CN115035491A - Driving behavior road condition early warning method based on federal learning - Google Patents
Driving behavior road condition early warning method based on federal learning Download PDFInfo
- Publication number
- CN115035491A CN115035491A CN202210587310.2A CN202210587310A CN115035491A CN 115035491 A CN115035491 A CN 115035491A CN 202210587310 A CN202210587310 A CN 202210587310A CN 115035491 A CN115035491 A CN 115035491A
- Authority
- CN
- China
- Prior art keywords
- driving
- driver
- model
- traffic
- vehicle
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/588—Recognition of the road, e.g. of lane markings; Recognition of the vehicle driving pattern in relation to the road
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- 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)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Multimedia (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Traffic Control Systems (AREA)
Abstract
The application provides a driving behavior road condition early warning method based on federal learning, which comprises the following steps: automatically identifying the identity of a driver and collecting driving behavior information of the driver; based on a differential privacy algorithm, combining federal learning to carry out driver information acquisition and transmission and privacy protection in use; performing driver driving habit and character analysis based on a deep learning method; analyzing road risk factors in the driving process based on an image processing algorithm; calculating the probability of various traffic accidents based on the bad driving habit data, the character portrait and the road condition analysis data of surrounding drivers; and planning a driving operation scheme according to the current potential traffic accident probability, wherein the planning of the driving operation scheme specifically comprises the steps of establishing a driving operation planning model and planning the driving operation scheme.
Description
Technical Field
The invention relates to the technical field of information, in particular to a driving behavior road condition early warning method based on federal learning.
Background
When a car is in the process of driving, various traffic accidents are often encountered, and the reasons for the accidents are usually careless carelessness and poor driving habits of drivers. Therefore, even if a driver in charge of driving the automobile with good driving habits and high driving technology is seriously responsible, the automobile is difficult to avoid traffic accidents such as collision, scratch and the like with other unqualified drivers on the road. Therefore, if the character characteristics and the driving habits of surrounding drivers can be obtained during driving, reference can be provided for the current driving operation, and surrounding drivers which may generate traffic friction and disputes with the drivers can be avoided.
Disclosure of Invention
The invention provides a driving behavior road condition early warning method based on federal learning, which mainly comprises the following steps:
automatically identifying the identity of a driver and collecting driving behavior information of the driver; based on a differential privacy algorithm, combining with federal learning, carrying out driver information acquisition and transmission and privacy protection in use; performing driver driving habit and character analysis based on a deep learning method; analyzing road risk factors in the driving process based on an image processing algorithm; calculating the probability of various traffic accidents based on the bad driving habit data, the character portrait and the road condition analysis data of surrounding drivers; planning a driving operation scheme according to the current potential traffic accident probability;
further optionally, the automatically recognizing the driver identity and collecting the driver driving behavior information comprises:
the method comprises the steps that an in-vehicle camera is used for collecting a face image of a driver when the driver enters a cab, and the identity of the driver is determined through a face recognition technology; after the identity is determined, acquiring driving behavior information aiming at the current driver, and acquiring a traffic violation penalty record of the driver by calling an inquiry interface of a traffic administration; then, acquiring current various in-vehicle activity records of a driver through an in-vehicle camera and an automobile data recorder; and finally, recording the collected traffic violation penalty records and the in-vehicle activities under the account number of the current driver, and if the current driver does not have the account number, indicating that the current driver uses the system for the first time, firstly creating an account number and then recording information.
Further optionally, the privacy protection in driver information collection transmission and use based on differential privacy algorithm in combination with federal learning includes:
privacy protection is carried out on driver driving behavior information and traffic violation records collected during driving through a difference privacy algorithm in combination with federal learning; training each neural network model based on a federal learning method; firstly, a local training number neural network model of a client side obtains local model parameters; then adding random noise to the local model parameters obtained after training based on a differential privacy algorithm; the server aggregates the local model parameters received from the client to obtain new global model parameters; finally, the server broadcasts the new global model parameters to each client, and the client receives the new global model parameters and carries out local calculation again; the method comprises the following steps: training a neural network model based on federal learning; adding noise into the client model parameters based on a differential privacy algorithm;
the training of the neural network model based on the federal learning specifically comprises the following steps:
training each neural network model based on a federal learning method, wherein the neural network model comprises the following steps: the system comprises a neural network model for analyzing the driving habits and characters of a driver, a road condition analysis model, a traffic accident analysis model and a driving operation planning model. Firstly, broadcasting each initial neural network model to a local client of each vehicle by a server; then, the local client trains each initial neural network model based on locally acquired personal information and driving data of a driver, and local model parameters obtained after training are fed back to the server; then the server uses a FedAVG algorithm to aggregate local model parameters received from the local client to obtain new global model parameters; finally, the server broadcasts the new global model parameters to each local client; and the local client receives the new global model parameters, updates each initial neural network model, and performs local calculation and the next round of model iteration again.
The method for adding noise to the client model parameters based on the differential privacy algorithm specifically comprises the following steps:
after a neural network model is trained locally at a client side based on federal learning to obtain model parameters, random noise is added to the trained local model parameters by using a differential privacy algorithm. Adding random dynamic disturbance noise to data through a Laplace mechanism aiming at numerical data in local model parameters; aiming at non-numerical and discrete data, an exponential mechanism is adopted to determine the probability of an output result, the output probability is determined according to the proportion of the discrete data, and the higher the proportion is, the higher the corresponding discrete data output probability is.
Further optionally, the performing of the driver driving habit and character analysis based on the deep learning method includes:
establishing an initial neural network model for driver driving habit and character analysis based on a deep learning method, wherein the initial neural network model comprises the following steps: a bad driving habit recognition model and a driver character analysis model; then broadcasting the two models to a local client of the user vehicle, and calculating the character characteristics and driving habits of each driver related to driving based on local data; then, adjusting model parameters based on the calculation result, and feeding back the updated model to the server; finally, the server side aggregates the local models received from the local client side to obtain new global model parameters, and broadcasts the new global model parameters to the local client side again; the method comprises the following steps: training a bad driving habit recognition model based on the traffic violation record; establishing a driver character analysis model based on a deep learning method; analyzing bad driving habits and character portraits of drivers around the vehicle in real time;
the recognition model for training bad driving habits based on the traffic violation records specifically comprises the following steps:
and training an adverse driving habit recognition model based on the historical traffic violation record and the deep learning method of the current driver. Firstly, collecting a large number of traffic violation records of various drivers as a training set and a testing set; then, carrying out data cleaning, data conversion and feature extraction on the training set, wherein the extracted features comprise: violation factors, penalty types, deductions on points or not, deduction values and violation times; then, carrying out label classification on the samples, and marking the samples with various bad driving habit labels, wherein the bad driving habit labels comprise: the method comprises the following steps of reverse driving, red light running, illegal lane changing, overspeed driving, illegal parking, line pressing and crossing, safety belt fastening, drunk driving, calling in driving, smoking and intentionally shielding a number plate; then inputting the training set into a bad driving habit recognition model to train a neural network, and establishing a classification decision rule; and finally, inputting a test set to test the recognition accuracy of the bad driving habit recognition model, and adjusting parameters of the bad driving habit recognition model according to a test result to improve the accuracy of the model recognition.
