Traffic geographic information data processing method and system based on deep learning

Abstract

The invention discloses a traffic geographic information data processing method and system based on deep learning, and the method comprises the steps: obtaining two-dimensional data and three-dimensional data of a target region in real time, and carrying out the preprocessing; matching the two-dimensional data and the three-dimensional data and integrating the two-dimensional data and the three-dimensional data into a unified coordinate system by adopting a coordinate conversion algorithm; fusing the two-dimensional data and the three-dimensional data which are integrated into the unified coordinate system to obtain fused traffic geographic information; constructing a traffic mode analysis model based on a deep learning network, and 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 a traffic mode and traffic information data. According to the invention, traffic mode analysis is carried out by constructing the deep learning network, and the traffic flow and the congestion condition can be effectively judged and predicted, so that a scientific basis is provided for urban traffic management, and the traffic efficiency and safety are improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a traffic geographic information data processing method and system based on deep learning. Background Art

[0002] Traditional traffic management methods often rely on manual monitoring and experience-based judgment, and are difficult to adapt to rapidly changing traffic conditions.

[0003] In this context, the transportation geographic information system (GIS) came into being and became an important tool for solving traffic problems. GIS can integrate and analyze a variety of geographic spatial data to provide a scientific basis for traffic management. However, with the advancement of data collection technology, the amount of traffic geographic information data has grown exponentially, and how to efficiently process and analyze this massive data has become an urgent problem to be solved.

[0004] At present, traffic data is mainly divided into two-dimensional data and three-dimensional data. Two-dimensional data usually includes information such as road network, traffic flow, speed, etc., while three-dimensional data provides richer spatial information such as road height, slope, and three-dimensional models of surrounding buildings. Although existing technologies have made some progress in two-dimensional data processing, there are still many challenges in the collection, processing and analysis of three-dimensional data.

[0005] In addition, the diversity and complexity of data make it difficult to effectively integrate data from different sources, leading to the emergence of information islands, which not only affects the accuracy and consistency of data, but also restricts the intelligent and real-time nature of traffic management.

[0006] Therefore, there is an urgent need for a new data processing method that can integrate two-dimensional and three-dimensional data and achieve efficient coordinate matching to improve the processing efficiency and accuracy of traffic geographic information. Summary of the invention

[0007] In view of this, the present invention provides a traffic geographic information data processing method and system based on deep learning to solve the problems existing in the background technology.

[0008] In order to achieve the above object, the present invention adopts the following technical solution: The first aspect of the present invention provides a method for processing traffic geographic information data based on deep learning, comprising: Acquire the two-dimensional data and three-dimensional data of the target area in real time, and pre-process the acquired two-dimensional data and three-dimensional data; Using coordinate conversion algorithm, the preprocessed 2D data and 3D data are matched and integrated into a unified coordinate system; Use deep learning algorithms to extract features from two-dimensional data and three-dimensional data integrated into a unified coordinate system, and fuse the extracted features through a fusion algorithm to obtain fused traffic geographic information; Building a traffic mode analysis model based on a deep learning network, and training the traffic mode analysis model using the fused traffic geographic information; The trained traffic mode analysis model is used to analyze the currently acquired traffic geographic information to obtain traffic mode and traffic information data.

[0009] Further, 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, as well as coordinate information and acquisition time information of corresponding acquisition devices; The three-dimensional data includes road height, slope, curvature information and / or three-dimensional data of the road surrounding environment and buildings obtained based on lidar, drone and / or ground measurement, as well as coordinate information and collection time information of the corresponding collection equipment.

[0010] Furthermore, the preprocessing step includes removing outliers, filling missing data and / or unifying data formats, wherein: Removing outliers includes setting a threshold based on the distribution of the data and removing data points that significantly deviate from other observations based on the threshold; Filling missing data includes using the mean or median to fill missing values ​​for numerical data; and using the previous or next valid value to fill missing values ​​for time series data. Unifying the data format includes converting the data types of all data fields to be consistent and performing unified conversion on data in different units.

