CN119181249A - A traffic geographic information data processing method and system based on deep learning - Google Patents
A traffic geographic information data processing method and system based on deep learning Download PDFInfo
- Publication number
- CN119181249A CN119181249A CN202411691166.2A CN202411691166A CN119181249A CN 119181249 A CN119181249 A CN 119181249A CN 202411691166 A CN202411691166 A CN 202411691166A CN 119181249 A CN119181249 A CN 119181249A
- Authority
- CN
- China
- Prior art keywords
- traffic
- data
- coordinate system
- dimensional data
- deep learning
- 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
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
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- 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
- G08G1/0129—Traffic data processing for creating historical data or processing based on historical data
-
- 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
- G08G1/0133—Traffic data processing for classifying traffic situation
-
- 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
- G08G1/0137—Measuring and analyzing of parameters relative to traffic conditions for specific applications
- G08G1/0145—Measuring and analyzing of parameters relative to traffic conditions for specific applications for active traffic flow control
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/123—Traffic control systems for road vehicles indicating the position of vehicles, e.g. scheduled vehicles; Managing passenger vehicles circulating according to a fixed timetable, e.g. buses, trains, trams
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Analytical Chemistry (AREA)
- Chemical & Material Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Molecular Biology (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a traffic geographic information data processing method and system based on deep learning, wherein the method comprises the steps of acquiring two-dimensional data and three-dimensional data of a target area in real time and preprocessing the two-dimensional data and the three-dimensional data; the method comprises the steps of matching two-dimensional data and three-dimensional data by adopting a coordinate conversion algorithm and integrating the two-dimensional data and the three-dimensional data into a unified coordinate system, fusing the two-dimensional data integrated into the unified coordinate system to obtain fused traffic geographic information, constructing a traffic mode analysis model based on a deep learning network, training the traffic mode analysis model by using the fused traffic geographic information, and analyzing the currently acquired traffic geographic information by using the trained traffic mode analysis model to obtain traffic mode and traffic information data. According to the invention, the traffic pattern analysis is carried out by constructing the deep learning network, so that the traffic flow and the congestion situation can be effectively judged and predicted, thereby providing scientific basis for urban traffic management and improving traffic efficiency and safety.
Description
Technical Field
The invention relates to the technical field of data processing, in particular to a traffic geographic information data processing method and system based on deep learning.
Background
Traditional traffic management means often rely on manual monitoring and empirical judgment, and are difficult to adapt to rapidly changing traffic conditions.
In this context, the Geographic Information System (GIS) of traffic has been developed as an important tool for solving traffic problems. The GIS can integrate and analyze various geographic space data and provide scientific basis for traffic management. However, with the progress of data acquisition technology, the data volume of traffic geographic information grows exponentially, and how to efficiently process and analyze such massive data is a problem to be solved.
Currently, traffic data is largely divided into two-dimensional data and three-dimensional data. Two-dimensional data typically includes information about road network, traffic flow, speed, etc., while three-dimensional data provides more spatial information about road elevation, grade, three-dimensional models of surrounding buildings, etc. Despite the advances made in two-dimensional data processing in the prior art, many challenges remain in the acquisition, processing, and analysis of three-dimensional data.
In addition, the diversity and complexity of the data make it difficult to effectively integrate data from different sources, resulting in the occurrence of information islanding. This not only affects the accuracy and consistency of the data, but also restricts the intellectualization and real-time of traffic management.
Therefore, a novel data processing method is needed, which can fuse two-dimensional data and three-dimensional data and realize efficient coordinate matching so as to improve the processing efficiency and accuracy of traffic geographic information.
Disclosure of Invention
In view of the above, the present invention provides a traffic geographic information data processing method and system based on deep learning, so as to solve the problems existing in the background art.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
The first aspect of the invention provides a traffic geographic information data processing method based on deep learning, which comprises the following steps:
acquiring two-dimensional data and three-dimensional data of a target area in real time, and preprocessing the acquired two-dimensional data and three-dimensional data;
Matching the preprocessed two-dimensional data with the preprocessed three-dimensional data by adopting a coordinate conversion algorithm, and integrating the two-dimensional data and the preprocessed three-dimensional data into a unified coordinate system;
Extracting features from the two-dimensional data and the three-dimensional data integrated into the unified coordinate system by using a deep learning algorithm, and fusing the extracted features by using a fusion algorithm to obtain fused traffic geographic information;
constructing a traffic pattern analysis model based on a deep learning network, and training the traffic pattern analysis model by using the fused traffic geographic information;
and analyzing the currently acquired traffic geographic information by using the trained traffic mode analysis model to obtain traffic modes and traffic information data.
Further, the two-dimensional data comprise traffic flow data, vehicle GPS data, traffic signal state data and/or accident data acquired based on a traffic camera, a vehicle GPS system and/or a sensor, and coordinate information and acquisition time information of corresponding acquisition equipment;
The three-dimensional data comprise road height, gradient, curvature information and/or road surrounding environment and building three-dimensional data acquired based on laser radar, unmanned aerial vehicle and/or ground measurement, and coordinate information and acquisition time information of corresponding acquisition equipment.