The driver character analysis model is established based on the deep learning method, and specifically comprises the following steps:
and establishing a driver character analysis model, combining the activity record in the vehicle and the personal information of the driver, automatically carrying out character portrayal on the driver, and analyzing the driving character characteristics of the driver. Firstly, establishing a driver character analysis model based on a deep learning method; extracting characteristic values related to the characters of various drivers from personal related information of the drivers, wherein the characteristic values include: sex, age, occupation, education degree, driving age and bad driving habit, speech data emotion keyword extraction in combination vehicle event recorder data and the car activity record, input driver character analysis neural network as the eigenvalue and carry out model training, then output driver character portrait, character portrait includes a plurality of dimensions: driving experience, driving preferences, driving caution, driving concentration, and driving tolerance.
The bad driving habits and the personality portrait of the driver around the real-time analysis vehicle specifically include:
and broadcasting the bad driving habit recognition model and the driver character analysis model to a local client of each user vehicle, and calculating the character portrait and the bad driving habit of each driver based on the current traffic violation record, the in-vehicle activity record and the personal information of each driver, which are locally collected in real time. And then the local client feeds back the character portrait and the bad driving habits of each driver to the server. The server side calculates the spatial distribution of each vehicle on the road based on the positioning of each vehicle, and then feeds back the personality portrait and bad driving habits of drivers on other vehicles which are adjacent to the current vehicle to the local client side according to the spatial distribution data.
Further optionally, the analyzing the road risk factors during driving based on the image processing algorithm includes:
establishing a road condition analysis model based on a target recognition algorithm, and recognizing and classifying objects in a vehicle visual field to obtain road condition analysis data of a current road; the method comprises the following steps: establishing a road condition analysis model based on a target recognition algorithm; identifying traffic condition influence factors of the surrounding environment of the vehicle in real time;
the road condition analysis model established based on the target recognition algorithm specifically comprises the following steps:
firstly, acquiring a large number of images of road environment outside a vehicle by a vehicle-mounted camera to serve as training data, and dividing the training data into a training set and a test set; and then carrying out image preprocessing on the training set to reduce invalid and interference information in the image, wherein the image preprocessing comprises the following steps: graying processing and histogram equalization. And then, carrying out image segmentation on the preprocessed image based on a simple linear iterative clustering SLIC algorithm, and segmenting the image into a plurality of super pixels according to the areas occupied by various objects in the image. Manually labeling superpixels generated by image segmentation, and taking object categories in the candidate frame as labels, wherein the label values are various traffic condition influence factors, and the traffic condition influence factors comprise: zebra crossing, deceleration strip, traffic light, roadblock, wide-angle mirror, protective fence, car stopper, toll gate, traffic sign, traffic prohibition sign, traffic direction sign, traffic warning sign, road construction safety sign, tourist area sign, auxiliary sign, road unevenness, accumulated snow, and passing pedestrians and animals. Then, inputting the super pixels with the labels into a CNN image classifier for training, extracting the image characteristics of the super pixels corresponding to each label by the CNN image classifier, and establishing a classification decision rule; and finally, inputting the test set into a CNN image classifier, performing classification test according to a classification decision rule established by the CNN image classifier, evaluating the accuracy of a classification result, and adjusting model parameters according to the test result.
The real-time identification of the traffic condition influencing factors of the vehicle surroundings specifically comprises:
the method comprises the steps that a local client side of a vehicle downloads a road condition analysis model from a server side, then objects in a vehicle visual field are identified and classified in real time based on images collected by a vehicle-mounted camera, the influence factors of the traffic condition existing in the current road are found, the positions, the number and the distances of the influence factors of the traffic condition are counted, and road condition analysis data are obtained.
Further optionally, the calculating the probability of various traffic accidents based on the driving habit data and the character representation of surrounding drivers and the road condition analysis data comprises:
establishing a traffic accident case base, and establishing a traffic accident analysis model based on the traffic accident case base; acquiring bad driving habits and character analysis data of drivers of surrounding vehicles and road condition analysis data of a current road in real time, inputting the bad driving habits and character analysis data into a traffic accident analysis model, and calculating various possible traffic accidents; the method comprises the following steps: establishing a traffic accident case library; establishing a traffic accident probability prediction model; calculating the probability of various traffic accidents in real time;
the establishing of the traffic accident case library specifically comprises the following steps:
firstly, collecting time, place, weather, accident cause, road condition images, personal information of drivers of vehicles involved in accidents and surrounding vehicles, traffic violation records and in-vehicle activity records when each traffic accident occurs; then inputting the road condition image into a road condition analysis model to obtain road condition analysis data of a traffic accident site; then personal information and in-vehicle activity records of drivers concerning vehicles and surrounding vehicles are input into the driver character analysis model, and traffic violation records are input into the bad driving habit recognition model, so that character images and bad driving habit data of the drivers concerning the vehicles and the surrounding vehicles are obtained. And finally, recording the time, the place, the weather, the accident type and the road condition analysis data of each traffic accident, and character images and bad driving habit data of drivers of the vehicles involved in the accident and surrounding vehicles into a traffic accident case library.
The establishing of the traffic accident probability prediction model specifically comprises the following steps:
and constructing a traffic accident probability prediction model based on the traffic accident case base and the SVM classification prediction model, and analyzing the probability of various traffic accidents. Firstly, taking various traffic accident cases as samples to obtain the influence factors of the traffic accidents, wherein the influence factors comprise: time, place, weather, road condition analysis data, and personality portraits and bad driving habit data of drivers of vehicles involved in the accident and surrounding vehicles; then constructing a single classification SVM model based on the traffic accident influence factors; and then mapping the value range of the distance from the sample output in the middle of the single classification SVM model to the spherical center of the hypersphere in the model to [0,1] by using an activation function, wherein the mapping result is the probability of a certain traffic accident with a certain vehicle around.