[0011] Furthermore, the coordinate conversion algorithm is used to match and integrate the preprocessed two-dimensional data and three-dimensional data into a unified coordinate system, including: According to the conversion model of the coordinate system, a coordinate transfer chain between the coordinate system to be converted and the target coordinate system is determined; Determine a dynamic coordinate system in the coordinate transfer chain, and obtain a common parent node coordinate system of the coordinate system to be converted and the target coordinate system according to the dynamic coordinate system; Respectively calculating the transformation relationship between the coordinate system to be transformed and the parent node coordinate system, and the transformation relationship between the parent node coordinate system of the coordinate system to be transformed and the target coordinate system; According to the transformation relationship between the coordinate system to be transformed and the parent node coordinate system, and the transformation relationship between the parent node coordinate system of the coordinate system to be transformed and the target coordinate system, the relationship between the coordinate system to be transformed and the target coordinate system is obtained; The acquired two-dimensional data and three-dimensional data are converted using the relationship between the coordinate system to be converted and the target coordinate system, and the acquired two-dimensional data and three-dimensional data are converted to the same coordinate system.

[0012] Furthermore, the deep learning algorithm is used to extract features from the two-dimensional data and three-dimensional data integrated into the unified coordinate system, and the extracted features are fused through a fusion algorithm to obtain the fused traffic geographic information including: Use convolutional neural networks to process image data and extract visual features in traffic scenes; Analyze the point cloud data collected by radar to obtain the movement characteristics of traffic elements; The acquired visual features and motion features are input into a multi-layer perceptron for feature fusion, and a comprehensive feature vector is generated according to the feature weights of each modal data, which is the fused traffic geographic information.

[0013] Furthermore, the traffic mode analysis model constructed based on the deep learning network includes: The stacked denoising autoencoder is used as the basic network, a regressor is superimposed on the top layer of the autoencoder network, and the ReLU function is selected as the activation function. The network is transformed as follows to serve as the traffic mode analysis model: Forward propagation: ; Backward Propagation: ; in ; in ; Where: x i is the ith input value, For the ly The output value of the layer, For the ly Output in layer i For input j The weight of For the ly Output in layer i For input j The weight change of LY is the total number of layers, is the loss function in the deep learning network.

[0014] Furthermore, the method also includes visually displaying the obtained traffic mode and traffic information data on a display screen in the form of a route map and a table graph.

[0015] The second aspect of the present invention provides a traffic geographic information data processing system based on deep learning, which is used to implement the traffic geographic information data processing method based on deep learning as described in the first aspect above, comprising: A data acquisition module is used to obtain two-dimensional and three-dimensional data of the target area in real time; A preprocessing module, used for preprocessing the collected two-dimensional and three-dimensional data; The data matching module is used to match the pre-processed two-dimensional data with the three-dimensional data using a coordinate conversion algorithm, and integrate the matched data into a unified coordinate system using GIS; The data fusion module is used to extract features from the two-dimensional data and three-dimensional data integrated into a unified coordinate system using a deep learning algorithm, and fuse the extracted features through a fusion algorithm to obtain fused traffic geographic information; The model building module is used to build a traffic mode analysis model based on a deep learning network, train the traffic mode judgment network, and obtain the traffic mode analysis model parameters; The result output module is used to analyze the fused traffic geographic information using the traffic pattern analysis model to obtain the traffic pattern, predict traffic flow and congestion.

[0016] The beneficial technical effects of the present invention are as follows: The present invention achieves accurate matching and integration of data by real-time acquisition and preprocessing of two-dimensional and three-dimensional data, thereby integrating information from different sources. This process not only improves the comprehensiveness and accuracy of traffic data, but also effectively determines and predicts traffic flow and congestion by constructing a deep learning network for traffic pattern analysis, thereby providing a scientific basis for urban traffic management and improving traffic efficiency and safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flow chart of an embodiment of the method for processing traffic geographic information data of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] The embodiment of the present invention discloses a method for processing traffic geographic information data based on deep learning, such as Figure 1 As shown, including: S1. Acquire the two-dimensional data and three-dimensional data of the target area in real time, and pre-process the acquired two-dimensional and three-dimensional data.

