CN112749210B - Vehicle collision recognition method and system based on deep learning - Google Patents

Vehicle collision recognition method and system based on deep learning Download PDF

Info

Publication number
CN112749210B
CN112749210B CN202110061524.1A CN202110061524A CN112749210B CN 112749210 B CN112749210 B CN 112749210B CN 202110061524 A CN202110061524 A CN 202110061524A CN 112749210 B CN112749210 B CN 112749210B
Authority
CN
China
Prior art keywords
data
vehicle
time
model
risk
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110061524.1A
Other languages
Chinese (zh)
Other versions
CN112749210A (en
Inventor
朱瑞
何道峰
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.)
Ubiai Information Technology Beijing Co ltd
Original Assignee
Ubiai Information Technology Beijing 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 Ubiai Information Technology Beijing Co ltd filed Critical Ubiai Information Technology Beijing Co ltd
Priority to CN202110061524.1A priority Critical patent/CN112749210B/en
Publication of CN112749210A publication Critical patent/CN112749210A/en
Application granted granted Critical
Publication of CN112749210B publication Critical patent/CN112749210B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • 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
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2474—Sequence data queries, e.g. querying versioned data
    • 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
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2477—Temporal data queries
    • 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
    • G06F16/25—Integrating or interfacing systems involving database management systems
    • G06F16/254—Extract, transform and load [ETL] procedures, e.g. ETL data flows in data warehouses
    • 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
    • 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
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services
    • G06Q50/265—Personal security, identity or safety
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/40—Business processes related to the transportation industry
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • Business, Economics & Management (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Molecular Biology (AREA)
  • Primary Health Care (AREA)
  • Probability & Statistics with Applications (AREA)
  • Economics (AREA)
  • Fuzzy Systems (AREA)
  • Human Resources & Organizations (AREA)
  • Computing Systems (AREA)
  • Strategic Management (AREA)
  • Evolutionary Computation (AREA)
  • Marketing (AREA)
  • Biophysics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Remote Sensing (AREA)
  • Educational Administration (AREA)
  • Traffic Control Systems (AREA)
  • Development Economics (AREA)
  • Computer Security & Cryptography (AREA)

Abstract

The invention relates to a vehicle collision recognition method and system based on deep learning, wherein the vehicle collision recognition method comprises the following steps: step 1: collecting real-time running data of a vehicle, and acquiring historical running data and historical risk-giving data of the vehicle; step 2: preprocessing vehicle operation data; step 3: performing low-frequency to high-frequency data interpolation on vehicle operation data; step 4: ETL engineering feature transformation; step 5: model training and iteration are carried out through a deep learning method and a mathematical statistics correlation principle, and a correlation model is obtained; step 6: and according to the real-time running data of the vehicle and the combination of the related models, obtaining a vehicle collision recognition result. The method adopts mass internet of vehicles data, models from 5 angles of abnormal running track, abnormal speed reduction, predicted braking distance, abnormal parking and abnormal turning, and identifies collision accidents, so that judgment trouble caused by complex geographic factors is avoided.

