CN115063976B - Vehicle conflict risk assessment and prediction method based on multichannel convolutional neural network - Google Patents
Vehicle conflict risk assessment and prediction method based on multichannel convolutional neural network Download PDFInfo
- Publication number
- CN115063976B CN115063976B CN202210649067.2A CN202210649067A CN115063976B CN 115063976 B CN115063976 B CN 115063976B CN 202210649067 A CN202210649067 A CN 202210649067A CN 115063976 B CN115063976 B CN 115063976B
- Authority
- CN
- China
- Prior art keywords
- vehicle
- formula
- data
- evaluation
- neural network
- 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
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/0104—Measuring and analyzing of parameters relative to traffic conditions
- G08G1/0125—Traffic data processing
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S19/00—Satellite radio beacon positioning systems; Determining position, velocity or attitude using signals transmitted by such systems
- G01S19/01—Satellite radio beacon positioning systems transmitting time-stamped messages, e.g. GPS [Global Positioning System], GLONASS [Global Orbiting Navigation Satellite System] or GALILEO
- G01S19/13—Receivers
- G01S19/35—Constructional details or hardware or software details of the signal processing chain
- G01S19/37—Hardware or software details of the signal processing chain
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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 discloses a vehicle conflict risk assessment and prediction method based on a multichannel convolutional neural network, which comprises the following steps: collecting satellite positioning data and personal information of a vehicle; establishing a data processing and analyzing model to obtain vehicle state data and road state data; the vehicle abnormal movement behavior evaluation system is used as an index, and weight distribution is carried out on the vehicle abnormal movement behavior based on the vehicle abnormal movement behavior entropy model; the vehicle collision risk is evaluated in real time by using the weight distribution result of the vehicle abnormal motion behavior entropy model; classifying, merging and average processing are carried out on the data to obtain a vehicle collision risk 2D tensor; and forming 2D tensor graphs by various vehicle movement behaviors, superposing the 2D tensor graphs to obtain 3D tensor graphs of various vehicle movement behaviors, pooling the 2D tensors of the vehicle collision risks based on a bidirectional multichannel convolutional neural network model, rolling and pooling the 3D tensors of various vehicle movement behaviors, and carrying out error back propagation to realize the prediction of the vehicle collision risks.
Description
Technical Field
The invention relates to the technical field of road traffic safety, in particular to a vehicle conflict risk assessment and prediction method based on a multichannel convolutional neural network.
Background
Urban road traffic safety problems are mostly caused by collision and possible collision among vehicles, and the movement behaviors of the vehicles can effectively reflect the occurrence frequency and probability of the collision among the vehicles. At present, most of researches on urban traffic conflict problems are focused on the aspects of vehicle motion model establishment, motion behavior simulation and the like, and the researches on road intersection vehicle conflict are emphasized. There is little research on vehicle collision in general urban roads, and no discussion is given about real-time evaluation and short-time prediction of the risk of vehicle collision in various urban road sections. However, the number of vehicles and the collision of vehicles under different time period road conditions can generate larger floating in practice. The traditional vehicle conflict risk assessment method cannot accurately study a large number of vehicle behaviors, cannot accurately correspond the vehicle behaviors to actual road traffic conditions, and cannot respectively dynamically assess and predict the vehicle conflict risks in real time under the traffic conditions of different dates and different time periods.
With the continuous development of internet technology, smart phones with real-time GPS positioning and data interaction functions are rapidly increased, and drivers can transmit vehicle behavior data including vehicle positions, speeds and running states in real time to determine driving behaviors, so that vehicle collision risks are estimated in real time. In addition, with the continuous development of computer technology, the application field of artificial intelligence is wider, and the results are more effective by the large-batch data processing and the establishment of related algorithms, so that the prediction of the collision risk of vehicles in urban roads is effectively realized.
The method evaluates and predicts the collision risk of the vehicles in the urban road, and has very important significance for improving the traffic safety service level of the urban road and improving the traffic jam of the urban road.
Disclosure of Invention
Aiming at solving the problems of long operation time and high data requirement of the existing vehicle conflict risk assessment and prediction algorithm, the invention discloses a vehicle conflict risk assessment and prediction method based on a multichannel convolutional neural network, which improves the effectiveness of vehicle conflict risk assessment and the accuracy of a prediction result.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
according to a first aspect of an embodiment of the present invention, a vehicle collision risk assessment and prediction method based on a multichannel convolutional neural network is provided.
In one embodiment, the vehicle collision risk assessment and prediction method based on the multichannel convolutional neural network comprises the following steps:
s1: collecting satellite positioning data and personal information of a vehicle;
s2: establishing a data processing and analyzing model to obtain vehicle state data and road state data;
s3: the vehicle abnormal movement behavior evaluation system is used as an index, and weight distribution is carried out on the vehicle abnormal movement behavior based on the vehicle abnormal movement behavior entropy model;
s4: the vehicle collision risk is evaluated in real time by using the weight distribution result of the vehicle abnormal motion behavior entropy model;
s5: classifying, merging and average processing are carried out on the data in the step S2, and a vehicle conflict risk 2D tensor is obtained;
s6: forming a 2D tensor graph by various vehicle movement behaviors, and superposing the 2D tensor graph to obtain a 3D tensor graph of various vehicle movement behaviors;
s7: based on a bidirectional multichannel convolutional neural network model, pooling the vehicle collision risk 2D tensors, rolling and pooling the 3D tensors of various vehicle movement behaviors, and carrying out error back propagation to realize vehicle collision risk prediction.