Further, the preprocessing step includes removing outliers, filling in missing data, and/or unifying data formats, wherein:
removing abnormal values comprises setting a threshold according to the distribution condition of data, and removing data points which are significantly deviated from other observed values according to the threshold;
filling the missing data comprises filling the missing value by adopting a mean value or a median for the numerical value data, and filling the missing value by adopting a previous or a next effective value for the time sequence data;
The unified data format includes converting the data types of all data fields to be consistent, and performing unified conversion on data of different units.
Further, the matching and integrating the preprocessed two-dimensional data and the preprocessed three-dimensional data into a unified coordinate system by adopting a coordinate conversion algorithm comprises the following steps:
according to the conversion model of the coordinate system, determining a coordinate transfer chain between the coordinate system to be converted and the target coordinate system;
Determining a dynamic coordinate system in the coordinate transfer chain, and acquiring a coordinate system to be converted and a coordinate system of a common father node of the target coordinate system according to the dynamic coordinate system;
Respectively calculating a transformation relation between a coordinate system to be converted and a parent node coordinate system and a transformation relation between the parent node coordinate system of the coordinate system to be converted and a target coordinate system;
acquiring the relation from the coordinate system to be converted to the target coordinate system according to the transformation relation from the coordinate system to be converted to the parent node coordinate system and the transformation relation from the parent node coordinate system to the target coordinate system;
And converting the acquired two-dimensional data and three-dimensional data by using the relation from the coordinate system to be converted to the target coordinate system, and converting the acquired two-dimensional data and three-dimensional data into the same coordinate system.
Further, the extracting features from the two-dimensional data and the three-dimensional data integrated into the unified coordinate system by using the deep learning algorithm, and fusing the extracted features by using the fusion algorithm, so as to obtain fused traffic geographic information, wherein the obtaining of the fused traffic geographic information comprises the following steps:
processing the image data by using a convolutional neural network, and extracting visual features in a traffic scene;
analyzing point cloud data acquired by a radar to acquire motion characteristics of traffic elements;
Inputting the acquired visual features and motion features into a multi-layer sensor for feature fusion, and generating a comprehensive feature vector which is the fused traffic geographic information according to the feature weights of the modal data.
Further, the constructing the traffic pattern analysis model based on the deep learning network includes:
a stack-based noise reduction self-encoder is adopted as a basic network, a regressor is overlapped on the top layer of the self-encoder network, a ReLU function is selected as an activation function, and the network is modified as follows to be used as the traffic pattern analysis model:
Forward propagation: ;
Back propagation: ;
Wherein the method comprises the steps of ;
Wherein the method comprises the steps of;
Where x i is the ith input value,For the output value of the ly-th layer,For the output i in the ly-th layer to the input j,For the weight change of the output i to the input j in the lyth layer, LY is the total layer number,Is a loss function in a deep learning network.
Further, the method also comprises the step of visually displaying the obtained traffic mode and traffic information data on a display screen in the form of a circuit diagram and a table diagram.
A second aspect of the present invention provides a traffic geographic information data processing system based on deep learning, for implementing the traffic geographic information data processing method based on deep learning according to the first aspect, including:
the data acquisition module is used for acquiring two-dimensional data and three-dimensional data of the target area in real time;
The preprocessing module is used for preprocessing the acquired two-dimensional and three-dimensional data;
The data matching module is used for matching the preprocessed two-dimensional data with the three-dimensional data by adopting a coordinate conversion algorithm, and integrating the matched data into a unified coordinate system by utilizing a GIS;
The data fusion module is used for extracting features from the two-dimensional data and the three-dimensional data integrated into the unified coordinate system by utilizing a deep learning algorithm, and fusing the extracted features by utilizing the fusion algorithm to obtain fused traffic geographic information;
The model building module is used for building a traffic mode analysis model based on the deep learning network, and carrying out judgment network training of the traffic mode to obtain traffic mode analysis model parameters;
and the result output module is used for analyzing the fused traffic geographic information by using the traffic mode analysis model to obtain a traffic mode, a predicted traffic flow and a congestion condition.
The beneficial technical effects of the invention are as follows:
according to the invention, through acquiring and preprocessing the two-dimensional and three-dimensional data in real time, the accurate matching and integration of the data are realized, and the information from different sources is fused. The traffic data analysis method not only improves the comprehensiveness and accuracy of traffic data, but also can effectively judge and predict traffic flow and congestion conditions by constructing a deep learning network so as to provide scientific basis for urban traffic management and improve traffic efficiency and safety.
Drawings
Fig. 1 is a flow chart of an embodiment of a traffic geographic information data processing method according to the present 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.
The embodiment of the invention discloses a traffic geographic information data processing method based on deep learning, which is shown in fig. 1 and comprises the following steps:
S1, acquiring two-dimensional data and three-dimensional data of a target area in real time, and preprocessing the acquired two-dimensional data and three-dimensional data.
In one specific embodiment, the two-dimensional data comprises traffic flow data, vehicle GPS data, traffic signal state data and/or accident data acquired based on a traffic camera, a vehicle GPS system and/or a sensor, and coordinate information and acquisition time information of corresponding acquisition equipment, and the two-dimensional data comprises:
Monitoring traffic flow, namely monitoring the number of vehicles, the running speed and the traffic flow density of a specific road section through a traffic camera and a sensor, and acquiring traffic flow data in real time;
The GPS data of the vehicle is that the GPS system of the vehicle is used for acquiring real-time position information, analyzing the running path and speed of the vehicle and providing more accurate traffic flow information;
the traffic signal state is connected with an urban traffic signal control system to acquire the signal lamp state of each intersection;
And the accident and event report is received from the traffic management center in real time, and the traffic flow data is updated in time, so that the accuracy and timeliness of the data are ensured.
