherein category sampling is adopted to ensure that the training set and the testing set have track change sample data with a certain proportion.
In this embodiment, the characteristic data (denoted as C) of the average lane change longitudinal speed is removed x ) And removing characteristic data of average lane change lateral speed (denoted as C y ) With the characteristic data (denoted as C) of the present invention containing five characteristics all ) And performing comparative analysis, setting the training times of the support vector machine to be 30, and performing model recognition analysis under three conditions of an Intel i5-6300HQ processor, a system memory of 8.0GB, a Windows10 (64 bits) system and a program language of Python3.7 on an experimental platform as shown in the following table.
| Accuracy (%) | Recall (%) | Accuracy (%) | |
| C x | 75.13 | 72.00 | 97.79 |
| C y | 78.88 | 81.33 | 98.24 |
| C all | 86.97 | 88.00 | 98.50 |
As can be seen from the above table:
the accuracy of the support vector machine model trained under the characteristic data lacking one characteristic is high, which reaches 97.79% and 98.24% respectively, but the accuracy and recall rate are low, because the non-lane change track is easy to identify, but the lane change track is difficult to identify, and the samples are few. The difference in average lane change lateral velocity is not significant, so C will be x The feature combination is changed into C y After training the model, the accuracy is improved by 3.75 percentThe recall rate is improved by 9.33%, the probability of recognizing the track change error as a normal track is greatly reduced, and the generalization capability of the model is enhanced; when all five features are used to train the support vector machine model, the precision and recall are further improved and approach 90%. Therefore, the method of the invention processes the radar data and extracts the characteristic data to carry out the training of the support vector machine model, so that the model has certain universality, and the identification result can provide decision basis for the service area management department.
Those skilled in the art will appreciate that all or part of the flow of the methods of the embodiments described above may be accomplished by way of a computer program to instruct associated hardware, where the program may be stored on a computer readable storage medium. Wherein the computer readable storage medium is a magnetic disk, an optical disk, a read-only memory or a random access memory, etc.
The present invention is not limited to the above-mentioned embodiments, and any changes or substitutions that can be easily understood by those skilled in the art within the technical scope of the present invention are intended to be included in the scope of the present invention.
Claims (7)
1. The vehicle channel change behavior recognition method based on the expressway radar data is characterized by comprising the following steps of:
acquiring radar data of a highway, and generating an initial track of a vehicle based on the radar data;
the radar data comprises acquisition time, target ID, vehicle transverse position, vehicle longitudinal position, transverse instantaneous speed and longitudinal instantaneous speed;
after the initial track of the vehicle is cleaned, matching is carried out based on an improved Frechet distance track matching algorithm, and a vehicle track is obtained; the method comprises the following steps:
selecting a track sequence with the number of track points larger than or equal to a length threshold value in the initial track of the vehicle as an initial long track of the vehicle, otherwise, selecting a track sequence with the number of track points larger than or equal to a length threshold value as an initial short track of the vehicle;
calculating the Frechet distance between each track sequence in the initial short track of the vehicle and each track in the initial long track of the vehicle;
if the Frechet distance is smaller than the set similarity threshold, averaging corresponding data in the two track sequences, and updating the track sequences;
wherein, the Frechet distance F (A, B) between the track sequence A in the initial short track and the track sequence B in the initial long track is calculated by the following formula:
F(A,B)＝inf{max{d(α(t),β(t))}+kσ(d)}；
where d (α (t), β (t))=y At -y Bt +ΔT(v At -v Bt )；
Wherein d (α (t), β (t)) represents the distance between the track points α (t) and β (t) of the track sequence a and the track sequence B at the time t, y At And y Bt Respectively representing the longitudinal position of the vehicle at the time t, v, of the track sequence A and the track sequence B At And v Bt The longitudinal instantaneous speeds of the track sequence A and the track sequence B at the moment T are respectively represented, delta T represents the acquisition time interval of radar data, k represents the discrete weight coefficient, sigma (d) represents the standard deviation of the distances of the track sequence A and the track sequence B at all moments, and inf { } represents the maximum lower bound of the function value;
after abnormality removal and smoothing are carried out on the vehicle track, extracting feature data, and adding a category label for the feature data to obtain sample data, wherein the category label comprises lane change or non-lane change;
training a support vector machine model based on the sample data to obtain an optimal support vector machine model;
and inputting the initial track characteristic data of the vehicle to be identified into an optimal support vector machine model to obtain the identification result of the lane change behavior of the vehicle.