The real-time calculation of the probability of the occurrence of various traffic accidents specifically comprises the following steps:
the method comprises the steps that a traffic accident probability prediction model is loaded to a local place by a vehicle from a server, analysis data of time, place, weather and road conditions of the current position are collected, bad driving habits and character images of drivers of surrounding vehicles are obtained from the server, the traffic accident probability prediction model is input, the probability of certain types of traffic accidents of the current vehicle and the surrounding vehicles in front, back, left and right and four diagonal angles is calculated in real time, then the probability of the various types of traffic accidents of the corresponding vehicles is sequenced, 8 traffic accident probability tables corresponding to the front, back, left and right and four diagonal angles are generated, and sequencing is dynamically updated; and finally, extracting the first traffic accident probability in the 8 traffic accident probability tables, forming a 3X3 danger matrix according to the vehicle directions, wherein the element at the center of the matrix corresponds to the current vehicle and has the value of 1, and the other 8 elements correspond to the vehicles in the 8 directions and are filled with corresponding numerical values.
Further optionally, the planning the driving operation scheme according to the current potential traffic accident probability comprises:
establishing a driving operation planning model based on a deep learning method, planning a plurality of next driving operation schemes according to the current danger matrix, and feeding back the driving operation schemes to a driver in real time; the method comprises the following steps: establishing a driving operation planning model; planning a driving operation scheme;
the establishing of the driving operation planning model specifically includes:
the method comprises the steps of establishing a driving operation planning model based on a deep learning method, firstly collecting a large number of danger matrixes in user historical data, extracting elements of the danger matrixes, using names, probability values and element coordinates of traffic accidents in the elements of the danger matrixes as features, using standard driving operations for handling the traffic accidents in the elements of the danger matrixes as labels, and training the driving operation planning model.
The planning driving operation scheme specifically comprises the following steps:
and the server side loads the driving operation planning models to each vehicle-mounted client side, the vehicle-mounted client sides plan driving operation schemes in real time according to the danger matrix data of the current vehicle, and inform drivers of implementing the driving operation schemes through voice prompt. The driving operation scheme is updated in real time according to the danger matrix data and the driving operation change of the driver; when the danger matrix data is updated or the system catches that the driver carries out driving operation, the system judges that the current node is the node for carrying out driving operation scheme planning, and replans the next driving operation scheme.
The technical scheme provided by the embodiment of the invention can have the following beneficial effects:
the invention can obtain the information of surrounding drivers and protect privacy based on a differential privacy algorithm and combined with federal learning, judge the probability of surrounding traffic accidents according to road condition data, character pictures and bad driving habits of the surrounding drivers, and decide in which direction the vehicle should walk to avoid various traffic accident dangers which are most likely to happen.
[ description of the drawings ]
Fig. 1 is a flowchart of a driving behavior road condition warning method based on federal learning according to the present invention.
Fig. 2 is a schematic diagram of a driving behavior road condition warning method based on federal learning according to the present invention.
[ detailed description ] embodiments
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in detail with reference to the accompanying drawings and specific embodiments.
Fig. 1 is a flow chart of a driving behavior road condition warning method based on federal learning according to the present invention. As shown in fig. 1, the driving behavior road condition warning method based on federal learning in this embodiment may specifically include:
step 101, automatically identifying the identity of a driver and collecting driving behavior information of the driver.
The method comprises the steps of collecting a face image of a driver when the driver enters a cab by using a camera in the vehicle, and determining the identity of the driver through a face recognition technology. After the identity is determined, acquiring driving behavior information aiming at the current driver, and acquiring a traffic violation penalty record of the driver by calling an inquiry interface of a traffic administration; then, acquiring current various in-vehicle activity records of a driver through an in-vehicle camera and an automobile data recorder; and finally, recording the collected traffic violation penalty records and the in-vehicle activities under the account number of the current driver, and if the current driver does not have the account number, indicating that the current driver uses the system for the first time, firstly creating an account number and then recording information.
And 102, carrying out driver information acquisition and transmission and privacy protection in use based on a differential privacy algorithm and combined with federal learning.
And privacy protection is carried out on driver driving behavior information and traffic violation records collected in driving by combining a differential privacy algorithm with federal learning. Training each neural network model based on a federal learning method; firstly, a local training number neural network model of a client side obtains local model parameters; then adding random noise to the local model parameters obtained after training based on a differential privacy algorithm; the server aggregates the local model parameters received from the client to obtain new global model parameters; and finally, the server broadcasts the new global model parameters to each client, and the client receives the new global model parameters and performs local calculation again.
The neural network model is trained based on federal learning.
Training each neural network model based on a federal learning method, wherein the neural network model comprises the following steps: the system comprises a neural network model for analyzing the driving habits and characters of a driver, a road condition analysis model, a traffic accident analysis model and a driving operation planning model. Firstly, a server side broadcasts each initial neural network model to a local client side of each vehicle; then, the local client trains each initial neural network model based on locally acquired personal information and driving data of the driver, and feeds back local model parameters obtained after training to the server; then the server uses a FedAVG algorithm to aggregate local model parameters received from the local client to obtain new global model parameters; finally, the server broadcasts the new global model parameters to each local client; and the local client receives the new global model parameters, updates each initial neural network model, and performs local calculation and the next round of model iteration again. For example: the in-vehicle activity record and the traffic violation penalty record of the driver reflect the driving habit and the character of the driver to a great extent, and the driving habit and the character of the driver are important influence factors of driving safety, so the in-vehicle activity record and the traffic violation penalty record of the driver need to be collected for analysis in the driving process. A vehicle may be driven by multiple people, so the identity of the driver is not fixed. When a driver enters a cab with Zusanli, the system carries out face recognition on the driver and calls an interface of a traffic control bureau to determine the identity of Zusanli and obtain a traffic violation penalty record of the Zusanli, and then a camera and a vehicle data recorder start to record various activities of Zusanli in a vehicle in real time. When Zhang three co-workers Li four enter a cab, the system carries out face recognition on the co-workers Li four and calls an interface of a traffic administration office to determine the identity of the Li four, but the Li four uses the system for the first time, and no account exists in the system, so that the system automatically establishes an account for the Li four first and then records the related information of the Li four.
And adding noise into the client model parameters based on a differential privacy algorithm.