Description

Vehicle collision recognition method and system based on deep learning
Technical Field
The invention relates to the technical field of vehicle collision detection, in particular to a vehicle collision recognition method and system based on deep learning.
Background
Along with the rapid increase of the number of automobiles in China, road traffic accidents are gradually frequent, thereby causing serious casualties and property loss, and the road traffic safety situation is severe, thereby attracting attention of the whole society. A large number of surveys show that after road traffic accidents, particularly after collision accidents, casualties caused by untimely rescue are far greater than casualties caused by accidents directly, so that the requirement of timely learning of vehicle collision accidents and obtaining rescue is particularly necessary. Due to the limitation of data collection conditions, in the prior art, the method for detecting the collision mainly uses a threshold value of a trigger gravity acceleration sensor as a judging condition for detecting the collision. Many current detection methods read the values of the three axes of the acceleration sensor X, Y, Z in a polling manner, calculate the difference between the maximum value and the minimum value of each axis, and calculate the vectors of the three axes. By comparing this vector value with a preset value, a collision is considered to occur if the vector value is greater than the preset value. However, when the vehicle is running on a complex road condition, the detection technology can have false alarm or missing alarm. For example, if the road condition exists in a sewer manhole cover or a deceleration strip, the vehicle can generate larger vibration when passing through the places, and some vehicle owners can generate larger braking force when passing through the places, so that the acceleration sensor arranged on the vehicle generates larger difference value, and false alarm is easy to generate. Meanwhile, because the parameters of the vehicle bodies of different types of vehicles are different, and the collision preset values of different products are also different, the collision detection technology is very difficult to realize in terms of use and maintenance aiming at various vehicle types and various application scenes, the accuracy of the output result is insufficient, a large number of misjudgment or missed judgment can be caused, and the correct judgment can not be timely made when an accident occurs. Some methods for detecting collision, such as a vehicle-mounted device judging method based on an OBD interface, have the defects of being not compatible with all vehicle types, high in device manufacturing cost, simple in detection algorithm and few in covering collision scene; the vehicle-mounted equipment judging method based on the video image has the defects of high equipment manufacturing cost, easy equipment damage in the collision process, low detection and calculation cost performance, less coverage on collision scenes and low detection precision.
Disclosure of Invention
In view of the above, the present invention aims to overcome the shortcomings of the prior art, and provide a vehicle collision recognition method and system based on deep learning.
In order to achieve the above purpose, the invention adopts the following technical scheme: a vehicle collision recognition method based on deep learning, comprising:
step 1: collecting real-time running data of a vehicle, and acquiring historical running data and historical risk-giving data of the vehicle;
step 2: preprocessing vehicle operation data;
step 3: performing low-frequency to high-frequency data interpolation processing on vehicle operation data;
step 4: ETL engineering feature conversion treatment;
step 5: model training and iteration are carried out through a deep learning method and a mathematical statistics correlation principle, and a correlation model is obtained;
step 6: and according to the real-time running data of the vehicle and the combination of the related models, obtaining a vehicle collision recognition result.
Optionally, the acquiring the vehicle history operation data and the vehicle history risk-free data includes:
acquiring historical operating data of the vehicle, the data comprising: time stamp, GPS speed, GPS direction, GPS altitude, device power status, device positioning status, vehicle movement status, GPS longitude, and GPS latitude;
acquiring historical risk information of the vehicle by a third party, wherein the risk information comprises: time of risk, place of risk and vehicle collision location;
the method comprises the steps of matching the risk information with historical operation data of the vehicle through risk time, intercepting different time sequence data with the time length of Ts containing risk moment through a data sliding method, marking the time sequence data as risk, intercepting a certain amount of non-risk time sequence data, and marking the time sequence data as non-risk.
Optionally, the preprocessing the vehicle operation data includes:
carrying out time sequence confusion processing on vehicle operation data;
performing GPS positioning drift outlier processing on the data processed by the timing disorder;
and carrying out signal loss data processing on the data processed by the GPS positioning drift abnormal value, and deleting the data which has no positioning signal and has the speed of 0.
Optionally, the performing timing sequence confusion processing on the vehicle operation data includes:
for the historical data used for model training, carrying out mode calculation on the minute min records of the TS time sequence data, then comparing each record with the current mode, calculating a time stamp difference value, and if the difference value is larger than the preset time, considering the record as carrying out complement operation on the previous data, and deleting the record;
for time sequence data calculated in real time on line of the model, calculating a time difference value between each data time stamp and the current system time, and if the time difference value is larger than the preset time, considering the data as operation of system complement data, and deleting the data;
and sorting the historical data or the data analyzed in real time after the processing is finished according to the time stamp.
Optionally, the processing the abnormal value of the GPS positioning drift of the data after the processing of the timing disorder includes:
sequentially calculating the time interval and the linear distance of the earth between the t and t+1 records;
acquiring suspicious linear distance;
and positioning the drift point according to the suspicious straight line distance, the time interval and the earth straight line distance.