Optionally, in step S1, the GPS module in the smart phone of the driver collects satellite positioning data and information of the vehicle in real time, where the information includes a driver identification number, a UNIX timestamp, a longitude coordinate and a latitude coordinate, performs preliminary screening on the collected data to remove information of obvious error data and abnormal offset of coordinates, and performs road binding processing on the data to accurately correspond to actual road coordinates.
Optionally, in step S1, the collecting the satellite positioning data and the personal information of the vehicle includes performing road binding processing and personal information processing on the satellite positioning data of the vehicle, where the personal information processing includes encrypting and dyeing the driver identification ID, and removing data processing on coordinates of a long-time parking point of the driver or a boarding point and a disembarking point of the passenger.
Optionally, in step S2, a data processing and analyzing model is established, modeling is performed on the vehicle movement behavior, and calculation and analysis are performed on the collected data to obtain vehicle movement state data including speed, acceleration, angular speed and angular acceleration and road traffic state data including average traffic speed;
wherein, establish data processing and analysis model includes:
the vehicle longitude and latitude coordinates of the nth moment and the (n+1) th moment are adopted to obtain the distance between two time nodes of the vehicle, as shown in the formula (1):
in the formula, lo ij n With La ij n Respectively representing longitude and latitude coordinates of the vehicle when the ith driver executes the j order at the nth time;
approximate displacement of vehicle at each time intervalAs shown in formula (2):
determining an instantaneous speed of a vehicleAnd instantaneous acceleration->As shown in formula (3) and formula (4):
calculating the change in the instantaneous azimuth angle of a vehicleAs shown in formula (5):
in the method, in the process of the invention,and->Representing the displacement of the n and n+1 th sections, dist 2 (n-1, n+1) represents a displacement between the n-1 time node and the n+1 time node;
determining an instantaneous angular velocity of a vehicleAnd angular acceleration->As shown in the formula (6) and the formula (7):
calculating the average speed of the vehicle in the estimated road section at the nth time nodeAs shown in formula (8):
in the method, in the process of the invention,representing the sum of the average speeds of all vehicles within the road segment being evaluated.
Optionally, the establishing of the abnormal movement behavior evaluation system of the vehicle in step S3 includes: the method comprises the steps of obtaining the relation between the collision type possibly occurring in a vehicle and the vehicle movement behavior through analysis and evaluation of various vehicle movement state data and road traffic state data, and establishing a vehicle abnormal movement behavior evaluation system;
the method for carrying out weight distribution on various abnormal behaviors of the vehicle in different types of road sections based on the vehicle abnormal motion behavior entropy model by taking the vehicle abnormal motion behavior evaluation system as an index comprises the following steps:
the vehicle speed and angular velocity indexes are adjusted by taking 85% of the maximum value of the vehicle speed and the angular velocity as evaluation criteria, as shown in the formula (9) and the formula (10):
wherein 0.85 represents δP with 85% of the maximum value of the vehicle speed and the angular velocity as the evaluation criterion v 、δP ω Respectively representing the vehicle speed and the angular speed index;
based on the constructed urban expressway vehicle conflict behavior evaluation indexes, a corresponding evaluation matrix P is constructed, and the evaluation matrix P is shown as a formula (11):
P＝(p ij ) n×m (11)
wherein p is ij J evaluation indexes of the ith sample;
normalizing the evaluation index, including processing the index with longitudinal acceleration, angular acceleration, etc. greater than zero as shown in formula (12), and processing the index with longitudinal deceleration, etc. less than zero as shown in formula (13), to obtain normalized evaluation matrix B ij ＝[b ij ]：
To determine the specific gravity of each value in the corresponding evaluation index in the evaluation system, calculating the characteristic specific gravity Z of the ith evaluation object under the jth evaluation index ij As shown in formula (14):
calculating the entropy value e of the j-th index j As shown in formula (15):
calculating the differentiation coefficient g of the j-th index j As shown in formula (16):
g j ＝1-e j (16)。
optionally, in step S4, the vehicle collision risk is evaluated in real time by using the weight distribution result of the vehicle abnormal movement behavior entropy model, the vehicle abnormal movement behavior entropy model is applied to the vehicle behavior data, and the vehicle collision risk concentration point is determined by establishing an urban road vehicle collision risk level thermodynamic diagram, so as to realize the vehicle collision risk evaluation;
calculating the weight coefficient w of the j-th evaluation index j Obtaining final evaluation results of various evaluation indexes, as shown in a formula (17):
optionally, in step S5, the data in step S2 is classified and combined with mean value processing to obtain a vehicle collision risk 2D tensor, the data is classified and combined, the vehicle movement behaviors in different road sections in each time period are determined, and the tabulated striped data are converted into grid data of various vehicle movement behaviors in different time periods;
comprising the following steps:
s51, meshing the longitude and latitude coordinates of the road, and adopting a nearby principle and an upper left principle to correspond punctiform longitude and latitude coordinate data to the inside of the meshed coordinates;
s52, calling a Pytorch API in the Python to correspondingly combine and average various vehicle motion behavior data;
and S53, establishing 1024 x 1024 2D space tensors, respectively inputting all kinds of well-processed vehicle movement behavior data into the tensors by taking 5min as a time node, and storing the tensors into a memory.