2. The method for identifying vehicle lane change behavior based on highway radar data according to claim 1, wherein the vehicle initial trajectory is generated based on the radar data by:
traversing radar data in sequence according to acquisition time:
initializing a first piece of data which is not added to the track sequence in the radar data as a first track point of a j=1th track sequence, and sequentially finding a next track point in the track sequence in the radar data: if there is data which is the same as the target ID of the first track point in the track sequence and is not added to the track sequence, judging the piece of data according to the following formula:
wherein x is z And y z The lateral and longitudinal positions of the vehicle, x, respectively, representing the last track point in the track sequence un And y un A vehicle lateral position and a longitudinal position respectively representing data in the radar data which is the same as the target ID of the first track point in the track sequence and which is not added to the track sequence;
if so, the piece of data is added to the track sequence, otherwise the piece of track ends, and j=j+1.
3. The method for identifying the lane-changing behavior of a vehicle based on highway radar data according to claim 1, wherein the cleaning of the initial trajectory of the vehicle is performed by power constraint, in particular:
each track sequence in the initial track of the vehicle is subjected to power constraint:
calculating the velocity v of adjacent track points in the track sequence by k ：
Wherein x is k And y k A vehicle transverse position and a vehicle longitudinal position respectively representing a kth track point of the track sequence; x is x k+1 And y k+1 The vehicle lateral position and the vehicle longitudinal position of the (k+1) th track point of the track sequence are respectively represented; t is t k And t k+1 Respectively represent the kth sum of the track sequenceAcquisition time of the (k+1) th track point;
calculating the acceleration a of adjacent speed according to the speeds of adjacent track points in the track sequence k ：
If all track points in the track sequence meet v k ＜v 0 And a k ＜a 0 The initial track of the vehicle keeps the track sequence, otherwise, the track sequence is deleted; wherein v is 0 And a 0 A speed threshold and an acceleration threshold, respectively.
4. The method for identifying the lane-changing behavior of a vehicle based on highway radar data according to claim 1, wherein the anomaly removal and smoothing of the vehicle track is performed, specifically:
if the number of track points of the track sequence in the vehicle track is greater than or equal to the length threshold value, the initial track of the vehicle keeps the track sequence, otherwise, the track sequence is deleted;
and carrying out track smoothing on the vehicle track after the abnormality is removed based on a Savitzky-Golay filter.
5. The method for identifying a lane change behavior of a vehicle based on highway radar data according to claim 1, wherein the feature data is extracted by:
the method comprises the steps of respectively taking a first track point and a last track point, of which the transverse instantaneous speed is greater than a track changing threshold value, in a track sequence of a vehicle as a start track point and an end track point of a track changing section in the track sequence to obtain the track changing section of the track sequence;
the lane change interval based on the track sequence extracts the following characteristics:
offset angle:
average lane change transverse velocity v xca ：
If t e -t s Is not 0, then
Otherwise v xca ＝0；
Average lane change longitudinal speed v yca ：
If t e -t s Is not 0, then
Otherwise v yca ＝0；
Lateral offset Δx c ：
Track change time delta t c ：
Δt c ＝t e -t s ；
Wherein Δx and Δy represent the vehicle lateral position change amount and the vehicle longitudinal position change amount of the start track point and the end track point of the lane change section, respectively; t is t e And t s Respectively representing acquisition moments of a start track point and an end track point of a track change interval; v xi And v yi Respectively representing the transverse instantaneous speed and the longitudinal instantaneous speed of the track point at the acquisition time i;
and->
Respectively at the acquisition time t e And t s The vehicle lateral position of the locus point at that time;
and carrying out normalization processing on the extracted features to obtain feature data of a track sequence in the vehicle track.