[0020] In a specific embodiment, the two-dimensional data includes traffic flow data, vehicle GPS data, traffic signal status data and / or accident data obtained based on traffic cameras, vehicle GPS systems and / or sensors, as well as coordinate information and collection time information of corresponding collection devices; specifically including: Traffic flow monitoring: Use traffic cameras and sensors to monitor the number of vehicles, driving speed, and traffic density on a specific road section to obtain real-time traffic flow data; Vehicle GPS data: Obtain real-time location information through the vehicle's GPS system, analyze the vehicle's driving path and speed, and provide more accurate traffic flow information; Traffic signal status: Connect to the city traffic signal control system to obtain the status of traffic lights at each intersection; Accident and incident reporting: Receive accident and incident reports from the traffic management center in real time, and update traffic flow data in a timely manner to ensure the accuracy and timeliness of the data.

[0021] The three-dimensional data includes road height, slope, curvature information and / or three-dimensional data of the road surrounding environment and buildings obtained based on lidar, drones and / or ground measurement, as well as coordinate information and collection time information of the corresponding collection equipment.

[0022] As a preferred implementation scheme, the three-dimensional data of the road and its surrounding environment can be obtained through a variety of advanced technical means, including: Laser Radar (LiDAR): Use laser radar equipment to perform high-precision three-dimensional scanning to obtain information such as road height, slope, curvature, etc. Laser radar can work in different weather conditions, provide high-density point cloud data, and is suitable for complex urban environments.

[0023] Unmanned aerial vehicles (UAVs): UAVs are equipped with high-definition cameras and lidars to capture and scan data from the air to obtain three-dimensional data of large areas. UAVs can quickly cover a wide geographical area and are suitable for traffic flow monitoring and environmental assessment.

[0024] Ground survey: Combined with traditional ground survey technology, use total station, GPS and other equipment for precise measurement to obtain three-dimensional coordinate information of roads and buildings. These data can be used to calibrate and verify the accuracy of other data sources.

[0025] 3D modeling software: Use professional 3D modeling software to process the collected point cloud data and generate 3D models of roads and surrounding environments. These models can intuitively display the spatial relationship between transportation facilities, buildings and natural environment, and provide support for traffic analysis.

[0026] An example of the above data collection is: 2D data acquisition: Traffic flow monitoring: On a main road in a city, traffic cameras and sensors are used to monitor the number of vehicles, driving speed, and traffic density of a specific road section (for example, from intersection A to intersection B, a length of 2 kilometers). In a certain time period (such as 9 am to 10 am), the average number of vehicles passing every 5 minutes is monitored to be 200, the average speed is 30 kilometers per hour, and the traffic density is 100 vehicles per kilometer.

[0027] Vehicle GPS data: The real-time location information of 100 randomly selected vehicles during the time period is obtained through the vehicle's GPS system to analyze the vehicle's driving path and speed. For example, one of the vehicles starts from the starting position (longitude 116.3°, latitude 39.9°), passes through multiple locations, and finally arrives at the destination (longitude 116.4°, latitude 40.0°), with an average speed of 32 km / h.

[0028] Traffic signal status: Connected to the city traffic signal control system to obtain the signal status of each intersection (such as intersection A, intersection B, etc., a total of 5 intersections). During the above time period, the average red light duration at intersection A was 60 seconds, the average green light duration was 40 seconds, and the average yellow light duration was 5 seconds. Analyzing the impact of signal changes on traffic flow, it was found that the traffic speed dropped significantly during the red light period, and the average speed dropped to 15 kilometers per hour.

[0029] Accident and incident reports: Receive accident and incident reports from the traffic management center in real time. During the monitoring period, a minor traffic accident occurred 500 meters away from intersection A, causing the traffic volume on this section to drop by 40% within 15 minutes after the accident.

[0030] 3D data acquisition: LiDAR acquisition: Use LiDAR equipment to perform high-precision three-dimensional scanning of the above-mentioned trunk roads and their surroundings. The scanning range is an area of ​​50 meters on both sides of the road. Obtain road height information, for example, the height of the middle part of the road is 0.2 meters, and gradually decreases to 0 meters on both sides; road slope information, such as the slope of a certain section of the road is 2°; and three-dimensional data of the surrounding environment, such as the height of roadside buildings is between 10-30 meters.