The three-dimensional data comprise road height, gradient, curvature information and/or road surrounding environment and building three-dimensional data acquired based on laser radar, unmanned aerial vehicle and/or ground measurement, and coordinate information and acquisition time information of corresponding acquisition equipment.
As a preferred embodiment, three-dimensional data of a road and its surrounding environment may be obtained by a variety of advanced technical means, including in particular:
Laser radar (LiDAR) uses laser radar equipment to perform high-precision three-dimensional scanning to obtain information such as height, gradient, curvature and the like of a road. The laser radar can work under different weather conditions, provides high-density point cloud data, and is suitable for complex urban environments.
Unmanned Aerial Vehicle (UAV) carries high-definition camera and laser radar to take and scan in the air to obtain three-dimensional data of large area. The unmanned aerial vehicle can quickly cover a wide geographic area, and is suitable for traffic flow monitoring and environment assessment.
And the ground measurement is carried out by combining the traditional ground measurement technology and using equipment such as a total station, a GPS and the like to carry out accurate measurement, so as to obtain the three-dimensional coordinate information of roads and buildings. These data can be used to correct and verify the accuracy of other data sources.
And the three-dimensional modeling software is used for processing the acquired point cloud data by utilizing professional three-dimensional modeling software to generate a three-dimensional model of the road and the surrounding environment. The models can intuitively display the spatial relationship among traffic facilities, buildings and natural environments, and provide support for traffic analysis.
One example of such data acquisition is:
Two-dimensional data acquisition:
traffic flow monitoring-the number of vehicles, the speed of travel and the density of traffic flow on a particular road segment (e.g., 2 km in length from an a-way junction to a B-way junction) is monitored by traffic cameras and sensors on a major thoroughfare of a city. In a certain period of time (such as 9 to 10 am), an average number of vehicles passing every 5 minutes is 200, an average vehicle speed is 30 km/h, and a vehicle flow density is 100 vehicles per km.
And the GPS data of the vehicle is that 100 pieces of real-time position information of the randomly extracted vehicle in the time period are acquired through a GPS system of the vehicle, and the running path and the speed of the vehicle are analyzed. For example, one of the vehicles starts from a starting location (116.3 ° in longitude and 39.9 ° in latitude), passes through a plurality of location points, and finally reaches a destination (116.4 ° in longitude and 40.0 ° in latitude), and the average speed is 32 km/h.
The traffic signal state is connected with the urban traffic signal control system to obtain the signal lamp states of all intersections (such as an A intersection, a B intersection and the like, and the total number of the intersections is 5). In the time period, the red light duration of the road A averages 60 seconds, the green light duration averages 40 seconds, and the yellow light duration averages 5 seconds. And analyzing the influence of signal change on traffic flow, and finding that the traffic flow speed during the red light is obviously reduced, and the average traffic speed is reduced to 15 km/h.
Accident and event report-the accident and event report from the traffic management center is received in real time. During monitoring, a slight traffic accident occurred together at 500 meters from the a-way opening, resulting in a 40% drop in traffic flow for that road segment within 15 minutes after the accident occurred.
Three-dimensional data acquisition:
And (3) laser radar acquisition, namely performing high-precision three-dimensional scanning on the trunk road and the surrounding environment by using laser radar equipment. The scanning range is an area of 50 meters on both sides of the road. The height information of the road is obtained, for example, the height of the middle part of the road is 0.2 m, both sides are gradually reduced to 0m, the road gradient information, such as the gradient of a certain section of road is 2 degrees, and the three-dimensional data of the surrounding environment, such as the height of a roadside building, is between 10 and 30 m.
Unmanned aerial vehicle acquisition is assisted in that high-definition cameras and laser radars are carried on the unmanned aerial vehicle, aerial shooting and scanning are carried out on a larger range (an area with a main road as a center and a radius of 1 km), and three-dimensional data are obtained for supplementing and verifying data acquired by the ground laser radars.
And (3) ground measurement calibration, namely combining the traditional ground measurement technology, and accurately measuring the key positions of roads and buildings by using equipment such as a total station, a GPS (global positioning system) and the like. For example, three-dimensional coordinates (116.35 degrees in longitude, 39.92 degrees in latitude, and 25 meters in height) of an important building are measured for correction and verification of accuracy of other data sources.
In a specific embodiment, preprocessing the collected two-dimensional and three-dimensional data includes removing outliers, filling in missing data, and unifying data formats, specifically:
Removing outliers by setting a threshold according to the distribution of the data to remove data points that deviate significantly from other observations according to the threshold, e.g., for traffic flow data if the number of vehicles monitored at a time exceeds 3 standard deviations (i.e., exceeds WhereinAs the number of vehicles passing every 5 minutes,For the average number of vehicles passing every 5 minutes, n is the number of monitors), the data point is considered an outlier and removed.