Optionally, the performing low-frequency to high-frequency data interpolation processing on the vehicle operation data includes:
performing low-frequency to high-frequency data interpolation processing on vehicle operation data by utilizing a differential integral and Bezier curve method;
specifically, a differential integration method is utilized to conduct speed item interpolation processing on vehicle operation data;
and performing GPS longitude and latitude item interpolation on the vehicle operation data subjected to the speed item interpolation processing by using a Bezier curve method.
Optionally, the model training and iteration are performed by using a deep learning method and a principle related to mathematical statistics, including:
dividing a training set and a testing set of the processed historical data;
LSTM model parameters are used for carrying out LSTM modeling on training set data, LSTM model optimization and evaluation parameters are used for evaluating the prediction results of the testing set data, and optimization iteration is carried out on the model parameters.
Optionally, the correlation model includes: a running track anomaly identification model, a sudden braking anomaly identification model, a turning anomaly identification model, a parking anomaly identification model and a braking distance anomaly model;
the method for obtaining the identification result according to the real-time operation data of the vehicle and by combining the related model comprises the following steps:
dividing the real-time running data of the vehicle according to the time interval of Ts;
carrying out real-time processing on the divided TS duration data according to the steps 2, 3 and 4;
inputting the processed data into a running track anomaly identification model, a sudden braking anomaly identification model, a turning anomaly identification model, a parking anomaly identification model and a braking distance anomaly model to respectively obtain driving anomaly scores of the models;
and determining the recognition result according to the driving abnormality score of each model.
Optionally, the determining the recognition result according to the driving anomaly score of each model includes:
obtaining the sum of products of driving abnormality scores and corresponding weights of the models, and taking the sum as a predictive score;
judging whether the predictive value is larger than an accident threshold value or not;
if the predictive value is greater than the accident threshold, judging that the vehicle has a collision accident; otherwise, judging that the vehicle has no collision accident.
The invention also provides a vehicle collision recognition system based on deep learning, which comprises:
the data acquisition module is used for acquiring real-time running data of the vehicle and acquiring historical running data and historical risk-giving data of the vehicle;
the preprocessing module is used for preprocessing vehicle operation data;
the data interpolation processing module is used for performing low-frequency to high-frequency data interpolation processing on the vehicle operation data;
the conversion module is used for ETL engineering characteristic conversion treatment;
the modeling module is used for carrying out model training and iteration through a deep learning method and a mathematical statistics correlation principle to obtain a correlation model;
and the recognition module is used for obtaining a vehicle collision recognition result according to the processed real-time vehicle operation data and the combination of the related models.
The invention adopts the technical scheme that the vehicle collision recognition method based on deep learning comprises the following steps: step 1: collecting real-time running data of a vehicle, and acquiring historical running data and historical risk-giving data of the vehicle; step 2: preprocessing vehicle operation data; step 3: performing low-frequency to high-frequency data interpolation processing on vehicle operation data; step 4: ETL engineering feature conversion treatment; step 5: model training and iteration are carried out through a deep learning method and a mathematical statistics correlation principle, and a correlation model is obtained; step 6: and according to the real-time running data of the vehicle and the combination of the related models, obtaining a vehicle collision recognition result. According to the identification method, the accuracy of the identification result is improved by adopting mass internet of vehicles data; in the model building process, the driving behavior factors, sex, weather conditions and other factors of the vehicle owner are added, so that the judging result is more accurate; the model identifies collision accidents from 5 angles of abnormal running track, abnormal speed reduction, prediction of braking distance, abnormal parking and abnormal direction change, and the judging angle is more comprehensive; the model judges whether a vehicle collision accident occurs according to the result presentation of time sequence data and the change rule of the data, replaces the mode that the traditional algorithm only uses one data record at a certain moment to judge, and has more reliable judging result; the identification method can avoid judgment trouble caused by complex geographic factors when the traditional equipment is used for collision induction identification.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of a vehicle collision recognition method based on deep learning according to an embodiment of the present invention;
FIG. 2 is a schematic flow chart of a vehicle collision recognition method based on deep learning according to a second embodiment of the present invention;
FIG. 3 is a schematic illustration of a vehicle traveling at constant speed;
FIG. 4 is a schematic illustration of a vehicle traveling at non-uniform speed;
FIG. 5 is a schematic illustration of first order Bezier curve interpolation GPS record points;
FIG. 6 is a schematic diagram of a third order Bezier curve interpolation GPS record;
fig. 7 is a schematic structural diagram of a vehicle collision recognition system based on deep learning according to an embodiment of the present invention.
In the figure: 1. a data acquisition module; 2. a preprocessing module; 3. a data interpolation processing module; 4. a conversion module; 5. a modeling module; 6. and an identification module.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be described in detail below. It will be apparent that the described embodiments are only some, but not all, embodiments of the invention. All other embodiments, based on the examples herein, which are within the scope of the invention as defined by the claims, will be within the scope of the invention as defined by the claims.