After a neural network model is trained locally at a client side based on federal learning to obtain model parameters, random noise is added to the trained local model parameters by using a differential privacy algorithm. Adding random dynamic disturbance noise to data through a Laplace mechanism aiming at numerical data in local model parameters; aiming at non-numerical and discrete data, an exponential mechanism is adopted to determine the probability of an output result, the output probability is determined according to the proportion of the discrete data, and the higher the proportion is, the higher the corresponding discrete data output probability is.
And 103, analyzing the driving habits and characters of the driver based on a deep learning method.
Establishing an initial neural network model for driver driving habit and character analysis based on a deep learning method, wherein the initial neural network model comprises the following steps: a bad driving habit recognition model and a driver character analysis model; then broadcasting the two models to a local client of a user vehicle, and calculating character features and driving habits of each driver related to driving based on local data; then, adjusting model parameters based on the calculation result, and feeding back the updated model to the server; and finally, the server side aggregates the local models received from the local client side to obtain new global model parameters, and broadcasts the new global model parameters to the local client side again.
And training a bad driving habit recognition model based on the traffic violation record.
And training an adverse driving habit recognition model based on the historical traffic violation record and the deep learning method of the current driver. Firstly, collecting a large number of traffic violation records of various drivers as a training set and a testing set; then, carrying out data cleaning, data conversion and feature extraction on the training set, wherein the extracted features comprise: violation factors, penalty types, deductions on points or not, deduction values and violation times; then, performing label classification on the sample, and marking the sample with various bad driving habit labels, wherein the bad driving habit labels comprise: the method comprises the following steps of reverse driving, red light running, illegal lane changing, overspeed driving, illegal parking, line pressing and crossing, safety belt fastening, drunk driving, calling in driving, smoking and intentionally shielding a number plate; then inputting the training set into a bad driving habit recognition model to train a neural network, and establishing a classification decision rule; and finally, inputting a test set to test the recognition accuracy of the bad driving habit recognition model, and adjusting parameters of the bad driving habit recognition model according to a test result to improve the accuracy of the model recognition. When a numerical parameter is considered, such as a speed value of 100km/h, the numerical value is modified into 120km/h by adding random disturbance noise; when faced with non-numerical parameters, such as direction: left and right sides, since the number of the two is 40% and 60% in the recording, respectively, the true value of the direction current is not output, but the left and right sides are randomly output according to the probabilities of 40% and 60%, respectively.
And establishing a driver character analysis model based on a deep learning method.
And establishing a driver character analysis model, combining the activity record in the vehicle and the personal information of the driver, automatically carrying out character portrayal on the driver, and analyzing the driving character characteristics of the driver. Firstly, establishing a driver character analysis model based on a deep learning method; extracting characteristic values related to the characters of various drivers from personal related information of the drivers, wherein the characteristic values comprise: sex, age, occupation, education degree, driving age and bad driving habit, speech data emotion keyword extraction in combination vehicle event recorder data and the car activity record, input driver character analysis neural network as the eigenvalue and carry out model training, then output driver character portrait, character portrait includes a plurality of dimensions: driving experience, driving preferences, driving caution, driving concentration, and driving tolerance.
And analyzing bad driving habits and character portraits of drivers around the vehicle in real time.
And broadcasting the bad driving habit recognition model and the driver character analysis model to a local client of each user vehicle, and calculating the character portrait and the bad driving habit of each driver based on the current traffic violation record, the in-vehicle activity record and the personal information of each driver, which are locally collected in real time. And then the local client side feeds back the character portrait and the bad driving habits of each driver to the server side. The server side calculates the spatial distribution of each vehicle on the road based on the positioning of each vehicle, and then feeds back the personality portrait and bad driving habits of drivers on other vehicles which are adjacent to the current vehicle to the local client side according to the spatial distribution data. For example: the record of the red light running for 6 times exists in the traffic violation record of the driver A, wherein two punishments are only two hundred yuan, 3 punishments are deducted by two hundred yuan for 3 times, and 1 punishment is deducted by two hundred yuan for 6 times; the system judges that the driver A has the bad driving habit of running the red light because the red light running for more than 3 times or the fine record in the classification decision rule of the bad driving habit recognition model can be judged as the bad driving habit of running the red light, so that the system judges that the driver A has the bad driving habit of running the red light.
And 104, analyzing road risk factors in the driving process based on an image processing algorithm.
And establishing a road condition analysis model based on a target recognition algorithm, and recognizing and classifying objects in the vehicle vision field to obtain road condition analysis data of the current road. For example: the driver A is a 40-year-old male truck driver with the driving age of ten years, the driver A learns the high school graduation and has bad driving habits of violation lane changing, the vehicle speed in the vehicle data recorder is far higher than the average level, and a large number of violent emotional words are extracted from the voice keywords recorded in the vehicle; through the analysis of the driver character analysis model, the character portrait of the driver nail is obtained as follows: the driving experience is rich, the driving preference is high, the driving caution degree is low, the driving concentration degree is low, and the driving endurance degree is low.
And establishing a road condition analysis model based on a target recognition algorithm.
Firstly, acquiring a large number of images of road environment outside a vehicle by a vehicle-mounted camera to serve as training data, and dividing the training data into a training set and a test set; and then carrying out image preprocessing on the training set to reduce invalid and interference information in the image, wherein the image preprocessing comprises the following steps: graying processing and histogram equalization. And then, carrying out image segmentation on the preprocessed image based on a simple linear iterative clustering SLIC algorithm, and segmenting the image into a plurality of super pixels according to the areas occupied by various objects in the image. Manually labeling superpixels generated by image segmentation, and taking object categories in the candidate frame as labels, wherein the label values are various traffic condition influence factors, and the traffic condition influence factors comprise: zebra crossing, deceleration strip, traffic lights, roadblocks, wide-angle mirror, guard rail, car stopper, toll station, traffic sign, traffic prohibition sign, traffic direction sign, traffic warning sign, road construction safety sign, tourist area sign, auxiliary sign, road unevenness, accumulated snow, and passing pedestrians and animals. Then, inputting the super-pixels with the labels into a CNN image classifier for training, extracting the image characteristics of the super-pixels corresponding to each label by the CNN image classifier, and establishing a classification decision rule; and finally, inputting the test set into a CNN image classifier, performing classification test according to a classification decision rule established by the CNN image classifier, evaluating the accuracy of a classification result, and adjusting model parameters according to the test result. For example: when the two drivers, the driver A and the driver A run side by side on the highway, the system finds that the two drivers are adjacent and drive in the same direction according to the positions of the two vehicles obtained by the vehicle-mounted GPS, and then the system feeds back the character portrait and the bad driving habit data obtained from the local clients of the two drivers to the local client of the opposite side respectively.