Filling missing data, namely filling missing values by adopting an average value or a median for numerical data, filling missing values by adopting a previous or a next effective value for time series data, for example, missing data of a plurality of position points of a certain vehicle in a certain period of time, calculating the average longitudes and latitudes of the same positions of other vehicles in the same period of time, and filling the same as the missing values.
For time series data, such as a data missing of traffic signal status in a certain period of time, the missing value is filled with the previous or next valid value. For example, an intersection is filled with one minute of signal light status when signal light status data for a minute is missing.
Unified data format, namely ensuring the consistent unit conversion of the data types of all data fields, namely carrying out unified conversion on the data of different units.
And S2, matching the preprocessed two-dimensional data and the preprocessed three-dimensional data by adopting a coordinate conversion algorithm, and integrating the two-dimensional data and the preprocessed three-dimensional data into a unified coordinate system.
In a specific embodiment, the above process includes:
And determining a coordinate transfer chain between the coordinate system to be converted and the target coordinate system according to the conversion model of the coordinate system. For example, the world coordinate system is used as the target coordinate system, the coordinate system of the acquisition device is used as the coordinate system to be converted, and the coordinate transfer chain is { acquisition device coordinate system } → { intermediate conversion coordinate system } → { world coordinate system }.
And determining a dynamic coordinate system in the coordinate transfer chain, and acquiring a common parent node coordinate system of the coordinate system to be converted and the target coordinate system according to the dynamic coordinate system. Assuming that the intermediate transformation coordinate system is a dynamic coordinate system, the common parent node coordinate system of the acquisition equipment coordinate system and the world coordinate system is { common parent node coordinate system }, which is obtained through calculation.
Respectively calculating a transformation relation between a coordinate system to be converted and a parent node coordinate system and a transformation relation between the parent node coordinate system of the coordinate system to be converted and a target coordinate system;
Obtaining the relation from the coordinate system to be converted to the target coordinate system according to the transformation relation from the coordinate system to be converted to the parent node coordinate system and the transformation relation from the parent node coordinate system to the target coordinate system;
and converting the acquired two-dimensional data and three-dimensional data by using the relation from the coordinate system to be converted to the target coordinate system, so as to ensure that all the data are analyzed under the same coordinate system.
In this embodiment, the parent node coordinate system refers to a node coordinate system having a directly subordinate node in one transfer chain.
The common parent node coordinate system refers to a common parent node coordinate system in which the intersection point of two transfer chains is called two subordinate nodes.
Specifically, the translational and rotational relationships of the acquisition device coordinate system to the common parent node coordinate system and the translational and rotational relationships of the common parent node coordinate system to the world coordinate system are calculated. Let the translation vector from the collection device coordinate system to the common parent node coordinate system be (x 1, y1, z 1), the rotation matrix be R1, the translation vector from the common parent node coordinate system to the world coordinate system be (x 2, y2, z 2), and the rotation matrix be R2.
And according to the relation, acquiring the relation from the coordinate system to be converted to the target coordinate system, and converting the acquired two-dimensional data and three-dimensional data by using the relation from the coordinate system to be converted to the target coordinate system. For example, for the coordinates of a certain acquired two-dimensional data point (x 3, y 3) in the acquisition device coordinate system, by calculating x4=r2 (R1 (x 3, y 3) + (x 1, y 1)) + (x 2, y 2), it is converted to coordinates (x 4, y 4) in the world coordinate system, and the three-dimensional data is converted in the same way, ensuring that all data are analyzed in the same coordinate system.
And S3, extracting features from the two-dimensional data and the three-dimensional data integrated into the unified coordinate system by using a deep learning algorithm, and fusing the extracted features by using a fusion algorithm to obtain fused traffic geographic information.
In a specific embodiment, the above process includes:
s31, processing image data by using a convolutional neural network, and extracting visual features in a traffic scene;
In this embodiment, the image data is processed using a convolutional neural network (Convolutional Neural Network, CNN). For example, resNet (residual network) or VGG (Visual Geometry Group) pre-trained models are used for visual feature extraction of images in traffic scenes. These models are able to identify important visual features of the vehicle, such as shape contours, size dimensions, color information, etc. The convolutional neural network (Convolutional Neural Network, CNN) is adopted to process the image data, for example, a pre-training model such as a residual network (ResNet) and a pre-training model such as a pre-training model (VGG (Visual Geometry Group)) are adopted to extract characteristics of traffic scene images, including visual characteristics such as vehicle shape outline, size and color information, so that traffic scenes can be more accurately understood and analyzed, and the efficiency of traffic management and safety monitoring is improved.
S32, analyzing point cloud data acquired by a radar, and acquiring motion characteristics of traffic elements;
In this embodiment, a point cloud processing algorithm based on deep learning, such as PointNet or PointNet ++, may be used to extract the motion characteristics of the vehicle, where the motion characteristics include key information such as speed, distance, and angle.
S33, inputting the visual features and the motions extracted in the steps S31 and S32 to a multi-layer sensor for feature fusion, and generating a comprehensive feature vector which is the fused traffic geographic information according to the feature weights of the modal data.