[0031] Drone collection assistance: Drones equipped with high-definition cameras and lidars can be used to conduct aerial photography and scanning of a larger area (an area with a radius of 1 km centered on the main road) to obtain three-dimensional data to supplement and verify the data collected by ground lidars.

[0032] Ground measurement calibration: Combined with traditional ground measurement technology, the key locations of roads and buildings are accurately measured using total stations, GPS and other equipment. For example, the three-dimensional coordinates of an important building are measured (longitude 116.35°, latitude 39.92°, height 25 meters), which are used to calibrate and verify the accuracy of other data sources.

[0033] In a specific embodiment, preprocessing the collected two-dimensional and three-dimensional data includes removing outliers, filling missing data, and unifying data formats, specifically: Remove outliers: Set a threshold based on the distribution of the data to remove data points that significantly deviate from other observations based on the threshold; for example, for traffic flow data, if the number of vehicles monitored at a certain moment exceeds 3 standard deviations of the average number of vehicles (i.e., exceeds ,in is the number of vehicles passing every 5 minutes, is the average number of vehicles passing every 5 minutes, and n is the number of monitoring times), then the data point is considered an outlier and is removed.

[0034] Filling missing data: For numerical data, use the mean or median to fill missing values; for time series data, use the previous or next valid value to fill missing values; for example, if data on several location points of a vehicle within a certain period of time are missing, calculate the average longitude and latitude of other vehicles at the same location within the same period of time and use them as missing values.

[0035] For time series data, if the data of traffic signal status in a certain period of time is missing, the missing value is filled with the previous or next valid value. For example, if the data of traffic light status at a certain minute at a certain intersection is missing, it is filled with the data of traffic light status in the previous minute.

[0036] Unified data format: Ensure that the data types of all data fields are consistent. Unit conversion: Perform unified conversion for data in different units.

[0037] S2. Use coordinate conversion algorithm to match and integrate the preprocessed two-dimensional data and three-dimensional data into a unified coordinate system.

[0038] In a specific embodiment, the above process includes: According to the conversion model of the coordinate system, the coordinate transfer chain between the coordinate system to be converted and the target coordinate system is determined. For example, the world coordinate system is the target coordinate system, the coordinate system of the acquisition device itself is the coordinate system to be converted, and the coordinate transfer chain is {acquisition device coordinate system} → {intermediate conversion coordinate system} → {world coordinate system}.

[0039] Determine the dynamic coordinate system in the coordinate transfer chain, and obtain the common parent node coordinate system of the coordinate system to be converted and the target coordinate system based on the dynamic coordinate system. Assuming that the intermediate conversion coordinate system is a dynamic coordinate system, the common parent node coordinate system of the acquisition device coordinate system and the world coordinate system is obtained by calculation as {common parent node coordinate system}.

[0040] Calculate the transformation relationship between the coordinate system to be transformed and the parent node coordinate system, and the transformation relationship between the parent node coordinate system of the coordinate system to be transformed and the target coordinate system respectively; According to the transformation relationship between the coordinate system to be transformed and the parent node coordinate system, and the transformation relationship between the parent node coordinate system of the coordinate system to be transformed and the target coordinate system, the relationship between the coordinate system to be transformed and the target coordinate system is obtained; The acquired 2D and 3D data are transformed using the relationship between the coordinate system to be transformed and the target coordinate system, ensuring that all data are analyzed in the same coordinate system.

[0041] In this embodiment, the parent node coordinate system refers to the node coordinate system of the node directly under it in a transmission chain.

[0042] The common parent node coordinate system refers to the intersection of two transmission chains, which is called the common parent node coordinate system of the two subordinate nodes.