Optionally, in step S6, the 2D tensor graphs are formed by the motion behaviors of the various vehicles, and the overlapping is performed to obtain the 3D tensor graphs of the motion behaviors of the various vehicles, including:
the data after the merging and mean processing is subjected to matrix classification processing, and data at intervals of 5min are extracted and divided as shown in the formulas (18) and (19):
wherein x is mn M represents the latitude number of the vehicle in the city, n represents the longitude number of the vehicle, and x represents the average displacement data of the vehicle;
in the formula, v mn Where m represents the latitude number of the vehicle in the city, n represents the longitude number thereof, and x represents the average speed data of the vehicle thereof.
Combining the two-dimensional matrices of average displacement, velocity, acceleration, angular displacement, angular velocity, angular acceleration into a 3D tensor, as shown in equation (20):
G＝[X 1 V 2 A 3 θ 4 ω 5 α 6 ] (20)
wherein X is 1 V 2 A 3 θ 4 ω 5 α 6 The average matrix of six indexes of average displacement, speed, acceleration, angular displacement, angular velocity and angular acceleration are represented respectively.
Optionally, in step S7, based on the bidirectional multichannel convolutional neural network model, pooling the vehicle collision risk 2D tensors, rolling and pooling the 3D tensors of the motion behaviors of various vehicles, and performing back propagation to implement vehicle collision risk prediction, including:
for the convolution layer, the data is convolved as shown in formula (21):
in the method, in the process of the invention,the j-th pattern on the first layer, ">An i-th element representing a j-th weight tensor at layer l, b being a bias factor;
for the pooling layer, the data is pooled as shown in formula (22):
pooling the data processed in the step S5, rolling up and pooling the 3D tensors of various vehicle motions, and performing error back propagation on the results obtained after pooling, wherein the error back propagation is shown as a formula (23) and a formula (24);
performing partial derivative solving on the generated error value, calculating a cost function, and reversely transmitting the cost function to the initial one-layer neural network model to update parameters, wherein the parameters are as shown in the formula (25) and the formula (26):
in the formula, G and b are finally processed based on a chain rule of a derivative function so as to realize the layer-by-layer updating of parameters.
According to a second aspect of an embodiment of the present invention, an application of the above method is presented.
In one embodiment, the method is applied to real-time assessment and prediction of vehicle collision risk for each road segment of a city.
The invention has the advantages that,
the invention discloses a vehicle conflict risk assessment and prediction method based on smart phone GPS data and a bidirectional multichannel convolutional neural network. In order to effectively evaluate the collision risk of the vehicle, a data processing and analyzing model is established by collecting GPS data of a smart phone of a driver of the motor vehicle in real time, and vehicle state data including speed, acceleration, angular speed, angular acceleration and the like and road state data including average traffic speed are obtained by screening, processing and analyzing the collected data; analyzing and studying the data to establish a vehicle abnormal movement behavior evaluation index system; establishing an entropy model of abnormal motion behaviors of the vehicle based on an entropy weight method, and carrying out weight distribution on different types of abnormal behaviors of the vehicle; and applying the vehicle abnormal movement behavior entropy evaluation system to the vehicle behavior data, thereby obtaining and determining more aggregation points of the vehicle abnormal movement behaviors so as to evaluate the risk of vehicle collision. In order to effectively predict the collision risk of vehicles, a bidirectional multichannel convolutional neural network model is established, and the screened and analyzed data are classified and combined to obtain the average value of various vehicle state data in different time periods respectively. Then, discretizing the whole urban road, dividing grids with different sizes according to the actual intersection density, and respectively fusing the grid data into a multi-channel 3D map according to the vehicle state type so as to carry out convolution treatment; and simultaneously, corresponding pooling processing is carried out on the vehicle conflict data so as to support training of a bidirectional multichannel convolutional neural network model, and finally, prediction of the vehicle conflict risk is realized.
Drawings
FIG. 1 is a schematic diagram of the instantaneous displacement of a vehicle according to an embodiment of the present invention;
FIG. 2 is a schematic illustration of instantaneous angular displacement of a vehicle according to an embodiment of the present invention;
FIG. 3 is a schematic diagram showing average speed of a road segment according to an embodiment of the present invention;
FIG. 4 is a diagram of data meshing according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of a bi-directional multi-channel convolutional neural network according to one embodiment of the present invention;
FIG. 6 is a schematic diagram of a vehicle collision risk prediction model according to an embodiment of the invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
A vehicle collision risk assessment and prediction method based on a multichannel convolutional neural network comprises the following steps:
s1: collecting satellite positioning data and personal information of a vehicle;
s2: establishing a data processing and analyzing model to obtain vehicle state data and road state data;
s3: the vehicle abnormal movement behavior evaluation system is used as an index, and weight distribution is carried out on the vehicle abnormal movement behavior based on the vehicle abnormal movement behavior entropy model;
s4: the vehicle collision risk is evaluated in real time by using the weight distribution result of the vehicle abnormal motion behavior entropy model;
s5: classifying, merging and average processing are carried out on the data in the step S2, and a vehicle conflict risk 2D tensor is obtained;
s6: forming a 2D tensor graph by various vehicle movement behaviors, and superposing the 2D tensor graph to obtain a 3D tensor graph of various vehicle movement behaviors;
s7: based on a bidirectional multichannel convolutional neural network model, pooling the vehicle collision risk 2D tensors, rolling and pooling the 3D tensors of various vehicle movement behaviors, and carrying out error back propagation to realize vehicle collision risk prediction.