Fig. 1 is a schematic flow chart of a vehicle collision recognition method based on deep learning according to an embodiment of the present invention.
As shown in fig. 1, a vehicle collision recognition method based on deep learning according to the present embodiment includes:
step 1: collecting real-time running data of a vehicle, and acquiring historical running data and historical risk-giving data of the vehicle;
step 2: preprocessing vehicle operation data;
step 3: performing low-frequency to high-frequency data interpolation processing on vehicle operation data;
step 4: ETL engineering feature conversion treatment;
step 5: model training and iteration are carried out through a deep learning method and a mathematical statistics correlation principle, and a correlation model is obtained;
step 6: and according to the real-time running data of the vehicle and the combination of the related models, obtaining a vehicle collision recognition result.
In practical use, as shown in fig. 2, the vehicle collision recognition method of the present embodiment specifically includes the following steps:
step 1: collecting real-time running data of the vehicle through front-loading equipment and rear-loading equipment, and obtaining historical danger-giving data of the vehicle through a 4s dealer;
firstly, acquiring vehicle history real-time operation data, wherein the acquisition frequency is 1 time/5 seconds, and the method comprises the following steps: time stamp, GPS speed, GPS direction, GPS altitude, device power status, device positioning status, vehicle movement status, GPS longitude, and GPS latitude.
Second, historical risk information is obtained by third parties such as insurance companies, dealers and vehicle owners, including: the time of the accident, the accident location and the vehicle collision location.
Finally, the risk information and the vehicle history real-time operation data are matched through risk time, different time sequence data with the time length of Ts (which can be set to 120 s) and containing the risk moment are intercepted through a data sliding method, the time sequence data are marked as risk, and meanwhile, a certain amount of non-risk time sequence data are intercepted and marked as non-risk.
Step 2: preprocessing vehicle operation data, wherein the preprocessing comprises the following steps: the method comprises the following specific processes of time sequence confusion processing, GPS positioning drift outlier processing and signal loss data processing:
1) First, timing chaotic processing: because of the influence of the defects of the data acquisition equipment, individual records can not be acquired in real time and transmitted into the system, and the records are returned and stacked after a certain time, so that misjudgment of a model on collision accidents is caused, 120s of time sequence data are required to be screened and identified, and the method comprises the following steps:
for the historical data used for model training, performing mode calculation on the minute min records of the 120s time sequence data, comparing each record with the current mode, calculating a time stamp difference value, and if the difference value is larger than a preset time (such as 250 s), considering the record as equipment to perform complement operation on the previous data, and processing the previous data by adopting a deletion method;
for the time sequence data calculated in real time on line of the model, calculating the time difference between each data time stamp and the current system time, and if the difference is larger than the preset time (such as 250 s), considering the time difference as the operation of system complement data, and further deleting the data; and finally, sorting the historical data processed and the data analyzed in real time according to the time stamp.
2) Secondly, GPS positioning drift outlier processing: because of the self reasons of the equipment, the problem of equipment positioning drift is caused, so that the abnormal amplification of the distance between the front point and the rear point of the GPS is caused, and the erroneous judgment of a model result is caused, and the equipment positioning drift judgment method is as follows:
(1) Sequentially calculating the time interval delta and the linear distance of the earth between the t and the t+1 records;
(2) Obtaining suspicious linear distance (distance > 5000 m);
(3) Sequentially judging the suspicious distances in the step (2), if only one suspicious distance exists and the distance is obtained by 120s time sequence data t 120 And t 119 Calculating the corresponding speed of the travel, if the speed is more than 40m/s, judging t 120 Recording as a positioning drift point, and deleting the record; if the suspected distance is determined by t 1 And t 2 Calculated, the corresponding time interval delta of the travel and the distance of the next distance are determined t23 If delta < 120s and distance t23 Less than 1000m, t is considered 1 For the drift point, if Δ < 120s, the next distance interval > 4000m, then t is considered 2 For drift point, if the corresponding time interval delta of the travel is more than 120s, and the distance of the next distance interval t23 More than 4000m, the next distance corresponds to a time interval of < 120s, then t is considered 2 Is a drift point; if the suspicious distance is not t 1 And t 2 Calculated, not t 120 And t 119 If the corresponding time interval delta of the travel is smaller than 120s and the next distance interval is larger than 4000m, the second record in the distance is considered to be calculated as a drift point, and if the corresponding time interval delta of the distance is larger than 120s and the next distance interval is larger than 4000m and the corresponding time interval delta of the next distance is smaller than 120s, the second record in the distance is considered to be calculated as a drift point.
3) And finally, processing the lost signal data, and deleting the data which has no positioning signal and has the speed of 0.
Step 3: and performing low-frequency to high-frequency data interpolation processing on the vehicle data by using a differential integration and Bezier curve method.
Because the frequency of data collection of the equipment is less than 5s and the time of collision accident is much less than 5s at a moment, the record is needed to be interpolated, the data frequency is improved, and the following method is adopted:
1) And (5) speed item interpolation: first, assuming that t is recorded, the vehicle is traveling at a constant speed (as shown in fig. 3) between t+1, and the travel distance in the time interval is calculated by using the integration principle, and the formula is as follows:
secondly, the linear distance k of the earth between the two points is calculated by using the corresponding GPS points, if s > k, the vehicle is not driven at a constant speed in the driving process, but is decelerated and accelerated firstly (as shown in fig. 4), so that a record t3 is interpolated in the interval between t1 and t 2.
Two interpolation principles are specified:
a first strip: the sum of the two trapezoidal areas of fig. 4 is equal to s;
and a second strip: the abnormality of the original data is not increased, and as much as possible, two pieces of data are smoothed, so the sum of accelerations of two strokes |a1|+|a2| is as small as possible.
The calculation formula is as follows:
t＝t 1 +t 2
due to the fact that y= |a is to be made 1 |+|a 2 The I reaches the minimum, so
the interpolation data t3, v3 between t1 and t2 is as shown in the above results, and so on.
2) GPS longitude and latitude item interpolation:
the scheme utilizes the Bezier curve to approximately interpolate the GPS record between two GPS positioning points, and is specific:
if the GPS recording points are on the same straight line in 120s time sequence data, the GPS recording points are interpolated by adopting a first-order Bezier curve, and the calculation formula is as follows:
B(t)＝(1-t)P 0 +tP 1 ，t∈[0,1]
t is t1 in the velocity interpolation, and is specifically shown in fig. 5.
If the GPS recording points are not in the same straight line in 120s time sequence data, the GPS recording points are interpolated by adopting a third-order Bezier curve, and firstly, 4 GPS recording points G1 (x 1, y 1), G2 (x 2, y 2), G3 (x 3, y 3) and G4 (x 4, y 4) are assumed to exist,
Midpoint1_x＝(x1+x2)/2，Midpoint1_y＝(y1+y2)/2
Midpoint2_x＝(x2+x3)/2，Midpoint1_y＝(y2+y3)/2
Midpoint3_x＝(x3+x4)/2，Midpoint1_y＝(y3+y4)/2
K1＝d1/(d1+d2)
K2＝d2/(d2+d3)
split1＝midpoint1+K1*(midpoint2-midpoint1)
split2＝midpoint2+K2*(midpoint3-midpoint2)
translating the segment of the migrant 1 or 2 until the split1 coincides with the G2, wherein the coincident point is marked as V1, and the point after the migratory of the migratory 1 or 2 is marked as V2 or V3
p0＝V2+K1*(V1-V2)
p1＝V3+K1*(V1-V3)
Similarly, p2, p3 can be calculated.
Note that G1, p0 are control points between G1 and G2, p1, p2 are control points between G2 and G3, and p2, G4 are control points between G3 and G4.
The GPS record of interpolation between two points can be calculated according to the following three-order Bezier curve smoothing formula, for example, the interpolation between two AB points is as follows:
M(t)＝(1-t) 3 A+3tB(1-t) 2 +3Ct 2 (1-t)+Dt 3 ，t∈[0,1]
wherein A, D are respectively a starting point and an ending point, B and C are control points, M (t) is a GPS recording point interpolated in A and D. t is t1 in the velocity interpolation, and specifically, as shown in fig. 6.
Step 4: ETL engineering feature conversion
The high-frequency data obtained by interpolation in the step 3 is subjected to characteristic transformation, and is shown in a table 1, and the characteristics are specifically as follows:
TABLE 1
Step 5: model training and iteration are carried out through a deep learning method and a principle related to mathematical statistics, wherein the model comprises the following components: the system comprises a running track anomaly identification model, a sudden braking anomaly identification model, a turning anomaly identification model, a parking anomaly identification model and a braking distance anomaly model.
1) The training and iteration process of the sudden braking anomaly model comprises the following steps:
firstly, taking real-time operation data of 1000 vehicles for 15 days, respectively calculating the absolute value of the negative acceleration of the vehicles at the current speed, and establishing a two-dimensional relation model of the speed-the absolute value of the negative acceleration;
secondly, segmenting the speed according to a rounding, for example, v=1m/s-2 m/s, and finally calculating 95% quantile (which can be set according to actual needs) of the absolute value of the negative acceleration corresponding to each speed interval as a threshold value threshold of deceleration abnormality;
finally, calculating a shift behavior score p for the relative anomalies of the single deceleration behavior i ，p i The calculation method of (2) is as follows:
when a is i At > threshold, the score is 100; when a is i When the threshold is less than or equal to threshold, the score is recorded as p i *100。
It should be noted that, taking real-time running data of 1000 vehicles for 15 days as a sample, the number of vehicles and the number of days of the sample can be set according to actual needs, and the protection scope is not limited by the example.
2) The training and iteration principle of the direction-changing abnormal recognition model is the same as that of the sudden braking abnormal model.
3) The training and iteration process of the parking anomaly identification model comprises the following steps:
the abnormal parking is mainly identified according to the abnormal parking condition of the expressway, if 120s time sequence data has records with the speed of more than 110KM/h and records with the speed of less than 3.6KM/h (considered as parking records), meanwhile, the vehicle is parked for 10s before the vehicle is parked, the high-speed running records are stored, the vehicle is not started all the time within 120s time sequence after being parked, the behavior is judged to be the abnormal parking, and the abnormal score is 100.
4) The training and iteration process of the braking distance anomaly model comprises the following steps:
at the same speed, the minimum value calculation formula of the vehicle braking distance is as follows
d＝0.75*v+0.0256*v 2 ；
Calculating the difference value s between the actual deceleration distance and the minimum braking distance, wherein the braking distance anomaly score calculation formula is as follows
5) The training and iteration process of the running track anomaly identification model comprises the following steps:
and (3) performing deep learning LSTM model learning on the model data obtained after ETL engineering feature conversion in the step (4), wherein the LSTM (long-short-term memory network/Long short term memory) principle is the prior art, and the description is omitted here.
Specifically, the running track anomaly recognition model training and model prediction steps are as follows:
firstly, training sets and test set division are carried out on the history data processed in the step 4.