[0043] Specifically, calculate the translation and rotation relationship from the acquisition device coordinate system to the common parent node coordinate system, and the translation and rotation relationship from the common parent node coordinate system to the world coordinate system. Assume that the translation vector from the acquisition device coordinate system to the common parent node coordinate system is (x1, y1, z1), and the rotation matrix is ​​R1; the translation vector from the common parent node coordinate system to the world coordinate system is (x2, y2, z2), and the rotation matrix is ​​R2.

[0044] According to the above relationship, the relationship between the coordinate system to be converted and the target coordinate system is obtained, and the collected two-dimensional data and three-dimensional data are converted by using the relationship between the coordinate system to be converted and the target coordinate system. For example, for the coordinates of a certain collected two-dimensional data point (x3, y3) in the acquisition device coordinate system, by calculating x4=R2(R1(x3, y3)+(x1, y1))+(x2, y2), it is converted to the coordinates (x4, y4) in the world coordinate system. Similarly, the three-dimensional data is converted to ensure that all data are analyzed in the same coordinate system.

[0045] S3. Use deep learning algorithms to extract features from two-dimensional data and three-dimensional data integrated into a unified coordinate system, and fuse the extracted features through a fusion algorithm to obtain fused traffic geographic information.

[0046] In a specific embodiment, the above process includes: S31. Use convolutional neural networks to process image data and extract visual features in traffic scenes. In this embodiment, a convolutional neural network (CNN) is used to process image data. For example, a ResNet (residual network) or VGG (Visual Geometry Group) pre-trained model is used to extract visual features from images in traffic scenes. These models can identify important visual features such as the shape, size, and color information of vehicles. A convolutional neural network (CNN) is used to process image data, such as using pre-trained models such as a residual network (ResNet) and VGG (Visual Geometry Group) to extract features from traffic scene images, including visual features such as the shape, size, and color information of vehicles, so as to more accurately understand and analyze traffic scenes, thereby improving the efficiency of traffic management and safety monitoring.

[0047] S32, analyzing the point cloud data collected by the radar to obtain the movement characteristics of the 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 features of the vehicle, wherein the motion features include key information such as speed, distance, and angle.

[0048] S33, inputting the visual features and motion extracted in steps S31 and S32 into a multi-layer perceptron for feature fusion, and generating a comprehensive feature vector according to the feature weights of each modal data, which is the fused traffic geographic information.

[0049] Specifically, it includes: assigning a weight to the features extracted from each modal data; performing weighted calculation on each feature according to the weight of each feature to obtain the corresponding weighted feature; wherein, the weighted calculation formula is: Fi′=Wi×Fi, Fi′ is the weighted feature, Fi is the original feature extracted from each modal data, and Wi is the weight determined by analyzing the importance of historical traffic element data in traffic flow and behavior monitoring; according to the pre-trained multilayer perceptron, the weighted feature is input into the multilayer perceptron, and the comprehensive feature vector V is obtained by forward propagation calculation of the network: wherein, the multilayer perceptron includes: an input layer, a hidden layer and an output layer, the number of neurons in the input layer is the same as the number of input features, V=σ(W2·σ(W1·F′)), W1 is the weight matrix from the input layer to the hidden layer, W2 is the weight matrix from the hidden layer to the output layer, and σ is the activation function.

[0050] S4. Construct a traffic pattern analysis model based on a deep learning network, and use the fused traffic geographic information to train the traffic pattern analysis model.

[0051] In a specific embodiment, due to the good performance of the Stacked Denoising AutoEncoder (SDAE) in processing nonlinear and high-dimensional recognition problems, the present invention adopts a stacked denoising autoencoder network as the basic network, superimposes a regressor on the top layer of the autoencoder network, and selects the ReLU function as the activation function. After the network is transformed, it is used as the traffic mode analysis model for feature extraction and prediction.

[0052] Specifically, the AutoEncoder (AE) is an unsupervised deep learning algorithm that reconstructs the input value through the back propagation algorithm during the learning process to obtain a more abstract expression of the input value, and adjusts the reconstruction parameters to make the reconstructed decoded value equal to the input value. The input value is extracted through multiple layers of reconstruction units. For an autoencoder structure with only a single hidden layer, the hidden layer output .