The traffic condition influencing factors of the vehicle surroundings are identified in real time.
The method comprises the steps that a local client side of a vehicle downloads a road condition analysis model from a server side, then objects in a vehicle visual field are identified and classified in real time based on images collected by a vehicle-mounted camera, the influence factors of the traffic condition existing in the current road are found, the positions, the number and the distances of the influence factors of the traffic condition are counted, and road condition analysis data are obtained.
And 105, calculating the probability of various traffic accidents based on the bad driving habit data, the character portrait and the road condition analysis data of surrounding drivers.
And establishing a traffic accident case base, and establishing a traffic accident analysis model based on the traffic accident case base. And acquiring the analysis data of bad driving habits and characters of drivers of surrounding vehicles and the analysis data of the road condition of the current road in real time, inputting the analysis data into a traffic accident analysis model, and calculating various traffic accidents which may occur.
And establishing a traffic accident case library.
Firstly, collecting time, place, weather, accident cause, road condition images, personal information of drivers of vehicles involved in accidents and surrounding vehicles, traffic violation records and in-vehicle activity records when each traffic accident occurs; then inputting the road condition image into a road condition analysis model to obtain road condition analysis data of a traffic accident site; then personal information and in-vehicle activity records of drivers concerning vehicles and surrounding vehicles are input into the driver character analysis model, and traffic violation records are input into the bad driving habit recognition model, so that character images and bad driving habit data of the drivers concerning the vehicles and the surrounding vehicles are obtained. And finally, recording the time, the place, the weather, the accident type and the road condition analysis data of each traffic accident, and character images and bad driving habit data of drivers of the vehicles involved in the accident and surrounding vehicles into a traffic accident case library.
And establishing a traffic accident probability prediction model.
And constructing a traffic accident probability prediction model based on a traffic accident case base and a Support Vector Machine (SVM) classification prediction model, and analyzing the occurrence probability of various traffic accidents. Firstly, taking various traffic accident cases as samples to obtain the influence factors of the traffic accidents, wherein the influence factors comprise: time, place, weather, road condition analysis data, and personality portraits and bad driving habit data of drivers of vehicles involved in the accident and surrounding vehicles; then constructing a single classification SVM model based on the traffic accident influence factors; and then mapping the value range of the distance from the sample output in the middle of the single classification SVM model to the spherical center of the hypersphere in the model to [0,1] by using an activation function, wherein the mapping result is the probability of a certain traffic accident with a certain vehicle around.
And calculating the probability of various traffic accidents in real time.
The method comprises the steps that a traffic accident probability prediction model is loaded to a local place by a vehicle from a server, time, place, weather and road condition analysis data of the current position are collected, then bad driving habits and character images of drivers of surrounding vehicles are obtained from the server, the traffic accident probability prediction model is input, the probability of a certain type of traffic accident occurring between the current vehicle and the surrounding vehicles in front, back, left and right and four diagonal angles is calculated in real time, then the probability of the occurrence of various types of traffic accidents corresponding to each vehicle is sequenced, 8 traffic accident probability tables corresponding to the front, back, left and right and four diagonal angles are generated, and sequencing is dynamically updated; and finally, extracting the first traffic accident probability in the 8 traffic accident probability tables, forming a 3X3 danger matrix according to the vehicle directions, wherein the element at the center of the matrix corresponds to the current vehicle and has the value of 1, and the other 8 elements correspond to the vehicles in the 8 directions and are filled with corresponding numerical values.
And 106, planning a driving operation scheme according to the current potential traffic accident probability.
And establishing a driving operation planning model based on a deep learning method, planning a plurality of next driving operation schemes according to the current danger matrix, and feeding back the driving operation schemes to the driver in real time. For example: in various traffic accident cases, accidents such as red light running, rear-end collision, road throwing and road side facility collision easily occur when a driver drives at an overspeed, and particularly the accident probability is suddenly increased under the conditions of poor weather, road conditions, time and places. Therefore, on a certain highway, when a driver runs at an overspeed, the probability change of various traffic accidents is analyzed in real time through the SVM classification prediction model by combining the bad driving habit and character portrait of the current driver and the current time, place, weather and road condition analysis data, and the probability of the traffic accidents is basically judged to be very high; the highest probability is to run the red light, then to collide roadside facilities and turn to throw out the road, and the lowest probability is to end up.
And establishing a driving operation planning model.
The method comprises the steps of establishing a driving operation planning model based on a deep learning method, firstly, collecting a large number of danger matrixes in user historical data, extracting elements of the danger matrixes, taking names, probability values and element coordinates of traffic accidents in the elements of the danger matrixes as characteristics, taking standard driving operations for dealing with the traffic accidents in the elements of the danger matrixes as labels, and training the driving operation planning model. For example: when the current vehicle is on a straight highway, acquiring time, place, weather and road condition analysis data of the current position, and combining bad driving habits and character images of drivers of surrounding vehicles to obtain a traffic accident probability table of the vehicle right ahead (10% of emergency brake injury, 7% of collision green belt, 4% of rear-end collision, 3% of scratch, 1% of collision and 0.1% of rollover); obtaining a danger matrix { 10% of emergency brake injury, 15% of drift and 10% of rollover ] by combining traffic accident probability tables of other surrounding vehicles; 7% of collision green belt, the vehicle: 1, 4 percent of rear-end collision; scratch is 3%, emergency brake is 1%, and collision to pedestrian is 0.1%.
And planning a driving operation scheme.
And the server side loads the driving operation planning model to each vehicle-mounted client side, and the vehicle-mounted client sides plan the driving operation scheme in real time according to the danger matrix data of the current vehicle and inform a driver of implementing the driving operation scheme through voice prompt. The driving operation scheme is updated in real time according to the danger matrix data and the driving operation change of the driver; when the danger matrix data is updated or the system catches that the driver carries out driving operation, the system judges that the current node is the node for carrying out driving operation scheme planning, and replans the next driving operation scheme.