Optionally, in step S1, the GPS module in the smart phone of the driver collects satellite positioning data and information of the vehicle in real time, where the information includes a driver identification number, a UNIX timestamp, a longitude coordinate and a latitude coordinate, performs preliminary screening on the collected data to remove information of obvious error data and abnormal offset of coordinates, and performs road binding processing on the data to accurately correspond to actual road coordinates.
Optionally, in step S1, the collecting the satellite positioning data and the personal information of the vehicle includes performing road binding processing and personal information processing on the satellite positioning data of the vehicle, where the personal information processing includes encrypting and dyeing the driver identification ID, and removing data processing on coordinates of a long-time parking point of the driver or a boarding point and a disembarking point of the passenger.
Optionally, in step S2, a data processing and analyzing model is established, modeling is performed on the vehicle movement behavior, and calculation and analysis are performed on the collected data to obtain vehicle movement state data including speed, acceleration, angular speed and angular acceleration and road traffic state data including average traffic speed;
wherein, establish data processing and analysis model includes:
the vehicle longitude and latitude coordinates of the nth moment and the (n+1) th moment are adopted to obtain the displacement of the vehicle between two time nodes, as shown in the formula (1):
in the formula, lo ij n With La ij n Respectively representing longitude and latitude coordinates of the vehicle when the ith driver executes the j order at the nth time;
as shown in FIG. 1, because the time node difference of the acquired data is 3s, the vehicle displacement can be considered to be in a shorter period of time than the complete movement duration of the vehicleMay be approximately equal to the vehicle forward path as shown in equation (2):
determining an instantaneous speed of a vehicleAnd instantaneous acceleration->As shown in the formula (3) and the formula (4):
lateral movement of the vehicle is also one of the main causes of vehicle collision, and further determination of angular velocity and angular acceleration of the vehicle is required for efficient evaluation of the relationship between vehicle behavior and vehicle collision.
To obtain the change of instantaneous azimuth angle of vehicleBy the formula (1), the displacement of the vehicle between two time nodes can be obtained by adopting the longitude and latitude coordinates of the vehicle at the nth time and the (n+1) th time, as shown in fig. 2.
The displacement of the n and n+1 th sections calculated by the formula (1)And->Carrying into formula (5) and calculating the change of the instantaneous azimuth angle of the vehicle +.>
In the method, in the process of the invention,and->Representing the displacement of the n and n+1 th sections, dist 2 (n-1, n+1) represents a displacement between the n-1 time node and the n+1 time node;
determining an instantaneous angular velocity of a vehicleAnd angular acceleration->As shown in the formula (6) and the formula (7):
in addition to the behavior of the vehicle itself, road traffic conditions, including the average speed of the vehicle, can also have an impact on the frequency and probability of vehicle collisions. Calculating the average speed of the vehicle at the nth time node by the formula (8)As shown in fig. 3:
in the method, in the process of the invention,representing the sum of the average speeds of all vehicles within the segment.
Optionally, the establishing of the abnormal movement behavior evaluation system of the vehicle in step S3 includes: the method comprises the steps of obtaining the relation between the collision type possibly occurring in a vehicle and the vehicle movement behavior through analysis and evaluation of various vehicle movement state data and road traffic state data, and establishing a vehicle abnormal movement behavior evaluation system;
the method for carrying out weight distribution on various abnormal behaviors of the vehicle in different types of road sections based on the vehicle abnormal motion behavior entropy model by taking the vehicle abnormal motion behavior evaluation system as an index comprises the following steps:
the vehicle speed and angular velocity indexes are adjusted by taking 85% of the maximum value of the vehicle speed and the angular velocity as evaluation criteria, as shown in the formula (9) and the formula (10):
wherein 0.85 represents δP with 85% of the maximum value of the vehicle speed and the angular velocity as the evaluation criterion v 、δP ω Respectively representing the vehicle speed and the angular speed index;
based on the constructed urban expressway vehicle conflict behavior evaluation indexes, a corresponding evaluation matrix P is constructed, and the evaluation matrix P is shown as a formula (11):
P＝(p ij ) n×m (11)
wherein p is ij J term for sample number iEvaluating the index;
to further eliminate the influence of different index dimensions, the evaluation indexes are standardized, and the vehicle collision evaluation indexes including the longitudinal acceleration and the angular acceleration of the vehicle are considered to be optimal in the stationary state, wherein the standardized evaluation matrix B is obtained by processing the indexes with the longitudinal acceleration, the angular acceleration and the like larger than zero as shown in a formula (12) and the indexes with the longitudinal deceleration and the like smaller than zero as shown in a formula (13) ij ＝[b ij ]：
To determine the specific gravity of each value in the corresponding evaluation index in the evaluation system, calculating the characteristic specific gravity Z of the ith evaluation object under the jth evaluation index ij As shown in formula (14):
calculating the entropy value e of the j-th index j As shown in formula (15):
calculating the differentiation coefficient g of the j-th index j As shown in formula (16):
g j ＝1-e j (16)。