[0053] Denoising AutoEncoder (DAE) adds noise to the autoencoder, sets some elements of the input vector to 0, and keeps the rest unchanged, and reconstructs the original input data through network model training, thereby improving network robustness. Stacked Denoising AutoEncoder (SDAE) stacks multiple layers of DAE to form a deep learning network to improve high-dimensional feature extraction performance.

[0054] The activation function is the nonlinear function part of the network, which makes the model meet the requirements of expressive ability. Its determination is an important process in building the network model. The traditional sigmoid transfer function is used as the activation function. When the error gradient is calculated by back propagation, the derivation involves division, the amount of calculation is relatively large, and the gradient disappears easily. That is, when the sigmoid approaches the saturation region, the derivative tends to 0, causing the regression result to tend to the two extremes of the value range. In order to overcome the above defects, the ReLU function is selected as the activation function of the SDAE network in the present invention, and the network is transformed as follows: Forward propagation: ; Backward Propagation: ; in ; in ; Where: x i is the input value, w, b are the parameters of the SDAE network, For the ly The output value of the layer, For the ly Output in layer i For input j The weight of For the ly Output in layer i For input j The weight change of LY is the total number of layers, is the loss function in the deep learning network; Furthermore, a regressor is superimposed on the top layer of SDAE, such as the output layer of a neural network, and then SDAE is trained through a standard multi-layer neural network supervised training method to output the regression result and analyze the fused traffic geographic information.

[0055] The structure of the regressor is as follows: 1. Input layer to hidden layer Let the input vector be , the output vector of the hidden layer neuron is , the weight matrix from the input layer to the hidden layer is (in represents the connection weight from the i-th neuron in the input layer to the j-th neuron in the hidden layer), the bias vector of the hidden layer is , then the output calculation formula of the hidden layer neurons is:

[0056] Where f is the activation function of the hidden layer. In the above scheme, the ReLU function is selected as the activation function, and its expression is .

[0057] Suppose the output vector of the output layer neuron is , the weight matrix from the hidden layer to the output layer is (That represents the connection weight from the jth neuron in the hidden layer to the kth neuron in the output layer), the bias vector of the output layer is , then the output calculation formula of the output layer neurons is:

[0058] Where g is the activation function of the output layer. For regression problems, if the range of predicted values ​​is unlimited, g can be a linear function (i.e. ); If the predicted value needs to meet a certain range (such as the congestion level between 0 and 1), it may be necessary to select a suitable activation function based on the specific situation.

[0059] It is necessary to further explain the input of the regressor: the input is the feature vector after SDAE multi-layer feature extraction. SDAE processes the fused traffic geographic information and converts it into an abstract feature representation suitable for regression analysis. These feature vectors serve as the input data of the regressor.

[0060] Regressor output: The output of the output layer depends on the traffic-related information to be predicted. For example, if the traffic flow is predicted, the output layer may have only one neuron, and its output value is the predicted traffic flow; if the traffic congestion level is predicted (assuming that the congestion level is represented by a value, such as a congestion index between 0 and 1), it can also be a neuron outputting the corresponding congestion index value; if multiple traffic attributes are to be predicted at the same time (such as traffic flow, average speed, congestion, etc.), the output layer will have the same number of neurons as the predicted attributes, and each neuron corresponds to the predicted value of one attribute.

[0061] Number of neurons in the regressor For example, in the case of predicting traffic flow values ​​or congestion index, the number of neurons in the output layer may be 1. If multiple attributes such as traffic flow, average speed, and congestion level are predicted at the same time, assuming that 3 attributes are predicted, the number of neurons in the output layer is 3, corresponding to the prediction results of these three attributes.

[0062] During training, the parameters of SDAE and regressor are optimized by iteratively executing the forward propagation and back propagation processes until the loss function converges. For example, the loss function is set to mean square error (MSE). When MSE is less than a certain threshold (such as 0.01), the loss function is considered to have converged.

[0063] S5. Analyze the currently acquired traffic geographic information using the trained traffic mode analysis model to obtain traffic mode and traffic information data.