Secondly, LSTM model parameters are used for carrying out LSTM modeling on training set data, LSTM model optimization and evaluation parameters are used for evaluating the prediction results of test set data, and therefore optimization iteration is carried out on model parameters.
Sixth step: the model is put into use, vehicle operation data are collected in real time, the vehicle operation data collected in real time are processed by utilizing the steps 1, 2, 3 and 4, and the vehicle operation data are input into the model in the step 5.
Specifically, the model is put into use, real-time processing is carried out on the real-time data of the vehicle according to the time intervals of 120s and the data according to the steps 1, 2, 3 and 4, then the real-time data are input into the 5 models in the step 5, driving abnormality scores of the models are respectively obtained, weighting and summarizing are carried out according to a certain weight to obtain a predictive value, and finally whether the vehicle has a collision accident or not is judged according to the magnitude relation between the predictive value and an accident threshold. If the predictive value is greater than the accident threshold, judging that the vehicle has a collision accident; otherwise, judging that the vehicle has no collision accident.
According to the identification method, the accuracy of the identification result is improved by adopting mass internet of vehicles data; in the model building process, the driving behavior factors, sex, weather conditions and other factors of the vehicle owner are added, so that the judging result is more accurate; the model identifies collision accidents from 5 angles of abnormal running track, abnormal speed reduction, prediction of braking distance, abnormal parking and abnormal direction change, and the judging angle is more comprehensive; the model judges whether a vehicle collision accident occurs according to the result presentation of time sequence data and the change rule of the data, replaces the mode that the traditional algorithm only uses one data record at a certain moment to judge, and has more reliable judging result; the identification method can avoid judgment trouble caused by complex geographic factors when the traditional equipment is used for collision induction identification.
Fig. 7 is a schematic structural diagram of a vehicle collision recognition system based on deep learning according to an embodiment of the present invention.
As shown in fig. 7, a vehicle collision recognition system based on deep learning according to the present embodiment includes:
the data acquisition module 1 is used for acquiring real-time running data of the vehicle and acquiring historical running data and historical risk-giving data of the vehicle;
a preprocessing module 2 for preprocessing vehicle operation data;
a data interpolation processing module 3 for performing low-frequency to high-frequency data interpolation processing on the vehicle operation data;
the conversion module 4 is used for ETL engineering characteristic conversion treatment;
the modeling module 5 is used for carrying out model training and iteration through a deep learning method and a mathematical statistics correlation principle to obtain a correlation model;
and the recognition module 6 is used for obtaining a vehicle collision recognition result according to the processed real-time vehicle operation data and the combination of the related models.
The working principle of the vehicle collision recognition system based on deep learning in this embodiment is the same as that of the vehicle collision recognition method based on deep learning in any of the above embodiments, and will not be described here again.
The vehicle collision recognition system based on deep learning provided by the invention adopts massive vehicle networking data, models and recognizes collision accidents from 5 angles of abnormal running track, abnormal speed reduction, predicted braking distance, abnormal parking and abnormal direction change, avoids judgment trouble caused by complex geographic factors, and has the advantages of wide coverage of collision scenes and high detection precision.
It is to be understood that the same or similar parts in the above embodiments may be referred to each other, and that in some embodiments, the same or similar parts in other embodiments may be referred to.
It should be noted that in the description of the present invention, the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance. Furthermore, in the description of the present invention, unless otherwise indicated, the meaning of "plurality" means at least two.
Any process or method descriptions in flow charts or otherwise described herein may be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing specific logical functions or steps of the process, and further implementations are included within the scope of the preferred embodiment of the present invention in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art of the present invention.
It is to be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above-described embodiments, the various steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, may be implemented using any one or combination of the following techniques, as is well known in the art: discrete logic circuits having logic gates for implementing logic functions on data signals, application specific integrated circuits having suitable combinational logic gates, programmable Gate Arrays (PGAs), field Programmable Gate Arrays (FPGAs), and the like.
Those of ordinary skill in the art will appreciate that all or a portion of the steps carried out in the method of the above-described embodiments may be implemented by a program to instruct related hardware, where the program may be stored in a computer readable storage medium, and where the program, when executed, includes one or a combination of the steps of the method embodiments.
In addition, each functional unit in the embodiments of the present invention may be integrated in one processing module, or each unit may exist alone physically, or two or more units may be integrated in one module. The integrated modules may be implemented in hardware or in software functional modules. The integrated modules may also be stored in a computer readable storage medium if implemented in the form of software functional modules and sold or used as a stand-alone product.
The above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, or the like.
In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
While embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and not to be construed as limiting the invention, and that variations, modifications, alternatives and variations may be made to the above embodiments by one of ordinary skill in the art within the scope of the invention.