Claims (7)
1. A driving behavior road condition early warning method based on federal learning is characterized by comprising the following steps:
automatically identifying the identity of a driver and collecting driving behavior information of the driver; based on difference privacy algorithm combines federal study to carry out driver information acquisition transmission and privacy protection in the use, specifically include: based on a federal learning training neural network model, noise is added into a client model parameter based on a differential privacy algorithm; the method comprises the following steps of carrying out analysis on driving habits and characters of a driver based on a deep learning method, wherein the analysis on the driving habits and the characters of the driver based on the deep learning method specifically comprises the following steps: training an adverse driving habit recognition model based on a traffic violation record, establishing a driver character analysis model based on a deep learning method, and analyzing the adverse driving habits and character images of drivers around the vehicle in real time; analyzing road risk factors in the driving process based on an image processing algorithm, wherein the analyzing of the road risk factors in the driving process based on the image processing algorithm specifically comprises the following steps: establishing a road condition analysis model based on a target identification algorithm, and identifying traffic condition influence factors of the surrounding environment of the vehicle in real time; calculate all kinds of traffic accident probability based on driver's bad driving habit data and personality portrait and road conditions analysis data around, specifically include: establishing a traffic accident case library, establishing a traffic accident probability prediction model, and calculating the occurrence probability of various traffic accidents in real time; planning a driving operation scheme according to the current potential traffic accident probability, wherein the planning of the driving operation scheme according to the current potential traffic accident probability specifically comprises the following steps: and establishing a driving operation planning model and planning a driving operation scheme.
2. The method of claim 1, wherein the automatically identifying a driver identity and collecting driver driving behavior information comprises:
the method comprises the steps that an in-vehicle camera is used for collecting a face image of a driver when the driver enters a cab, and the identity of the driver is determined through a face recognition technology; after the identity is determined, acquiring driving behavior information aiming at the current driver, and acquiring a traffic violation penalty record of the driver by calling an inquiry interface of a traffic administration; then, acquiring current various in-vehicle activity records of a driver through an in-vehicle camera and an automobile data recorder; and finally, recording the collected traffic violation penalty records and the in-vehicle activities under the account number of the current driver, and if the current driver does not have the account number, indicating that the current driver uses the system for the first time, firstly creating an account number and then recording information.
3. The method of claim 1, wherein the differential privacy based algorithm in combination with federal learning for driver information collection transmission and in-use privacy protection comprises:
privacy protection is carried out on driver driving behavior information and traffic violation records collected during driving through a difference privacy algorithm in combination with federal learning; training each neural network model based on a federal learning method; firstly, a local training number neural network model of a client side obtains local model parameters; then adding random noise to the local model parameters obtained after training based on a differential privacy algorithm; the server aggregates the local model parameters received from the client to obtain new global model parameters; finally, the server broadcasts the new global model parameters to each client, and the client receives the new global model parameters and carries out local calculation again; the method comprises the following steps: training a neural network model based on federal learning; adding noise into the client model parameters based on a differential privacy algorithm;
the training of the neural network model based on federal learning specifically comprises:
training each neural network model based on a federal learning method, wherein the neural network model comprises the following steps: the system comprises a neural network model for analyzing the driving habits and characters of a driver, a road condition analysis model, a traffic accident analysis model and a driving operation planning model; firstly, a server side broadcasts each initial neural network model to a local client side of each vehicle; then, the local client trains each initial neural network model based on locally acquired personal information and driving data of the driver, and feeds back local model parameters obtained after training to the server; then the server uses a FedAVG algorithm to aggregate local model parameters received from the local client to obtain new global model parameters; finally, the server broadcasts the new global model parameters to each local client; the local client receives the new global model parameters, updates each initial neural network model, and performs local calculation and the next round of model iteration again;
the method for adding noise to the client model parameters based on the differential privacy algorithm specifically comprises the following steps:
after a neural network model is trained locally at a client side based on federal learning to obtain model parameters, random noise is added to the local model parameters obtained after training by using a differential privacy algorithm; adding random dynamic disturbance noise to data through a Laplace mechanism aiming at numerical data in local model parameters; aiming at non-numerical and discrete data, an exponential mechanism is adopted to determine the probability of an output result, the output probability is determined according to the proportion of the discrete data, and the higher the proportion is, the higher the corresponding discrete data output probability is.
4. The method of claim 1, wherein the deep learning based approach to driver driving habits and personality analysis comprises:
establishing an initial neural network model for driver driving habit and character analysis based on a deep learning method, wherein the initial neural network model comprises the following steps: a bad driving habit recognition model and a driver character analysis model; then broadcasting the two models to a local client of a user vehicle, and calculating character features and driving habits of each driver related to driving based on local data; then, adjusting model parameters based on the calculation result, and feeding back the updated model to the server; finally, the server side aggregates the local models received from the local client side to obtain new global model parameters, and broadcasts the new global model parameters to the local client side again; the method comprises the following steps: training a bad driving habit recognition model based on the traffic violation record; establishing a driver character analysis model based on a deep learning method; analyzing bad driving habits and character portraits of drivers around the vehicle in real time;
the model for recognizing the bad driving habits based on the traffic violation record training specifically comprises the following steps:
training an adverse driving habit recognition model based on a historical traffic violation record and a deep learning method of a current driver; firstly, collecting a large number of traffic violation records of various drivers as a training set and a test set; then, carrying out data cleaning, data conversion and feature extraction on the training set, wherein the extracted features comprise: violation factors, penalty types, deductions on points or not, deduction values and violation times; then, performing label classification on the sample, and marking the sample with various bad driving habit labels, wherein the bad driving habit labels comprise: the method comprises the following steps of reverse driving, red light running, illegal lane changing, overspeed driving, illegal parking, line pressing and crossing, safety belt fastening, drunk driving, calling in driving, smoking and intentionally shielding a number plate; then inputting the training set into a bad driving habit recognition model to train a neural network, and establishing a classification decision rule; finally, inputting a test set to test the recognition accuracy of the bad driving habit recognition model, and adjusting parameters of the bad driving habit recognition model according to a test result to improve the accuracy of the model recognition;
the driver character analysis model is established based on the deep learning method, and specifically comprises the following steps:
establishing a driver character analysis model, automatically carrying out character portrayal on a driver by combining the activity record in the vehicle and the personal information of the driver, and analyzing the driving character characteristics of the driver; firstly, establishing a driver character analysis model based on a deep learning method; extracting characteristic values related to the characters of various drivers from personal related information of the drivers, wherein the characteristic values comprise: sex, age, occupation, education degree, driving age and bad driving habit, speech data emotion keyword extraction in combination vehicle event recorder data and the car activity record, input driver character analysis neural network as the eigenvalue and carry out model training, then output driver character portrait, character portrait includes a plurality of dimensions: driving experience, driving preferences, driving caution, driving concentration, and driving tolerance;
the bad driving habits and the personality portrait of the driver around the real-time analysis vehicle specifically include:
broadcasting the bad driving habit recognition model and the driver character analysis model to a local client of each user vehicle, and calculating a character portrait and bad driving habits of each driver based on current traffic violation records, in-vehicle activity records and personal information of each driver, which are locally collected in real time; then the local client side feeds back the character portrait and the bad driving habits of each driver to the server side; the server side calculates the spatial distribution of each vehicle on the road based on the positioning of each vehicle, and then feeds back the personality portrait and bad driving habits of drivers on other vehicles which are adjacent to the current vehicle to the local client side according to the spatial distribution data.