[0064] As a preferred implementation scheme, the trained traffic mode analysis model 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 usually contains various spatial features, such as road network structure, intersection distribution, traffic flow density, etc. The representation of these spatial features can be effectively learned through autoencoders.

[0065] Temporal characteristics: Traffic flow and transportation patterns usually change over time, so the temporal characteristics extracted from geographic information (such as the difference in flow during peak hours and non-peak hours) are also very important.

[0066] Multi-attribute features: This includes attributes of different dimensions such as vehicle speed, waiting time, traffic signal status, weather conditions (such as rain and snow), timestamps, etc. The autoencoder can effectively integrate this information and extract features that are meaningful for traffic pattern analysis.

[0067] Abnormal patterns: Abnormal situations stored in the data, such as changes in traffic flow caused by sudden traffic accidents, SDAE can identify and extract the features of these abnormal patterns during training.

[0068] Clustering features: SDAE can also help with classification and prediction by learning clustering features of data with similar properties, such as similar traffic flow patterns or geographic location features.

[0069] After obtaining the above features, the corresponding traffic mode and traffic information data can be obtained.

[0070] For example, the analysis results show that within the next hour, the traffic pattern of the above-mentioned trunk road will be at peak hours, and the traffic volume is predicted to increase by 20%, and congestion may occur between intersection A and intersection B.

[0071] As a further preferred implementation, the method in this embodiment further includes visually displaying the obtained traffic mode, predicted traffic flow and congestion situation, displaying them on a display screen in the form of a route map and a table graphic, and displaying the traffic information sign of each road on the display screen, so that the user can obtain the traffic information of each road. For example, on the display screen, a red line segment is used to represent a congested section, a green line segment is used to represent an unobstructed section, and a table is used to list the traffic flow prediction value and actual value of each section.

[0072] Another embodiment of the present invention also shows a traffic geographic information data processing system based on deep learning, which is used in the traffic geographic information data processing method shown in the above embodiment, including: A data acquisition module is used to obtain two-dimensional and three-dimensional data of the target area in real time; A preprocessing module, used for preprocessing the collected two-dimensional and three-dimensional data; The data matching module is used to match the pre-processed two-dimensional data with the three-dimensional data using a coordinate conversion algorithm, and integrate the matched data into a unified coordinate system using GIS; The data fusion module is used to extract features from the two-dimensional data and three-dimensional data integrated into a unified coordinate system using a deep learning algorithm, and fuse the extracted features through a fusion algorithm to obtain fused traffic geographic information; The model building module is used to build a traffic mode analysis model based on a deep learning network, train the traffic mode judgment network, and obtain the traffic mode analysis model parameters; The result output module is used to analyze the fused traffic geographic information using the traffic pattern analysis model to obtain the traffic pattern, predict traffic flow and congestion.

[0073] It should be noted that, for the system and its various functional modules disclosed in this embodiment, since they correspond to the methods disclosed in the aforementioned embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0074] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for processing traffic geographic information data based on deep learning, characterized in that: include: Acquire the two-dimensional data and three-dimensional data of the target area in real time, and pre-process the acquired two-dimensional data and three-dimensional data; Using coordinate conversion algorithm, the preprocessed 2D data and 3D data are matched and integrated into a unified coordinate system; Use deep learning algorithms to extract features from two-dimensional data and three-dimensional data integrated into a unified coordinate system, and fuse the extracted features through a fusion algorithm to obtain fused traffic geographic information; Building a traffic mode analysis model based on a deep learning network, and training the traffic mode analysis model using the fused traffic geographic information; The trained traffic mode analysis model is used to analyze the currently acquired traffic geographic information to obtain traffic mode and traffic information data.

2. The method for processing traffic geographic information data based on deep learning according to claim 1, characterized in that: 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, as well as coordinate information and acquisition time information of corresponding acquisition devices; The three-dimensional data includes road height, slope, curvature information and / or three-dimensional data of the road surrounding environment and buildings obtained based on lidar, drone and / or ground measurement, as well as coordinate information and collection time information of the corresponding collection equipment.