optionally, in step S4, the vehicle collision risk is evaluated in real time by using the weight distribution result of the vehicle abnormal movement behavior entropy model, the vehicle abnormal movement behavior entropy model is applied to the vehicle behavior data, and the vehicle collision risk concentration point is determined by establishing an urban road vehicle collision risk level thermodynamic diagram, so as to realize the vehicle collision risk evaluation;
calculating the weight coefficient w of the j-th evaluation index j Obtaining final evaluation results of various evaluation indexes, as shown in a formula (17):
optionally, in step S5, the data in step S2 is classified and combined with mean value processing to obtain a vehicle collision risk 2D tensor, the data is classified and combined, the vehicle movement behaviors in different road sections in each time period are determined, and the tabulated striped data are converted into grid data of various vehicle movement behaviors in different time periods; the original data are all tabular stripe vehicle data, so that the tabular stripe data are required to be converted for effective graph prediction, so that various vehicle motion behaviors in different roads under each period are respectively obtained, and the convolutional neural network learning of the vehicle collision risk in an image mode is realized. Thus, the following method is adopted, based on the analysis processing of the NumPyAPI and Pandas API data under Python.
Comprising the following steps:
s51, meshing longitude and latitude coordinates, and adopting a nearby principle and an upper left principle to correspond punctiform longitude and latitude coordinate data to the inside of the divided meshing coordinates;
s52, calling a Pytorch API in the Python to correspondingly combine and average various vehicle motion behavior data;
and S53, establishing 1024 x 1024 2D space tensors, respectively inputting all kinds of well-processed vehicle movement behavior data into the tensors by taking 5min as a time node, and storing the tensors into a memory.
Optionally, in step S6, the 2D tensor graphs are formed by the motion behaviors of the various vehicles, and the overlapping is performed to obtain the 3D tensor graphs of the motion behaviors of the various vehicles, including:
the data are classified and combined, the data after one-time processing are striped vehicle state data, the vehicle motion state under specific time cannot be represented, the mutual association relation between the vehicle and the road cannot be represented, the data after combination and mean processing are subjected to matrix classification processing, the data are extracted and divided every 5min, and positioning is carried out according to the vehicle positioning coordinates. If there are multiple pieces of car data at a certain location, i.e. data combination and mean processing are performed on the pieces of car data, if there is only single piece of car data at the location, the piece of data is considered as the car movement state data of the area, if there is no car movement state data in the area within a time interval, the average speed of the car in the area is considered as 0, as shown in the formula (18) and the formula (19):
wherein x is mn M represents the latitude number of the vehicle in the city, n represents the longitude number of the vehicle, and x represents the average displacement data of the vehicle;
in the formula, v mn Where m represents the latitude number of the vehicle in the city, n represents the longitude number thereof, and x represents the average speed data of the vehicle thereof.
According to equation (20), two-dimensional matrices of average displacement, velocity, acceleration, angular displacement, angular velocity, angular acceleration are combined into a 3D tensor to support the prediction of subsequent bi-directional multi-channel convolutional neural networks, as shown in fig. 4:
G＝[X 1 V 2 A 3 θ 4 ω 5 α 6 ] (20)
wherein X is 1 V 2 A 3 θ 4 ω 5 α 6 The average matrix of six indexes of average displacement, speed, acceleration, angular displacement, angular velocity and angular acceleration are represented respectively.
In order to effectively evaluate the collision risk of the vehicle, a bidirectional multichannel convolutional neural network model is established, and a special neural network model training method is provided. The typical convolutional neural network model comprises five processes of a convolutional layer, a pooling layer, a full connection layer, back propagation and parameter updating, so that the learning of parameters in the model is realized. The bidirectional multichannel convolutional neural network model provided by the invention respectively learns various vehicle motion behavior 3D tensors including road position information and vehicle conflict risk 2D tensors including road position information which are established in the formula (20) and are realized in a vehicle conflict risk assessment module, and carries out error back propagation on the final convolution or pooling result so as to realize the bidirectional multichannel convolutional neural network model. The model structure of the established bidirectional multichannel convolutional neural network is shown in fig. 5.
Optionally, in step S7, based on the bidirectional multichannel convolutional neural network model, the vehicle collision risk 2D tensors are pooled, rolling and pooling are performed on the 3D tensors of the motion behaviors of various vehicles, and back propagation is performed, so as to implement vehicle collision risk prediction, as shown in fig. 6, including:
for the convolution layer, the data is convolved as shown in formula (21):
in the method, in the process of the invention,the j-th pattern on the first layer, ">An i-th element representing a j-th weight tensor at layer l, b being a bias factor;
for the pooling layer, the data is pooled as shown in formula (22):
the 3D tensors of various movement behaviors of the vehicle, including road position status information, are processed according to the mode of the table 1, and the 2D tensors with the size of 1,8,8 are finally obtained through convolution and pooling.