6. A vehicle lane change behavior recognition system based on highway radar data, comprising:
the data acquisition module is used for acquiring radar data of the expressway and generating an initial track of the vehicle based on the radar data;
the radar data comprises acquisition time, target ID, vehicle transverse position, vehicle longitudinal position, transverse instantaneous speed and longitudinal instantaneous speed;
the vehicle track generation module is used for carrying out matching on the basis of an improved French distance track matching algorithm after the initial track of the vehicle is cleaned, so as to obtain a vehicle track; the method comprises the following steps:
selecting a track sequence with the number of track points larger than or equal to a length threshold value in the initial track of the vehicle as an initial long track of the vehicle, otherwise, selecting a track sequence with the number of track points larger than or equal to a length threshold value as an initial short track of the vehicle;
calculating the Frechet distance between each track sequence in the initial short track of the vehicle and each track in the initial long track of the vehicle;
if the Frechet distance is smaller than the set similarity threshold, averaging corresponding data in the two track sequences, and updating the track sequences;
wherein, the Frechet distance F (A, B) between the track sequence A in the initial short track and the track sequence B in the initial long track is calculated by the following formula:
F(A,B)＝inf{max{d(α(t),β(t))}+kσ(d)}；
where d (α (t), β (t))=y At -y Bt +ΔT(v At -v Bt )；
Wherein d (α (t), β (t)) represents the distance between the track points α (t) and β (t) of the track sequence a and the track sequence B at the time t, y At And y Bt Respectively representing the longitudinal position of the vehicle at the time t, v, of the track sequence A and the track sequence B At And v Bt The longitudinal instantaneous speeds of the track sequence A and the track sequence B at the moment T are respectively represented, delta T represents the acquisition time interval of radar data, k represents the discrete weight coefficient, sigma (d) represents the standard deviation of the distances of the track sequence A and the track sequence B at all moments, and inf { } represents the maximum lower bound of the function value;
the sample data generation module is used for extracting characteristic data after abnormality removal and smoothing of the vehicle track, and adding a category label for the characteristic data to obtain sample data, wherein the category label comprises lane change or non-lane change;
the model training module is used for training the support vector machine model based on the sample data to obtain an optimal support vector machine model;
the recognition module is used for inputting the initial track characteristic data of the vehicle to be recognized into the optimal support vector machine model to obtain the recognition result of the lane change behavior of the vehicle.
7. The highway radar data based vehicle lane change behavior recognition system of claim 6, wherein the radar data comprises acquisition time, target ID, vehicle lateral position, vehicle longitudinal position, lateral instantaneous speed, and longitudinal instantaneous speed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110997952.5A CN115731261B (en) | 2021-08-27 | 2021-08-27 | Vehicle lane change behavior recognition method and system based on expressway radar data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110997952.5A CN115731261B (en) | 2021-08-27 | 2021-08-27 | Vehicle lane change behavior recognition method and system based on expressway radar data |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115731261A CN115731261A (en) | 2023-03-03 |
| CN115731261B true CN115731261B (en) | 2023-06-16 |
Family
ID=85290420
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110997952.5A Active CN115731261B (en) | 2021-08-27 | 2021-08-27 | Vehicle lane change behavior recognition method and system based on expressway radar data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115731261B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102011016770A1 (en) * | 2011-04-12 | 2011-11-03 | Daimler Ag | Method for assisting vehicle driver for lane change from current lane to adjacent target lane, involves changing intervention into steering device, brake device and drive train if sufficient space for vehicle for lane change is established |
| CN111273668A (en) * | 2020-02-18 | 2020-06-12 | 福州大学 | Unmanned vehicle motion track planning system and method for structured road |
Family Cites Families (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9977123B2 (en) * | 2014-05-20 | 2018-05-22 | Bae Systems Information And Electronic Systems Integration Inc. | Automated track projection bias removal using frechet distance and road networks |
| CN106777776A (en) * | 2017-01-10 | 2017-05-31 | 长沙理工大学 | A kind of vehicle lane-changing decision-making technique based on supporting vector machine model |
| WO2019048034A1 (en) * | 2017-09-06 | 2019-03-14 | Swiss Reinsurance Company Ltd. | Electronic logging and track identification system for mobile telematics devices, and corresponding method thereof |