3. The method for processing traffic geographic information data based on deep learning according to claim 1, characterized in that: The preprocessing step includes removing outliers, filling missing data and / or unifying data formats, wherein: Removing outliers includes setting a threshold based on the distribution of the data and removing data points that significantly deviate from other observations based on the threshold; Filling missing data includes using the mean or median to fill missing values ​​for numerical data; and using the previous or next valid value to fill missing values ​​for time series data. Unifying the data format includes converting the data types of all data fields to be consistent and performing unified conversion on data in different units.

4. The method for processing traffic geographic information data based on deep learning according to claim 1, characterized in that: The coordinate conversion algorithm is used to match the preprocessed two-dimensional data and three-dimensional data and integrate them into a unified coordinate system, including: According to the conversion model of the coordinate system, a coordinate transfer chain between the coordinate system to be converted and the target coordinate system is determined; Determine a dynamic coordinate system in the coordinate transfer chain, and obtain a common parent node coordinate system of the coordinate system to be converted and the target coordinate system according to the dynamic coordinate system; Respectively calculating the transformation relationship between the coordinate system to be transformed and the parent node coordinate system, and the transformation relationship between the parent node coordinate system of the coordinate system to be transformed and the target coordinate system; According to the transformation relationship between the coordinate system to be transformed and the parent node coordinate system, and the transformation relationship between the parent node coordinate system of the coordinate system to be transformed and the target coordinate system, the relationship between the coordinate system to be transformed and the target coordinate system is obtained; The acquired two-dimensional data and three-dimensional data are converted using the relationship between the coordinate system to be converted and the target coordinate system, and the acquired two-dimensional data and three-dimensional data are converted to the same coordinate system.

5. The method for processing traffic geographic information data based on deep learning according to claim 1, characterized in that: The deep learning algorithm is used to extract features from the two-dimensional data and three-dimensional data integrated into the unified coordinate system, and the extracted features are fused through the fusion algorithm to obtain the fused traffic geographic information including: Use convolutional neural networks to process image data and extract visual features in traffic scenes; Analyze the point cloud data collected by radar to obtain the movement characteristics of traffic elements; The acquired visual features and motion features are input into a multi-layer perceptron for feature fusion, and a comprehensive feature vector is generated according to the feature weights of each modal data, which is the fused traffic geographic information.

6. The method for processing traffic geographic information data based on deep learning according to claim 1, characterized in that: The traffic mode analysis model constructed based on the deep learning network includes: The stacked denoising autoencoder is used as the basic network, a regressor is superimposed on the top layer of the autoencoder network, and the ReLU function is selected as the activation function. The network is transformed as follows to serve as the traffic mode analysis model: Forward propagation: ; Backward Propagation: ; in ; in ; Where: x i is the ith input value, For the ly The output value of the layer, For the ly Output in layer i For input j The weight of For the ly Output in layer i For input j The weight change of LY is the total number of layers, is the loss function in the deep learning network.

7. The method for processing traffic geographic information data based on deep learning according to claim 1, characterized in that: It also includes visually displaying the obtained traffic mode and traffic information data on a display screen in the form of a route map and a table graphic.

8. A traffic geographic information data processing system based on deep learning, used to implement the traffic geographic information data processing method based on deep learning as described in any one of claims 1 to 7, characterized in that: include: A data acquisition module is used to obtain two-dimensional and three-dimensional data of the target area in real time; A preprocessing module, used for preprocessing the collected two-dimensional and three-dimensional data; The data matching module is used to match the pre-processed two-dimensional data with the three-dimensional data using a coordinate conversion algorithm, and integrate the matched data into a unified coordinate system using GIS; The data fusion module is used to extract features from the two-dimensional data and three-dimensional data integrated into a unified coordinate system using a deep learning algorithm, and fuse the extracted features through a fusion algorithm to obtain fused traffic geographic information; The model building module is used to build a traffic mode analysis model based on a deep learning network, train the traffic mode judgment network, and obtain the traffic mode analysis model parameters; The result output module is used to analyze the fused traffic geographic information using the traffic pattern analysis model to obtain the traffic pattern, predict traffic flow and congestion.

Images