Claims (7)

1. A vehicle collision recognition method based on deep learning, characterized by comprising:
step 1: collecting real-time running data of a vehicle, and acquiring historical running data and historical risk-giving data of the vehicle;
step 2: preprocessing vehicle operation data;
step 3: performing low-frequency to high-frequency data interpolation processing on vehicle operation data;
step 4: ETL engineering feature conversion treatment;
step 5: model training and iteration are carried out through a deep learning method and a mathematical statistics correlation principle, and a correlation model is obtained;
step 6: according to the real-time running data of the vehicle and the combination of the related models, a vehicle collision recognition result is obtained;
the acquiring vehicle history operation data and vehicle history risk data includes:
acquiring historical operating data of the vehicle, the data comprising: time stamp, GPS speed, GPS direction, GPS altitude, device power status, device positioning status, vehicle movement status, GPS longitude, and GPS latitude;
acquiring historical risk information of the vehicle by a third party, wherein the risk information comprises: time of risk, place of risk and vehicle collision location;
matching the risk information with historical operation data of the vehicle through risk time, intercepting different time sequence data with the time length of Ts containing risk moment through a data sliding method, marking the time sequence data as risk, intercepting a certain amount of non-risk time sequence data, and marking the time sequence data as non-risk;
the correlation model includes: a running track anomaly identification model, a sudden braking anomaly identification model, a turning anomaly identification model, a parking anomaly identification model and a braking distance anomaly model;
the method for obtaining the identification result according to the real-time operation data of the vehicle and by combining the related model comprises the following steps:
dividing the real-time running data of the vehicle according to the time interval of Ts;
carrying out real-time processing on the divided TS duration data according to the steps 2, 3 and 4;
inputting the processed data into a running track anomaly identification model, a sudden braking anomaly identification model, a turning anomaly identification model, a parking anomaly identification model and a braking distance anomaly model to respectively obtain driving anomaly scores of the models;
determining a recognition result according to the driving anomaly score of each model;
the determining the recognition result according to the driving abnormality score of each model comprises the following steps:
obtaining the sum of products of driving abnormality scores and corresponding weights of the models, and taking the sum as a predictive score;
judging whether the predictive value is larger than an accident threshold value or not;
if the predictive value is greater than the accident threshold, judging that the vehicle has a collision accident; otherwise, judging that the vehicle has no collision accident.
2. The vehicle collision recognition method according to claim 1, wherein the preprocessing of the vehicle operation data includes:
carrying out time sequence confusion processing on vehicle operation data;
performing GPS positioning drift outlier processing on the data processed by the timing disorder;
and carrying out signal loss data processing on the data processed by the GPS positioning drift abnormal value, and deleting the data which has no positioning signal and has the speed of 0.
3. The vehicle collision recognition method according to claim 2, wherein the performing timing confusion processing on the vehicle operation data includes:
for the historical data used for model training, carrying out mode calculation on the minute min records of the TS time sequence data, then comparing each record with the current mode, calculating a time stamp difference value, and if the difference value is larger than the preset time, considering the record as carrying out complement operation on the previous data, and deleting the record;
for time sequence data calculated in real time on line of the model, calculating a time difference value between each data time stamp and the current system time, and if the time difference value is larger than the preset time, considering the data as operation of system complement data, and deleting the data;
and sorting the historical data or the data analyzed in real time after the processing is finished according to the time stamp.
4. The vehicle collision recognition method according to claim 2, wherein the GPS positioning drift outlier processing of the time-lapse processed data includes:
sequentially calculating the time interval and the linear distance of the earth between the t and t+1 records;
acquiring suspicious linear distance;
and positioning the drift point according to the suspicious straight line distance, the time interval and the earth straight line distance.
5. The vehicle collision recognition method according to claim 1, wherein the performing the low-frequency to high-frequency data interpolation processing on the vehicle operation data includes:
performing low-frequency to high-frequency data interpolation processing on vehicle operation data by utilizing a differential integral and Bezier curve method;
specifically, a differential integration method is utilized to conduct speed item interpolation processing on vehicle operation data;
and performing GPS longitude and latitude item interpolation on the vehicle operation data subjected to the speed item interpolation processing by using a Bezier curve method.
6. The vehicle collision recognition method according to claim 1, wherein the model training and iteration by the deep learning method and the principle related to mathematical statistics comprises:
dividing a training set and a testing set of the processed historical data;
LSTM model parameters are used for carrying out LSTM modeling on training set data, LSTM model optimization and evaluation parameters are used for evaluating the prediction results of the testing set data, and optimization iteration is carried out on the model parameters.
7. A deep learning based vehicle collision recognition system, comprising:
the data acquisition module is used for acquiring real-time running data of the vehicle and acquiring historical running data and historical risk-giving data of the vehicle;
the preprocessing module is used for preprocessing vehicle operation data;
the data interpolation processing module is used for performing low-frequency to high-frequency data interpolation processing on the vehicle operation data;
the conversion module is used for ETL engineering characteristic conversion treatment;
the modeling module is used for carrying out model training and iteration through a deep learning method and a mathematical statistics correlation principle to obtain a correlation model;
the recognition module is used for obtaining a vehicle collision recognition result according to the processed real-time vehicle operation data and the combination of the related models;
the acquiring vehicle history operation data and vehicle history risk data includes:
acquiring historical operating data of the vehicle, the data comprising: time stamp, GPS speed, GPS direction, GPS altitude, device power status, device positioning status, vehicle movement status, GPS longitude, and GPS latitude;
acquiring historical risk information of the vehicle by a third party, wherein the risk information comprises: time of risk, place of risk and vehicle collision location;
matching the risk information with historical operation data of the vehicle through risk time, intercepting different time sequence data with the time length of Ts containing risk moment through a data sliding method, marking the time sequence data as risk, intercepting a certain amount of non-risk time sequence data, and marking the time sequence data as non-risk;
the correlation model includes: a running track anomaly identification model, a sudden braking anomaly identification model, a turning anomaly identification model, a parking anomaly identification model and a braking distance anomaly model;
the method for obtaining the identification result according to the real-time operation data of the vehicle and by combining the related model comprises the following steps:
dividing the real-time running data of the vehicle according to the time interval of Ts;
carrying out real-time processing on the divided TS duration data according to the steps 2, 3 and 4;
inputting the processed data into a running track anomaly identification model, a sudden braking anomaly identification model, a turning anomaly identification model, a parking anomaly identification model and a braking distance anomaly model to respectively obtain driving anomaly scores of the models;
determining a recognition result according to the driving anomaly score of each model;
the determining the recognition result according to the driving abnormality score of each model comprises the following steps:
obtaining the sum of products of driving abnormality scores and corresponding weights of the models, and taking the sum as a predictive score;
judging whether the predictive value is larger than an accident threshold value or not;
if the predictive value is greater than the accident threshold, judging that the vehicle has a collision accident; otherwise, judging that the vehicle has no collision accident.
CN202110061524.1A 2021-01-18 2021-01-18 Vehicle collision recognition method and system based on deep learning Active CN112749210B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110061524.1A CN112749210B (en) 2021-01-18 2021-01-18 Vehicle collision recognition method and system based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110061524.1A CN112749210B (en) 2021-01-18 2021-01-18 Vehicle collision recognition method and system based on deep learning