The method comprises the steps of distributing a weight to each mode data extracted feature, carrying out weighted calculation on each feature according to the weight of each feature to obtain a corresponding weighted feature, wherein a weighted calculation formula is that Fi '=wi×Fi, fi' is the weighted feature, fi is an original feature extracted from each mode data, wi is a weight determined by analyzing importance of historical traffic element data in traffic flow and behavior monitoring, the weighted feature is input into a multi-layer perceptron obtained through pre-training, and a comprehensive feature vector V is obtained through forward propagation calculation of a network, wherein the multi-layer perceptron comprises an input layer, a hidden layer and an output layer, the number of neurons of the input layer is the same as the number of the input feature, V=sigma (W 2·σ(W1·F′)）,W1 is a weight matrix from the input layer to the hidden layer, W 2 is a weight matrix from the hidden layer to the output layer, and sigma is an activation function.
S4, constructing a traffic pattern analysis model based on the deep learning network, and training the traffic pattern analysis model by using the fused traffic geographic information.
In a specific embodiment, because of the good performance of the stacked noise reduction self-encoder (Stacked Denoising AutoEncoder, SDAE) in dealing with the nonlinear and high-dimensional recognition problems, the invention adopts a stacked noise reduction self-encoder network as a basic network, a regressor is overlapped on the top layer of the self-encoder network, a ReLU function is selected as an activation function, and the network is modified and then is used as the traffic pattern analysis model to carry out feature extraction and prediction.
Specifically, an Automatic Encoder (AE) is an unsupervised deep learning algorithm, which reconstructs an input value through a back propagation algorithm in a learning process, obtains a more abstract expression of the input value, and equalizes a reconstructed decoding value with the input value by adjusting parameters during reconstruction. The input value realizes the feature extraction of the input value through the multi-layer reconstruction unit. For an automatic encoder structure having only a single hidden layer, wherein the hidden layer outputs。
Noise is added to the noise-reducing self-encoder (Denoising AutoEncoder, DAE) on the basis of the automatic encoder, part of elements of the input vector are set to 0, the rest elements are unchanged, and original input data is reconstructed through training of a network model, so that network robustness is improved. Stacked noise reduction self-encoder (SDAE) stacks multiple DAE to form a deep learning network to improve high-dimensional feature extraction performance.
The activation function is a nonlinear function part of the network, making the model meet the expression capability requirements, and its determination is an important process for constructing the network model. When the conventional sigmoid transfer function is used as an activation function and error gradients are calculated by back propagation, the derivation involves division, the calculated amount is relatively large, and the situation that the gradients disappear easily occurs, namely when the sigmoid approaches a saturation region, the derivative tends to be 0, and the regression result tends to be at two poles of a value range. In order to overcome the defects, the ReLU function is selected as an activation function of the SDAE network, and the network is modified as follows:
Forward propagation: ;
Back propagation: ;
Wherein the method comprises the steps of ;
Wherein the method comprises the steps of;
Wherein x i is an input value, w, b is a parameter of the SDAE network,For the output value of the ly-th layer,For the output i in the ly-th layer to the input j,For the weight change of the output i to the input j in the lyth layer, LY is the total layer number,A loss function in a deep learning network;
further, a regressor, such as an output layer of a neural network, is superimposed on the top layer of the SDAE, and then the SDAE is trained by a standard multi-layer neural network supervision training method, so that a regressive result can be output, and the fused traffic geographic information is analyzed.
The regressive device has the following structure:
1. input layer to hidden layer
Let the input vectorThe output vector of hidden layer neuron isThe weight matrix from the input layer to the hidden layer is(WhereinRepresenting the connection weight of the ith neuron of the input layer to the jth neuron of the hidden layer), the bias vector of the hidden layer isThe output calculation formula of hidden layer neurons is:
Wherein f is the activation function of the hidden layer, and the expression of selecting the ReLU function as the activation function is as follows 。
Let the output vector of the output layer neuron beThe weight matrix from the hidden layer to the output layer is(It isRepresenting the connection weight of the jth neuron of the hidden layer to the kth neuron of the output layer), the bias vector of the output layer isThe output calculation formula of the output layer neuron is:
where g is the activation function of the output layer, for regression problems, g may be a linear function if the predicted value range is unlimited (i.e ) If the predicted value needs to meet a certain range (e.g. the congestion level is between 0 and 1), then it may be necessary to select an appropriate activation function according to the circumstances.
It should be further described that the input to the regressor is a feature vector after the extraction of the SDAE multi-layer features. The SDAE processes the fused traffic geographic information and converts the traffic geographic information into abstract feature representations suitable for regression analysis, and the feature vectors are used as input data of a regressor.
Regressor output-the output of the output layer depends on the traffic-related information to be predicted. For example, if the predicted traffic flow value is a predicted traffic flow value, the output layer may have only one neuron, the output value is the predicted traffic flow value, if the predicted traffic congestion degree is a predicted traffic congestion degree (the congestion degree is represented by a value, such as a congestion index between 0 and 1), the output layer may also output a corresponding congestion index value, and if multiple traffic attributes (such as traffic flow, average speed, congestion condition, etc.) are predicted at the same time, the output layer may have neurons with the same number as the predicted attributes, and each neuron corresponds to a predicted value of one attribute.
Number of neurons of regressor
As in the case of the predicted traffic flow value or congestion index described above, the number of output layer neurons may be 1. If a plurality of attributes such as traffic flow, average speed, congestion degree and the like are predicted at the same time, and 3 attributes are predicted, the number of neurons of the output layer is 3, and the prediction results of the three attributes are respectively corresponding.