| CN107826118B (en) * | 2017-11-01 | 2019-08-30 | 南京阿尔特交通科技有限公司 | A kind of method and device differentiating abnormal driving behavior |
| US11675359B2 (en) * | 2018-06-13 | 2023-06-13 | Nvidia Corporation | Path detection for autonomous machines using deep neural networks |
| US11194331B2 (en) * | 2018-10-30 | 2021-12-07 | The Regents Of The University Of Michigan | Unsupervised classification of encountering scenarios using connected vehicle datasets |
| CN111291141B (en) * | 2018-12-07 | 2023-06-20 | 阿里巴巴集团控股有限公司 | Track similarity determination method and device |
| CN109816606B (en) * | 2019-01-18 | 2022-01-04 | 中国科学院空天信息创新研究院 | Method for tracking target by using optical remote sensing satellite |
| CN110244713B (en) * | 2019-05-22 | 2023-05-12 | 江苏大学 | Intelligent vehicle lane change track planning system and method based on artificial potential field method |
| CN110466516B (en) * | 2019-07-11 | 2020-08-28 | 北京交通大学 | Curve road automatic vehicle lane change track planning method based on nonlinear programming |
| CN110569783B (en) * | 2019-09-05 | 2022-03-25 | 吉林大学 | Method and system for identifying lane changing intention of driver |
| CN111143495B (en) * | 2019-12-18 | 2024-01-30 | 北京中交兴路车联网科技有限公司 | Road missing discovery method and device |
| CN111209838B (en) * | 2019-12-31 | 2022-11-18 | 清华大学 | Driving intention-based dynamic identification method for lane change behavior of vehicles |
| CN111932887B (en) * | 2020-08-17 | 2022-04-26 | 武汉四维图新科技有限公司 | Method and equipment for generating lane-level track data |
-
2021
- 2021-08-27 CN CN202110997952.5A patent/CN115731261B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE102011016770A1 (en) * | 2011-04-12 | 2011-11-03 | Daimler Ag | Method for assisting vehicle driver for lane change from current lane to adjacent target lane, involves changing intervention into steering device, brake device and drive train if sufficient space for vehicle for lane change is established |
| CN111273668A (en) * | 2020-02-18 | 2020-06-12 | 福州大学 | Unmanned vehicle motion track planning system and method for structured road |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115731261A (en) | 2023-03-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113486822B (en) | Surrounding vehicle track prediction method and system based on driving intention | |
| CN103473540B (en) | The modeling of intelligent transportation system track of vehicle increment type and online method for detecting abnormality | |
| CN111291790B (en) | Turning path extraction and road network topology change detection framework method based on track similarity | |
| CN112053589B (en) | Target vehicle lane changing behavior adaptive identification model construction method | |
| CN106956680B (en) | Electric automobile driving behavior recognition analysis method | |
| JP2013228259A (en) | Object identification device and object identification method | |
| CN105448108A (en) | Overspeed discrimination method based on hidden Markov road network matching | |
| CN113127466B (en) | Vehicle track data preprocessing method and computer storage medium | |
| CN104951764A (en) | Identification method for behaviors of high-speed vehicle based on secondary spectrum clustering and HMM (Hidden Markov Model)-RF (Random Forest) hybrid model | |
| CN110569855A (en) | Long-time target tracking algorithm based on correlation filtering and feature point matching fusion | |
| CN108344997B (en) | Road guardrail rapid detection method based on trace point characteristics | |
| CN110736999A (en) | Railway turnout detection method based on laser radar | |
| CN112560915A (en) | Urban expressway traffic state identification method based on machine learning | |
| Cassity et al. | Applying weibull distribution and discriminant function techniques to predict damaged cup anemometers in the 2011 PHM competition | |
| US20200174488A1 (en) | False target removal device and method for vehicles and vehicle including the device | |
| CN115731261B (en) | Vehicle lane change behavior recognition method and system based on expressway radar data | |
| CN113269042B (en) | Intelligent traffic management method and system based on driving vehicle violation identification | |
| Maghsood et al. | Detection of steering events based on vehicle logging data using hidden Markov models | |
| CN111796250A (en) | False trace point multi-dimensional hierarchical suppression method based on risk assessment | |
| CN107622667B (en) | Method and system for detecting altered license plate number | |
| CN113722982A (en) | Automobile sensor attack detection and defense method based on multi-model fusion | |
| CN113157761B (en) | Electric vehicle charging and discharging fault analysis method based on association rule mining | |
| CN109448377A (en) | A method of vehicle safety evaluation is carried out using satellite location data | |
| CN117237911A (en) | Image-based dynamic obstacle rapid detection method and system | |
| CN116680519A (en) | GM-HMM model-based vehicle lane change intention recognition method |
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 |