TABLE 1 method for processing various sports behaviors of vehicle
The established 2D tensor of risk of collision of the vehicle, including road location information, is processed in the manner of table 2, and only by pooling the function, a 2D tensor of size is finally obtained (1,8,8).
TABLE 2 vehicle conflict risk processing method
Pooling the vehicle collision risk 2D tensors, rolling and pooling the 3D tensors of various vehicle motions, and performing error back propagation on the results after the pooling treatment, wherein the error back propagation is shown as a formula (23) and a formula (24);
performing partial derivative solving on the generated error value, calculating a cost function, and reversely transmitting the cost function to the initial one-layer neural network model to update parameters, wherein the parameters are as shown in the formula (25) and the formula (26):
in the formula, G and b are finally processed based on a chain rule of a derivative function so as to realize the layer-by-layer updating of parameters.
The operation software adopted by the neural network model is Python3.8 and Pytorch1.10, and CUDA is used for acceleration; the hardware device used was a CPU Intel core i7-8750H CPU@2.20GHz, GPU NVIDIA GeForce RTX 2070with Max-Q Design.
Evaluation index: the invention adopts cross entropy, mean square error and root mean square error as evaluation indexes of a model, wherein, the formula (27) is a cross entropy evaluation function, the formula (28) is a mean square error evaluation function and the formula (29) is a root mean square error evaluation function:
wherein y is ic Representing a sign function, p ic To observe sample probability;
in the formulas (28) and (29),for average value,/->Is the predicted value of the sample.
Finally, based on the PyTorrchAPI under the Python, the model is built and trained, and through training and evaluation of the model of the bidirectional multichannel convolutional neural network, various errors can be ensured to be within a reasonable range, and the model is effective.
It should be understood that the above description is not intended to limit the invention to the particular embodiments disclosed, but to limit the invention to the particular embodiments disclosed, and that the invention is not limited to the particular embodiments disclosed, but is intended to cover modifications, adaptations, additions and alternatives falling within the spirit and scope of the invention.
Claims (7)
1. The vehicle collision risk assessment and prediction method based on the multichannel convolutional neural network is characterized by comprising the following steps of:
s1: collecting satellite positioning data and personal information of a vehicle;
s2: establishing a data processing and analyzing model to obtain vehicle state data and road state data;
s3: the vehicle abnormal movement behavior evaluation system is used as an index, and weight distribution is carried out on the vehicle abnormal movement behavior based on the vehicle abnormal movement behavior entropy model;
s4: the vehicle collision risk is evaluated in real time by using the weight distribution result of the vehicle abnormal motion behavior entropy model;
s5: classifying, merging and average processing are carried out on the data in the step S2, and a vehicle conflict risk 2D tensor is obtained;
s6: forming a 2D tensor graph by various vehicle movement behaviors, and superposing the 2D tensor graph to obtain a 3D tensor graph of various vehicle movement behaviors;
s7: based on a bidirectional multichannel convolutional neural network model, pooling the vehicle collision risk 2D tensors, rolling and pooling the 3D tensors of various vehicle movement behaviors, and carrying out error back propagation to realize vehicle collision risk prediction;
in step S2, a data processing and analyzing model is established, including:
the vehicle longitude and latitude coordinates of the nth moment and the (n+1) th moment are adopted to obtain the distance between two time nodes of the vehicle, as shown in the formula (1):
in the formula, lo ij n With La ij n Respectively representing longitude and latitude coordinates of the vehicle when the ith driver executes the j order at the nth time;
approximate displacement of vehicle at each time intervalAs shown in formula (2):
determining an instantaneous speed of a vehicleAnd instantaneous acceleration->As shown in formula (3) and formula (4):
calculating the change in the instantaneous azimuth angle of a vehicleAs shown in formula (5):
in the method, in the process of the invention,and->Representing the displacement of the n and n+1 th sections, dist 2 (n-1, n+1) represents a displacement between the n-1 time node and the n+1 time node;
determining an instantaneous angular velocity of a vehicleAnd angular acceleration->As shown in formula (6) and formula (7):
calculating the average speed of the vehicle in the estimated road section at the nth time nodeAs shown in formula (8):
in the method, in the process of the invention,representing the sum of the average speeds of all vehicles within the road segment being evaluated;
in step S3, the abnormal vehicle behavior evaluation system is used as an index, and the weight distribution is performed on the abnormal vehicle behavior based on the entropy model of the abnormal vehicle behavior, which includes:
the vehicle speed and angular velocity indexes are adjusted by taking 85% of the maximum value of the vehicle speed and the angular velocity as evaluation criteria, as shown in the formula (9) and the formula (10):
wherein 0.85 represents δP with 85% of the maximum value of the vehicle speed and the angular velocity as the evaluation criterion v 、δP ω Respectively representing the vehicle speed and the angular speed index;
based on the constructed urban expressway vehicle conflict behavior evaluation indexes, a corresponding evaluation matrix P is constructed, and the evaluation matrix P is shown as a formula (11):
P＝(p ij ) n×m (11)
wherein p is ij J evaluation indexes of the ith sample;
normalizing the evaluation index, including processing the index with longitudinal acceleration, angular acceleration, etc. greater than zero as shown in formula (12), and processing the index with longitudinal deceleration, etc. less than zero as shown in formula (13), to obtain normalized evaluation matrix B ij ＝[b ij ]：
To determine the specific gravity of each value in the corresponding evaluation index in the evaluation system, the j-th evaluation index is calculatedCharacteristic specific gravity Z of the i-th evaluation object under the heading ij As shown in formula (14):
calculating the entropy value e of the j-th index j As shown in formula (15):
calculating the differentiation coefficient g of the j-th index j As shown in formula (16):
g j ＝1-e j (16)。
2. the method for estimating and predicting collision risk of vehicle based on multi-channel convolutional neural network as claimed in claim 1, wherein the collecting the satellite positioning data and the personal information in step S1 comprises road binding processing and personal information processing of the satellite positioning data of vehicle, wherein the personal information processing comprises encrypting and dyeing the driver identification ID, and removing the coordinates of the vehicle stop for a long time or the vehicle get-on and get-off of the passenger.