During training, the parameters of the SDAE and the regressor are optimized through the processes of iterative forward propagation and backward propagation until the loss function converges, and the training of the model can be completed. For example, the loss function is set to Mean Square Error (MSE), and is considered to converge when MSE is less than a certain threshold (e.g., 0.01).
And S5, analyzing the currently acquired traffic geographic information by using the trained traffic mode analysis model to obtain traffic modes and traffic information data.
As a preferred embodiment, the traffic pattern analysis model after training can perform feature learning and multi-attribute feature extraction on the input traffic geographic information to obtain one or more of the following features:
spatial features-traffic geographic information typically includes various spatial features such as road network structure, intersection distribution, traffic flow density, and the like. The representation of these spatial features can be efficiently learned by the self-encoder.
Time characteristics-traffic flow and transportation patterns typically vary over time, so that time characteristics extracted from geographic information (e.g., peak hours, off-peak hours, flow differences, etc.) are also important.
Multi-attribute features-this includes attributes of different dimensions of vehicle speed, waiting time, traffic signal status, weather conditions (e.g., rain and snow), time stamps, etc. The self-encoder can effectively integrate the information and extract features that are meaningful for traffic pattern analysis.
Abnormal patterns-abnormal conditions stored in data, such as traffic flow changes caused by sudden traffic accidents, the SDAE can identify and extract features of these abnormal patterns in training.
Clustering features the SDAE may also aid in classification and prediction by learning data clustering features of similar nature, such as similar traffic flow patterns or geographic location features.
After the characteristics are obtained, the corresponding traffic mode and traffic information data can be obtained.
For example, the analysis result shows that in the future one hour, the traffic mode of the arterial road is the peak period, the predicted traffic flow will increase by 20%, and a congestion situation may occur between the road a and the road B.
As a further preferred implementation manner, the method in this embodiment further includes visually displaying the obtained traffic mode, the predicted traffic flow and the congestion situation, displaying the traffic information identifier of each road on the display screen in a form of a circuit diagram and a table diagram, and enabling the user to obtain information of the traffic information of each road. For example, a red line segment is used for representing a congestion road section, a green line segment is used for representing a clear road section, and information such as a traffic flow predicted value and an actual value of each road section is listed in a table.
Another embodiment of the present invention also shows a traffic geographic information data processing system based on deep learning, which is used for the traffic geographic information data processing method shown in the foregoing embodiment, and includes:
the data acquisition module is used for acquiring two-dimensional data and three-dimensional data of the target area in real time;
The preprocessing module is used for preprocessing the acquired two-dimensional and three-dimensional data;
The data matching module is used for matching the preprocessed two-dimensional data with the three-dimensional data by adopting a coordinate conversion algorithm, and integrating the matched data into a unified coordinate system by utilizing a GIS;
The data fusion module is used for extracting features from the two-dimensional data and the three-dimensional data integrated into the unified coordinate system by utilizing a deep learning algorithm, and fusing the extracted features by utilizing the fusion algorithm to obtain fused traffic geographic information;
The model building module is used for building a traffic mode analysis model based on the deep learning network, and carrying out judgment network training of the traffic mode to obtain traffic mode analysis model parameters;
and the result output module is used for analyzing the fused traffic geographic information by using the traffic mode analysis model to obtain a traffic mode, a predicted traffic flow and a congestion condition.
It should be noted that, for the system and its functional modules disclosed in this embodiment, since they correspond to the methods disclosed in the foregoing embodiments, the description is relatively simple, and the relevant points refer to the description of the method section.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (8)
1. The traffic geographic information data processing method based on deep learning is characterized by comprising the following steps of:
acquiring two-dimensional data and three-dimensional data of a target area in real time, and preprocessing the acquired two-dimensional data and three-dimensional data;
Matching the preprocessed two-dimensional data with the preprocessed three-dimensional data by adopting a coordinate conversion algorithm, and integrating the two-dimensional data and the preprocessed three-dimensional data into a unified coordinate system;
Extracting features from the two-dimensional data and the three-dimensional data integrated into the unified coordinate system by using a deep learning algorithm, and fusing the extracted features by using a fusion algorithm to obtain fused traffic geographic information;
constructing a traffic pattern analysis model based on a deep learning network, and training the traffic pattern analysis model by using the fused traffic geographic information;
and analyzing the currently acquired traffic geographic information by using the trained traffic mode analysis model to obtain traffic modes and traffic information data.
2. The deep learning-based traffic geographic information data processing method according to claim 1, wherein the two-dimensional data includes traffic flow data, vehicle GPS data, traffic signal status data and/or accident data acquired based on traffic cameras, vehicle GPS systems and/or sensors, and coordinate information and acquisition time information of corresponding acquisition devices;
The three-dimensional data comprise road height, gradient, curvature information and/or road surrounding environment and building three-dimensional data acquired based on laser radar, unmanned aerial vehicle and/or ground measurement, and coordinate information and acquisition time information of corresponding acquisition equipment.
3. The deep learning based traffic geographic information data processing method of claim 1 wherein the preprocessing step comprises removing outliers, filling in missing data and/or unifying data formats, wherein:
removing abnormal values comprises setting a threshold according to the distribution condition of data, and removing data points which are significantly deviated from other observed values according to the threshold;
filling the missing data comprises filling the missing value by adopting a mean value or a median for the numerical value data, and filling the missing value by adopting a previous or a next effective value for the time sequence data;
The unified data format includes converting the data types of all data fields to be consistent, and performing unified conversion on data of different units.