Publications (2)

Publication Number Publication Date
CN112749210A CN112749210A (en) 2021-05-04
CN112749210B true CN112749210B (en) 2024-03-12

Family

ID=75652302

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110061524.1A Active CN112749210B (en) 2021-01-18 2021-01-18 Vehicle collision recognition method and system based on deep learning

Country Status (1)

Country Link
CN (1) CN112749210B (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114054736B (en) * 2021-10-12 2022-10-18 中国重型机械研究院股份公司 Buggy ladle parking system and method
CN114291081B (en) * 2021-12-31 2023-07-04 成都路行通信息技术有限公司 Vehicle collision detection method based on artificial intelligence algorithm
CN116189114B (en) * 2023-04-21 2023-07-14 西华大学 Method and device for identifying collision trace of vehicle
CN116343484B (en) * 2023-05-12 2023-10-03 天津所托瑞安汽车科技有限公司 Traffic accident identification method, terminal and storage medium

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104882025A (en) * 2015-05-13 2015-09-02 东华大学 Crashing detecting and warning method based on vehicle network technology
CN109344903A (en) * 2018-10-12 2019-02-15 苏交科华东(浙江)工程设计有限公司 Urban road surfaces failure real-time detection method based on vehicle-mounted perception data
CN110599353A (en) * 2018-06-13 2019-12-20 百度在线网络技术（北京）有限公司 Vehicle insurance and claims rate prediction method, device, equipment and medium
CN111079590A (en) * 2019-12-04 2020-04-28 东北大学 Peripheral vehicle behavior pre-judging method of unmanned vehicle

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10950130B2 (en) * 2018-03-19 2021-03-16 Derq Inc. Early warning and collision avoidance

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104882025A (en) * 2015-05-13 2015-09-02 东华大学 Crashing detecting and warning method based on vehicle network technology
CN110599353A (en) * 2018-06-13 2019-12-20 百度在线网络技术（北京）有限公司 Vehicle insurance and claims rate prediction method, device, equipment and medium
CN109344903A (en) * 2018-10-12 2019-02-15 苏交科华东(浙江)工程设计有限公司 Urban road surfaces failure real-time detection method based on vehicle-mounted perception data
CN111079590A (en) * 2019-12-04 2020-04-28 东北大学 Peripheral vehicle behavior pre-judging method of unmanned vehicle

Also Published As

Publication number Publication date
CN112749210A (en) 2021-05-04

Similar Documents

Publication Publication Date Title
CN112749210B (en) Vehicle collision recognition method and system based on deep learning
CN103176185B (en) Method and system for detecting road barrier
JP5895162B2 (en) Traffic accident detection device and traffic accident detection method
CN109459045B (en) Improved interactive voting matching method for low-frequency GPS track
CN109784254B (en) Vehicle violation event detection method and device and electronic equipment
CN108877213B (en) Traffic accident identification method and system
CN107293117A (en) A kind of determination methods of highway anomalous event
CN110009903B (en) Traffic accident scene restoration method
CN100466010C (en) Different species traffic information real time integrating method
CN113936465B (en) Traffic event detection method and device
CN106339692A (en) Fatigue driving state information determination method based on route offset detection and system
CN110781578A (en) Intelligent network connection algorithm testing and evaluating method based on accident scene
CN113096397A (en) Traffic jam analysis method based on millimeter wave radar and video detection
Wei et al. Trajectory-based identification of critical instantaneous decision events at mixed-flow signalized intersections
CN116135640A (en) Anti-collision early warning method and system for vehicle and vehicle
CN109720353B (en) Driving behavior detection method based on smart phone
CN114771548A (en) Data logging for advanced driver assistance system testing and verification
CN116665188B (en) Bus image system data analysis method
Scora et al. Real-time roadway emissions estimation using visual traffic measurements
CN102930722B (en) Traffic flow video detection device and detection method thereof
CN117037088A (en) Positioning method and system for thermal power plant coal ash transport vehicle based on edge calculation
Das et al. Why slammed the brakes on? auto-annotating driving behaviors from adaptive causal modeling
Higuchi et al. Monitoring live parking availability by vision-based vehicular crowdsensing
Shen et al. Data Collection and Processing Methods for the Evaluation of Vehicle Road Departure Detection Systems
Li et al. A robust terrain-based road vehicle localization algorithm

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