3. The method for estimating and predicting collision risk of vehicle based on multi-channel convolutional neural network as set forth in claim 1, wherein in step S4, the collision risk of vehicle is estimated in real time by using the weight distribution result of the entropy model of abnormal motion behavior of vehicle, and the weight coefficient w of the j-th evaluation index is calculated j Obtaining final evaluation results of various evaluation indexes, as shown in a formula (17):
4. the method for estimating and predicting collision risk of a vehicle based on a multi-channel convolutional neural network as set forth in claim 1, wherein the classifying, merging and averaging the data of step S2 in step S5 to obtain a vehicle collision risk 2D tensor comprises:
s51, meshing the longitude and latitude coordinates of the road, and adopting a nearby principle and an upper left principle to correspond punctiform longitude and latitude coordinate data to the inside of the meshed coordinates;
s52, calling a Pytorch API in the Python to correspondingly combine and average various vehicle motion behavior data;
and S53, establishing 1024 x 1024 2D space tensors, respectively inputting all kinds of well-processed vehicle movement behavior data into the tensors by taking 5min as a time node, and storing the tensors into a memory.
5. The method for evaluating and predicting risk of vehicle collision based on multichannel convolutional neural network as set forth in claim 1, wherein in step S6, each kind of vehicle motion behavior is formed into a 2D tensor graph, and the 2D tensor graphs are superimposed to obtain each kind of vehicle motion behavior 3D tensor graph, including:
the data after the merging and mean processing is subjected to matrix classification processing, and data at intervals of 5min are extracted and divided as shown in the formulas (18) and (19):
wherein x is mn M represents the latitude number of the vehicle in the city, n represents the longitude number of the vehicle, and x represents the average displacement data of the vehicle;
in the formula, v mn Wherein m represents the latitude number of the vehicle in the city, n represents the longitude number thereof, and v represents the longitude number thereofVehicle average speed data;
combining the two-dimensional matrices of average displacement, velocity, acceleration, angular displacement, angular velocity, angular acceleration into a 3D tensor, as shown in equation (20):
G＝[X 1 V 2 A 3 θ 4 ω 5 α 6 ] (20)
wherein X is 1 V 2 A 3 θ 4 ω 5 α 6 The average matrix of six indexes of average displacement, speed, acceleration, angular displacement, angular velocity and angular acceleration are represented respectively.
6. The method for estimating and predicting collision risk of a vehicle based on a multi-channel convolutional neural network according to claim 1, wherein in step S7, based on a two-way multi-channel convolutional neural network model, pooling the 2D tensors of collision risk of the vehicle, rolling and pooling the 3D tensors of motion behaviors of various vehicles, and back-propagating the 3D tensors to realize collision risk prediction of the vehicle, comprising:
for the convolution layer, the data is convolved as shown in formula (21):
in the method, in the process of the invention,the j-th pattern on the first layer, ">An i-th element representing a j-th weight tensor at layer l, b being a bias factor;
for the pooling layer, the data is pooled as shown in formula (22):
pooling the data processed in the step S5, rolling up and pooling the 3D tensors of various vehicle motions, and performing error back propagation on the results obtained after pooling, wherein the error back propagation is shown as a formula (23) and a formula (24);
performing partial derivative solving on the generated error value, calculating a cost function, and reversely transmitting the cost function to the initial one-layer neural network model to update parameters, wherein the parameters are as shown in the formula (25) and the formula (26):
in the formula, G and b are finally processed based on a chain rule of a derivative function so as to realize the layer-by-layer updating of parameters.