4. The method for processing traffic geographic information data based on deep learning according to claim 1, wherein the matching and integrating the preprocessed two-dimensional data and three-dimensional data into a unified coordinate system by using a coordinate conversion algorithm comprises:
according to the conversion model of the coordinate system, determining a coordinate transfer chain between the coordinate system to be converted and the target coordinate system;
Determining a dynamic coordinate system in the coordinate transfer chain, and acquiring a coordinate system to be converted and a coordinate system of a common father node of the target coordinate system according to the dynamic coordinate system;
Respectively calculating a transformation relation between a coordinate system to be converted and a parent node coordinate system and a transformation relation between the parent node coordinate system of the coordinate system to be converted and a target coordinate system;
acquiring the relation from the coordinate system to be converted to the target coordinate system according to the transformation relation from the coordinate system to be converted to the parent node coordinate system and the transformation relation from the parent node coordinate system to the target coordinate system;
And converting the acquired two-dimensional data and three-dimensional data by using the relation from the coordinate system to be converted to the target coordinate system, and converting the acquired two-dimensional data and three-dimensional data into the same coordinate system.
5. The method for processing traffic geographic information data based on deep learning according to claim 1, wherein the step of extracting features from two-dimensional data and three-dimensional data integrated into a unified coordinate system by using a deep learning algorithm, and fusing the extracted features by using a fusion algorithm, the step of obtaining fused traffic geographic information comprises the steps of:
processing the image data by using a convolutional neural network, and extracting visual features in a traffic scene;
analyzing point cloud data acquired by a radar to acquire motion characteristics of traffic elements;
Inputting the acquired visual features and motion features into a multi-layer sensor for feature fusion, and generating a comprehensive feature vector which is the fused traffic geographic information according to the feature weights of the modal data.
6. The method for processing traffic geographic information data based on deep learning according to claim 1, wherein the constructing a traffic pattern analysis model based on the deep learning network comprises:
a stack-based noise reduction self-encoder is adopted as a basic network, a regressor is overlapped on the top layer of the self-encoder network, a ReLU function is selected as an activation function, and the network is modified as follows to be used as the traffic pattern analysis model:
Forward propagation: ;
Back propagation: ;
Wherein the method comprises the steps of ;
Wherein the method comprises the steps of;
Where x i is the ith input value,For the output value of the ly-th layer,For the output i in the ly-th layer to the input j,For the weight change of the output i to the input j in the lyth layer, LY is the total layer number,Is a loss function in a deep learning network.
7. The deep learning-based traffic geographic information data processing method of claim 1 further comprising visually displaying the obtained traffic pattern and traffic information data on a display screen in the form of a route map and a tabular graphic.
8. A deep learning-based traffic geographic information data processing system for implementing the deep learning-based traffic geographic information data processing method as claimed in any one of claims 1 to 7, comprising:
the data acquisition module is used for acquiring two-dimensional data and three-dimensional data of the target area in real time;
The preprocessing module is used for preprocessing the acquired two-dimensional and three-dimensional data;
The data matching module is used for matching the preprocessed two-dimensional data with the three-dimensional data by adopting a coordinate conversion algorithm, and integrating the matched data into a unified coordinate system by utilizing a GIS;
The data fusion module is used for extracting features from the two-dimensional data and the three-dimensional data integrated into the unified coordinate system by utilizing a deep learning algorithm, and fusing the extracted features by utilizing the fusion algorithm to obtain fused traffic geographic information;
The model building module is used for building a traffic mode analysis model based on the deep learning network, and carrying out judgment network training of the traffic mode to obtain traffic mode analysis model parameters;
and the result output module is used for analyzing the fused traffic geographic information by using the traffic mode analysis model to obtain a traffic mode, a predicted traffic flow and a congestion condition.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411691166.2A CN119181249A (en) | 2024-11-25 | 2024-11-25 | A traffic geographic information data processing method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202411691166.2A CN119181249A (en) | 2024-11-25 | 2024-11-25 | A traffic geographic information data processing method and system based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN119181249A true CN119181249A (en) | 2024-12-24 |
Family
ID=93898210
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202411691166.2A Pending CN119181249A (en) | 2024-11-25 | 2024-11-25 | A traffic geographic information data processing method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN119181249A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN120318446A (en) * | 2025-04-11 | 2025-07-15 | 陕西缜密技术有限公司 | A method and system for processing two-dimensional and three-dimensional data of spatiotemporal geographic information |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108960488A (en) * | 2018-06-13 | 2018-12-07 | 国网山东省电力公司经济技术研究院 | A kind of accurate prediction technique of saturation loading spatial distribution based on deep learning and Multi-source Information Fusion |