5. The method of claim 1, wherein the analyzing road hazard factors during driving based on image processing algorithms comprises:
establishing a road condition analysis model based on a target recognition algorithm, and recognizing and classifying objects in a vehicle view to obtain road condition analysis data of a current road; the method comprises the following steps: establishing a road condition analysis model based on a target recognition algorithm; identifying traffic condition influence factors of the surrounding environment of the vehicle in real time;
the method for establishing the road condition analysis model based on the target recognition algorithm specifically comprises the following steps:
firstly, acquiring a large number of images of road environment outside a vehicle by a vehicle-mounted camera to serve as training data, and dividing the training data into a training set and a test set; and then carrying out image preprocessing on the training set to reduce invalid and interference information in the image, wherein the image preprocessing comprises the following steps: graying processing and histogram equalization; then, carrying out image segmentation on the preprocessed image based on a simple linear iterative clustering SLIC algorithm, and segmenting the image into a plurality of super pixels according to the areas occupied by various objects in the image; manually labeling superpixels generated by image segmentation, and taking object categories in the candidate frame as labels, wherein the label values are various traffic condition influence factors, and the traffic condition influence factors comprise: zebra crossing, deceleration strip, traffic lights, roadblock, wide-angle mirror, guard rail, car stopper, toll station, traffic indicator sign, traffic prohibition sign, traffic direction sign, traffic warning sign, road construction safety sign, tourist area sign, auxiliary sign, road unevenness, accumulated snow, and passing pedestrians and animals; then, inputting the super-pixels with the labels into a CNN image classifier for training, extracting the image characteristics of the super-pixels corresponding to each label by the CNN image classifier, and establishing a classification decision rule; finally, inputting the test set into a CNN image classifier, performing classification test according to a classification decision rule established by the CNN image classifier, evaluating the accuracy of a classification result, and adjusting model parameters according to the test result;
the real-time identification of the traffic condition influencing factors of the vehicle surroundings specifically comprises:
the method comprises the steps that a local client side of a vehicle downloads a road condition analysis model from a server side, then objects in a vehicle visual field are identified and classified in real time based on images collected by a vehicle-mounted camera, the influence factors of the traffic condition existing in the current road are found, the positions, the number and the distances of the influence factors of the traffic condition are counted, and road condition analysis data are obtained.
6. The method of claim 1, wherein the calculating of the probabilities of various types of traffic accidents based on the bad driving habit data and the personality images of surrounding drivers and the traffic analysis data comprises:
establishing a traffic accident case base, and establishing a traffic accident analysis model based on the traffic accident case base; acquiring bad driving habits and character analysis data of drivers of surrounding vehicles and road condition analysis data of a current road in real time, inputting the bad driving habits and character analysis data into a traffic accident analysis model, and calculating various possible traffic accidents; the method comprises the following steps: establishing a traffic accident case library; establishing a traffic accident probability prediction model; calculating the probability of various traffic accidents in real time;
the establishing of the traffic accident case library specifically comprises the following steps:
firstly, collecting time, place, weather, accident cause, road condition images, personal information of drivers of vehicles involved in accidents and surrounding vehicles, traffic violation records and in-vehicle activity records when each traffic accident occurs; then inputting the road condition image into a road condition analysis model to obtain road condition analysis data of a traffic accident site; then inputting personal information and in-vehicle activity records of drivers of the vehicles involved in the accident and surrounding vehicles into a driver character analysis model, and inputting traffic violation records into a bad driving habit recognition model to obtain character pictures and bad driving habit data of the drivers of the vehicles involved in the accident and surrounding vehicles; finally, recording time, place, weather, accident types, road condition analysis data, personality pictures of drivers related to vehicles and surrounding vehicles and bad driving habit data when each traffic accident occurs into a traffic accident case library;
the establishing of the traffic accident probability prediction model specifically comprises the following steps:
constructing a traffic accident probability prediction model based on a traffic accident case base and a Support Vector Machine (SVM) classification prediction model, and analyzing the occurrence probability of various traffic accidents; firstly, taking various traffic accident cases as samples to obtain the influence factors of the traffic accidents, wherein the influence factors comprise: time, place, weather, road condition analysis data, and personality portraits and bad driving habit data of drivers of vehicles involved in the accident and surrounding vehicles; then constructing a single classification SVM model based on the traffic accident influence factors; then mapping a value interval of the distance from the middle output sample of the single classification SVM model to the spherical center of the hypersphere in the model to [0,1] by using an activation function, wherein the mapping result is the probability of a certain traffic accident with a certain vehicle around;
the real-time calculation of the probability of the occurrence of various traffic accidents specifically comprises the following steps:
the method comprises the steps that a traffic accident probability prediction model is loaded to a local place by a vehicle from a server, analysis data of time, place, weather and road conditions of the current position are collected, bad driving habits and character images of drivers of surrounding vehicles are obtained from the server, the traffic accident probability prediction model is input, the probability of certain types of traffic accidents of the current vehicle and the surrounding vehicles in front, back, left and right and four diagonal angles is calculated in real time, then the probability of the various types of traffic accidents of the corresponding vehicles is sequenced, 8 traffic accident probability tables corresponding to the front, back, left and right and four diagonal angles are generated, and sequencing is dynamically updated; and finally, extracting the first traffic accident probability in the 8 traffic accident probability tables, forming a 3X3 danger matrix according to the vehicle directions, wherein the element at the center of the matrix corresponds to the current vehicle and has the value of 1, and the other 8 elements correspond to the vehicles in the 8 directions and are filled with corresponding numerical values.