7. A method for estimating and predicting risk of vehicle collision based on a multi-channel convolutional neural network according to any one of claims 1 to 6, wherein the risk of vehicle collision applied to each road section of a city is estimated and predicted in real time.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210649067.2A CN115063976B (en) | 2022-06-09 | 2022-06-09 | Vehicle conflict risk assessment and prediction method based on multichannel convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210649067.2A CN115063976B (en) | 2022-06-09 | 2022-06-09 | Vehicle conflict risk assessment and prediction method based on multichannel convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115063976A CN115063976A (en) | 2022-09-16 |
| CN115063976B true CN115063976B (en) | 2023-08-29 |
Family
ID=83200879
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210649067.2A Active CN115063976B (en) | 2022-06-09 | 2022-06-09 | Vehicle conflict risk assessment and prediction method based on multichannel convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115063976B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115620199B (en) * | 2022-10-24 | 2023-06-13 | 四川警察学院 | Traffic safety risk diagnosis method and device |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| ITAN20110017A1 (en) * | 2011-02-09 | 2012-08-10 | Infomobility It S P A | SAFETY DEVICE FOR VEHICLE. |
| CN111311904A (en) * | 2020-01-17 | 2020-06-19 | 东南大学 | Traffic state estimation method based on floating car data weighted tensor reconstruction |
| CN112201038A (en) * | 2020-09-28 | 2021-01-08 | 同济大学 | Road network risk assessment method based on risk of bad driving behavior of single vehicle |
| CN112258875A (en) * | 2020-10-26 | 2021-01-22 | 青岛理工大学 | Intersection intelligent control method based on vehicle-road cooperation and automatic driving |
| CN113689720A (en) * | 2021-07-14 | 2021-11-23 | 东南大学 | Automatic intersection traffic decision method based on convolutional neural network |
| JP6980958B1 (en) * | 2021-06-23 | 2021-12-15 | 中国科学院西北生態環境資源研究院 | Rural area classification garbage identification method based on deep learning |
-
2022
- 2022-06-09 CN CN202210649067.2A patent/CN115063976B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| ITAN20110017A1 (en) * | 2011-02-09 | 2012-08-10 | Infomobility It S P A | SAFETY DEVICE FOR VEHICLE. |
| EP2487659A1 (en) * | 2011-02-09 | 2012-08-15 | Infomobility.it S.p.A. | A system for characterizing the driving style of vehicle drivers |
| CN111311904A (en) * | 2020-01-17 | 2020-06-19 | 东南大学 | Traffic state estimation method based on floating car data weighted tensor reconstruction |
| CN112201038A (en) * | 2020-09-28 | 2021-01-08 | 同济大学 | Road network risk assessment method based on risk of bad driving behavior of single vehicle |
| CN112258875A (en) * | 2020-10-26 | 2021-01-22 | 青岛理工大学 | Intersection intelligent control method based on vehicle-road cooperation and automatic driving |
| JP6980958B1 (en) * | 2021-06-23 | 2021-12-15 | 中国科学院西北生態環境資源研究院 | Rural area classification garbage identification method based on deep learning |
| CN113689720A (en) * | 2021-07-14 | 2021-11-23 | 东南大学 | Automatic intersection traffic decision method based on convolutional neural network |
Non-Patent Citations (1)
| Title |
|---|
| Risk Measurement, Risk Entropy, and Autonomous Driving Risk Modeling;Jiamin Yu;《Cornell University》;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115063976A (en) | 2022-09-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | A review of surrogate safety measures and their applications in connected and automated vehicles safety modeling | |
| Xing et al. | Examining traffic conflicts of up stream toll plaza area using vehicles’ trajectory data | |
| Wang et al. | How much data are enough? A statistical approach with case study on longitudinal driving behavior | |
| CN112700072B (en) | Traffic condition prediction method, electronic device, and storage medium | |
| CN112487617B (en) | Collision model-based risk prevention method, device, equipment and storage medium | |
| CN109871599B (en) | Unmanned vehicle operation safety index evaluation method | |
| CN109272745B (en) | Vehicle track prediction method based on deep neural network | |
| Hu et al. | Efficient mapping of crash risk at intersections with connected vehicle data and deep learning models | |
| CN109118787B (en) | Vehicle speed prediction method based on deep neural network | |
| Du et al. | Predicting freeway work zone delays and costs with a hybrid machine-learning model | |
| CN108091132A (en) | A kind of traffic flow forecasting method and device | |
| CN115063976B (en) | Vehicle conflict risk assessment and prediction method based on multichannel convolutional neural network | |
| CN114446049A (en) | Traffic flow prediction method, system, terminal and medium based on social value orientation | |
| Beauchamp et al. | Study of automated shuttle interactions in city traffic using surrogate measures of safety | |
| Swain et al. | Evolution of machine learning algorithms for enhancement of self-driving vehicles security | |
| Ghoul et al. | Real-time safest route identification: Examining the trade-off between safest and fastest routes | |
| Eggert et al. | The foresighted driver: Future ADAS based on generalized predictive risk estimation | |
| Wu et al. | Lane-GNN: Integrating GNN for Predicting Drivers' Lane Change Intention | |
| Aghabayk et al. | New car-following model considering impacts of multiple lead vehicle types | |
| Elleuch et al. | Towards an efficient traffic congestion prediction method based on neural networks and big GPS data | |
| CN106157657A (en) | A kind of moving state identification system and method for mobile subscriber | |
| Nilsson | Simulation-based analysis of partially automated vehicular networks | |
| Li et al. | A probabilistic framework for estimating the risk of pedestrian-vehicle conflicts at intersections | |
| Lee et al. | Evaluating the effects of highway traffic accidents in the development of a vehicle accident queue length estimation model | |
| Anagnostopoulos et al. | Application of Artificial Neural Network for Modelling and Predicting Roundabout Capacity |
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 |