| US20210209939A1 (en) * | 2020-12-08 | 2021-07-08 | Harbin Engineering University | Large-scale real-time traffic flow prediction method based on fuzzy logic and deep LSTM |
| CN115464643A (en) * | 2022-09-13 | 2022-12-13 | 江苏小野智能装备有限公司 | Parallel robot motion coordinate system following conversion method and equipment |
| CN115984351A (en) * | 2022-12-26 | 2023-04-18 | 淮安中科晶上智能网联研究院有限公司 | Unmanned vehicle multi-sensor fusion map building method based on self-attention mechanism learning |
| CN117876642A (en) * | 2024-03-08 | 2024-04-12 | 杭州海康威视系统技术有限公司 | Method for constructing digital model, computer program product and electronic device |
| CN118212775A (en) * | 2024-03-18 | 2024-06-18 | 湖南银河伟业数字技术服务有限公司 | A road detection alarm system based on image recognition and analysis |
| CN118366312A (en) * | 2024-06-19 | 2024-07-19 | 徐州市交通控股集团智能科技有限公司 | Traffic detection system and method |
| CN118366311A (en) * | 2024-06-17 | 2024-07-19 | 中咨泰克交通工程集团有限公司 | Intelligent traffic monitoring method and device based on multi-mode data fusion and graph neural network, and electronic equipment |
-
2024
- 2024-11-25 CN CN202411691166.2A patent/CN119181249A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108960488A (en) * | 2018-06-13 | 2018-12-07 | 国网山东省电力公司经济技术研究院 | A kind of accurate prediction technique of saturation loading spatial distribution based on deep learning and Multi-source Information Fusion |
| US20210209939A1 (en) * | 2020-12-08 | 2021-07-08 | Harbin Engineering University | Large-scale real-time traffic flow prediction method based on fuzzy logic and deep LSTM |
| CN115464643A (en) * | 2022-09-13 | 2022-12-13 | 江苏小野智能装备有限公司 | Parallel robot motion coordinate system following conversion method and equipment |
| CN115984351A (en) * | 2022-12-26 | 2023-04-18 | 淮安中科晶上智能网联研究院有限公司 | Unmanned vehicle multi-sensor fusion map building method based on self-attention mechanism learning |
| CN117876642A (en) * | 2024-03-08 | 2024-04-12 | 杭州海康威视系统技术有限公司 | Method for constructing digital model, computer program product and electronic device |
| CN118212775A (en) * | 2024-03-18 | 2024-06-18 | 湖南银河伟业数字技术服务有限公司 | A road detection alarm system based on image recognition and analysis |
| CN118366311A (en) * | 2024-06-17 | 2024-07-19 | 中咨泰克交通工程集团有限公司 | Intelligent traffic monitoring method and device based on multi-mode data fusion and graph neural network, and electronic equipment |
| CN118366312A (en) * | 2024-06-19 | 2024-07-19 | 徐州市交通控股集团智能科技有限公司 | Traffic detection system and method |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN120318446A (en) * | 2025-04-11 | 2025-07-15 | 陕西缜密技术有限公司 | A method and system for processing two-dimensional and three-dimensional data of spatiotemporal geographic information |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113705636B (en) | An automatic driving vehicle trajectory prediction method, device and electronic equipment | |
| CN118025203B (en) | A method and system for predicting autonomous driving vehicle behavior by integrating complex networks and graph transformers | |
| CN113936139B (en) | A scene bird's-eye view reconstruction method and system combining visual depth information and semantic segmentation | |
| WO2022141910A1 (en) | Vehicle-road laser radar point cloud dynamic segmentation and fusion method based on driving safety risk field | |
| CN115062529B (en) | Vehicle behavior prediction method based on fusion of long short-term memory and multi-graph convolutional network | |
| CN108986207A (en) | A kind of road based on true road surface data and emulation modelling method is built along the line | |
| CN116129066A (en) | Automatic driving high-precision map model based on digital twin and high-precision static map manufacturing method | |
| CN115257785B (en) | A method and system for producing an autonomous driving data set | |
| CN118644616B (en) | Scene modeling method and system based on digital twin for realizing intelligent traffic | |
| CN117372991A (en) | Autonomous driving method and system based on multi-view and multi-modal fusion | |
| CN119181249A (en) | A traffic geographic information data processing method and system based on deep learning | |
| CN118982661A (en) | A semantic segmentation method of farm point cloud based on deep learning | |
| CN110796360A (en) | Fixed traffic detection source multi-scale data fusion method | |
| CN119693933B (en) | A real-time 3D target detection method for lidar point clouds suitable for unmanned vehicles | |
| CN112686988B (en) | Three-dimensional modeling method, three-dimensional modeling device, electronic equipment and storage medium | |
| CN114419877A (en) | Vehicle track prediction data processing method and device based on road characteristics | |
| CN117835158A (en) | Vehicle position real-time tracking method and system based on wireless communication | |
| CN113963334B (en) | A vehicle trajectory defective data repair method for roadside perception unit | |
| Bai et al. | Cyber mobility mirror for enabling cooperative driving automation: A co-simulation platform | |
| CN118172385A (en) | A method for extracting trajectory data from drone aerial video of traffic participants at complex intersections | |
| CN115994985A (en) | Automobile domain controller and automatic vehicle map building and positioning method on commute road | |
| CN118587710B (en) | Map element labeling method and device, electronic equipment and storage medium | |
| Kumar | Mobile sensor systems for field estimation and" hot spot" identification | |
| CN118397605B (en) | New energy automobile target detection method based on depth perception | |
| CN117115777B (en) | A method for identifying urban traffic risk hotspots |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20241224 |