7. The method of claim 1, wherein planning a driving maneuver according to the current potential traffic accident probability comprises:
establishing a driving operation planning model based on a deep learning method, planning a plurality of next driving operation schemes according to the current danger matrix, and feeding back the driving operation schemes to a driver in real time; the method comprises the following steps: establishing a driving operation planning model; planning a driving operation scheme;
the establishing of the driving operation planning model specifically includes:
establishing a driving operation planning model based on a deep learning method, firstly, collecting a large number of danger matrixes in user historical data, extracting elements of the danger matrixes, taking names, probability values and element coordinates of traffic accidents in the elements of the danger matrixes as characteristics, and taking standard driving operations for dealing with the traffic accidents in the elements of the danger matrixes as labels to train the driving operation planning model;
the planning driving operation scheme specifically comprises the following steps:
the server side loads the driving operation planning model to each vehicle-mounted client side, the vehicle-mounted client sides plan driving operation schemes in real time according to danger matrix data of the current vehicle, and inform drivers of implementing the driving operation schemes through voice prompt; the driving operation scheme is updated in real time according to the danger matrix data and the driving operation change of the driver; when the danger matrix data is updated or the system catches that the driver carries out driving operation, the system judges that the current node is the node for carrying out driving operation scheme planning, and replans the next driving operation scheme.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210587310.2A CN115035491A (en) | 2022-05-27 | 2022-05-27 | Driving behavior road condition early warning method based on federal learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210587310.2A CN115035491A (en) | 2022-05-27 | 2022-05-27 | Driving behavior road condition early warning method based on federal learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115035491A true CN115035491A (en) | 2022-09-09 |
Family
ID=83121978
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210587310.2A Pending CN115035491A (en) | 2022-05-27 | 2022-05-27 | Driving behavior road condition early warning method based on federal learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115035491A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116226787A (en) * | 2023-05-04 | 2023-06-06 | 中汽信息科技(天津)有限公司 | Commercial vehicle danger probability prediction method, equipment and medium |
| CN116311952A (en) * | 2023-05-25 | 2023-06-23 | 广州豪进摩托车股份有限公司 | Motorcycle accident prediction system based on GPS positioning analysis |
| CN116383270A (en) * | 2023-03-30 | 2023-07-04 | 万联易达物流科技有限公司 | Mining method for on-line driver transportation interests |
| CN116911610A (en) * | 2023-07-20 | 2023-10-20 | 上海钢联物流股份有限公司 | Method and system for monitoring, evaluating and early warning of driving safety risk of transport vehicle |
| CN117485265A (en) * | 2024-01-02 | 2024-02-02 | 珠海格力电器股份有限公司 | Vehicle-mounted terminal control method and device and vehicle-mounted terminal |
-
2022
- 2022-05-27 CN CN202210587310.2A patent/CN115035491A/en active Pending
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116383270A (en) * | 2023-03-30 | 2023-07-04 | 万联易达物流科技有限公司 | Mining method for on-line driver transportation interests |
| CN116383270B (en) * | 2023-03-30 | 2023-08-29 | 万联易达物流科技有限公司 | Mining method for on-line driver transportation interests |
| CN116226787A (en) * | 2023-05-04 | 2023-06-06 | 中汽信息科技(天津)有限公司 | Commercial vehicle danger probability prediction method, equipment and medium |
| CN116311952A (en) * | 2023-05-25 | 2023-06-23 | 广州豪进摩托车股份有限公司 | Motorcycle accident prediction system based on GPS positioning analysis |
| CN116311952B (en) * | 2023-05-25 | 2023-08-08 | 广州豪进摩托车股份有限公司 | Motorcycle accident prediction system based on GPS positioning analysis |
| CN116911610A (en) * | 2023-07-20 | 2023-10-20 | 上海钢联物流股份有限公司 | Method and system for monitoring, evaluating and early warning of driving safety risk of transport vehicle |
| CN117485265A (en) * | 2024-01-02 | 2024-02-02 | 珠海格力电器股份有限公司 | Vehicle-mounted terminal control method and device and vehicle-mounted terminal |
| CN117485265B (en) * | 2024-01-02 | 2024-05-24 | 珠海格力电器股份有限公司 | Vehicle-mounted terminal control method and device and vehicle-mounted terminal |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11113961B2 (en) | Driver behavior monitoring | |
| CN115035491A (en) | Driving behavior road condition early warning method based on federal learning | |
| Wang et al. | How much data are enough? A statistical approach with case study on longitudinal driving behavior | |
| CN107430006B (en) | Method and system for avoiding accidents | |
| CN109637151B (en) | Method for identifying illegal driving of emergency lane on highway | |
| CN111862606B (en) | Illegal operating vehicle identification method based on multi-source data | |
| Chung | Injury severity analysis in taxi-pedestrian crashes: An application of reconstructed crash data using a vehicle black box | |
| EP3403219A1 (en) | Driver behavior monitoring | |
| US20070282519A1 (en) | System and method for analyzing traffic disturbances reported by vehicles | |
| CN111422202B (en) | Intelligent driving test judging method | |
| CN106448223A (en) | Haze-auto-adaptive expressway traveling speed early-warning device and method | |
| CN114446057A (en) | Accident high-incidence point identification early warning method and system based on intelligent network connection environment | |
| Xu et al. | A machine learning approach to vehicle occupancy detection | |
| CN114550445A (en) | Urban area traffic safety state evaluation method and device | |
| CN113469105A (en) | Method and device for monitoring surrounding vehicle and computer readable storage medium | |
| Huang et al. | Objective and subjective analysis to quantify influence factors of driving risk | |
| Van Houten et al. | 17-05 Effects of Safe Bicycle Passing Laws on Drivers' Behavior and Bicyclists' Safety | |
| CN115757628A (en) | Vehicle collision detection method, device, equipment and medium | |
| CN113147781A (en) | Intelligent driving information display system for automobile | |
| Srinivas et al. | Influence area of speed cameras: based on naturalistic driving data | |
| CN112200148B (en) | Decentralizing traffic integrity evaluation system | |
| CN111968365B (en) | Non-signalized intersection vehicle behavior analysis method and system and storage medium | |
| Pereira et al. | NATURALISTIC DRIVING DATABASE DEVELOPMENT AND ANALYSIS OF CRASH AND NEAR-CRASH TRAFFIC EVENTS IN HONOLULU | |
| Ibriaeva et al. | The Use of the YOLO neural network in the task of separating vehicles and pedestrians at a signal-controlled intersection | |
| CN118072551A (en) | Safety early warning method, device and system based on space-time track and electronic equipment |
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 |