A traffic generation prediction method based on GIS and image vectorization

Abstract

This invention relates to the field of traffic information processing technology, specifically to a traffic generation prediction method based on GIS and image vectorization. The method includes georegistring a color raster image of a planning area to obtain a calibrated image; selecting color regions from this image to construct a training sample set labeled with land use type; training an image classification model using a GIS platform and reclassifying the calibrated image; aggregating adjacent pixels of the same type in the classification results into independent vector plots to form a plot vector map; calculating the actual geometric area of ​​each plot in the map and aggregating them according to traffic zones; and finally, combining the traffic parameters and area of ​​each plot to calculate the future traffic generation and attraction of each traffic zone. This invention solves the problems of traditional methods relying on vector planning files, low efficiency, and poor classification accuracy, and achieves automatic and accurate extraction of land use information and traffic generation prediction from easily accessible planning images.

Description

Technical Field

[0001] This invention relates to the field of traffic information processing technology, and in particular to a traffic generation and prediction method based on GIS and image vectorization. Background Technology

[0002] In the feasibility study and preliminary design phases of road engineering projects, traffic volume forecasting is one of the key bases for determining the project scale. Among them, traffic generation forecasting has a decisive impact on subsequent traffic allocation. Traditional traffic generation forecasting methods usually rely on obtaining accurate vector planning files, such as land use planning data in CAD or GIS format. Through this data, different types of land parcels are identified and their areas are calculated. Then, by combining the travel rates, occurrence and attraction ratios of various land uses, the traffic generation volume of each traffic zone is predicted.

[0003] However, in the early stages of actual engineering projects, especially during the project initiation phase, it is often difficult to obtain complete and accurate electronic planning documents in a timely manner. Issues such as changes to planning schemes, data sharing barriers between departments, incompatible file formats, or copyright restrictions often prevent design units from obtaining vector planning data suitable for traffic forecasting. In contrast, pixel-based images of planning schemes (such as JPG and PNG formats) are easier to obtain, for example, from planning texts, reports, or publicly available maps. Therefore, utilizing pixel-based planning maps for rapid approximate vectorization has become an important way to improve the efficiency of traffic forecasting.

[0004] In existing technologies, attempts have been made to extract land use information from images through manual visual interpretation or simple color differentiation. Common practices include: manually georeferencing the image to give it actual coordinate information; then manually drawing or using semi-automatic tools (such as magic wand selection) to identify different color areas corresponding to different land use types; finally, calculating the area of ​​each plot and predicting traffic generation. However, these methods have obvious shortcomings: (1) georeferencing relies on manual operation, the accuracy is greatly affected by subjective experience, and the efficiency is low; (2) the identification of land use types relies on color differentiation, but actual images may have color differences, gradients, noise, or complex legends, resulting in a high classification error rate and inaccurate extraction of plot boundaries; (3) the manual processing is cumbersome and time-consuming, which is not conducive to quickly responding to project needs. In addition, existing methods mostly lack systematic machine learning classification support, making it difficult to adapt to the generalization needs of different planning legend styles, which limits their application in actual engineering.

[0005] Therefore, overcoming the above problems and providing a method that can automatically, accurately, and efficiently extract vector land use information based on common planning images and realize traffic generation prediction has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] The purpose of this invention is to provide a traffic generation and prediction method based on GIS and image vectorization, in order to solve the problems of existing technologies that rely on vector planning files, have low efficiency in manual processing, poor classification accuracy, and are difficult to quickly and automatically extract land use information from planning images and perform traffic generation and prediction.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: This invention discloses a traffic generation and prediction method based on GIS and image vectorization, comprising: A color raster image of the target planning area is acquired, and georegistration processing is performed on the color raster image to generate a calibrated image that maps the image pixel coordinate system to the real geographic coordinate system. A training sample set is constructed by selecting multiple color regions corresponding to useful land type labels from the calibrated image; An image classification model trained on the training sample set is established based on GIS. The image classification model is used to reclassify the calibrated image, and pixels with the same land use type label and spatially adjacent pixels in the classification results are aggregated into independent vector plots to obtain a plot vector map. Obtain the geometric area of ​​each vector plot in the real geographic coordinate system from the plot vector map, and collect all the vector plots belonging to the target traffic zone from the plot vector map; Obtain the traffic parameters of each vector plot in the target traffic zone, and calculate the future traffic generation and traffic attraction of the target traffic zone by combining the geometric area of ​​each vector plot in the target traffic zone.

[0008] Optionally, the georegistration process performed on the color raster image to generate a calibrated image mapping the image pixel coordinate system to the real geographic coordinate system includes: The color raster image is cropped at the boundary to obtain a registration image focused on the target planning region; Multiple pixels with known real geographic coordinates are selected in the image to be registered, and coordinate transformation parameters are obtained according to the mapping relationship between the image pixel coordinates and the known real geographic coordinates of each selected pixel in the image to be registered. The coordinate transformation parameters are used to convert the image pixel coordinates of all pixels in the image to be registered into real geographic coordinates to obtain a preliminary calibration image. The geolocation accuracy of each pixel in the image to be registered is verified until the verification result meets the preset accuracy condition, and the calibrated image is obtained.

[0009] Optionally, the step of constructing a training sample set by selecting multiple color regions corresponding to useful land type labels from the calibrated image includes: In the calibrated image, color regions representing different land use types are identified and delineated. According to the land use type represented by different colors, add a corresponding and unique land use type label to each of the color areas; Select a portion of the color region from the calibrated image, and extract color feature data from the selected color region; The extracted color feature data is associated with the land use type label corresponding to the color region to obtain training samples; All the training samples are collected and output as a structured data file for recognition and reading by the GIS platform, thus obtaining the training sample set.

[0010] Optionally, establishing the image classification model based on GIS and trained on the training sample set includes: The training sample set is read from the GIS platform, and the color feature data of each training sample is extracted. The color feature data is quantized into a feature vector, and the land use type label corresponding to the color region is used as a classification supervision signal for the feature vector. All the feature vectors and the corresponding classification supervision signals are taken as input, and a preset supervised classification algorithm is called to perform the model training process to generate an initial classification model; Cross-validation is used to optimize and adjust the parameters of the initial classification model to obtain the trained image classification model.

[0011] Optionally, the step of reclassifying the calibrated image using the image classification model and aggregating spatially adjacent pixels with the same land use type label into independent vector plots to obtain a plot vector map includes: Each pixel of the calibrated image is input into the image classification model for reclassification, and the land use type label of each pixel is output. In the reclassified and calibrated image, all pixels with the same land use type label and spatially adjacent pixels are aggregated into independent polygonal vector plots, and the shared land use type label is assigned to the corresponding vector plot. All vector plots with the aforementioned land use type label are integrated into a single vector map layer, and the resulting vector map of the plots is output.

[0012] Optionally, the traffic generation prediction method further includes, before obtaining the land parcel vector map, performing a fusion and removal process on the aggregated multiple vector land parcels, including: Read the multiple vector land parcels and their corresponding land use type labels generated after reclassification and aggregation, and identify all vector land parcels with the same land use type label and adjacent boundaries to obtain a land parcel set; All vector plots in each set of plots are merged into a continuous and single vector plot, and the land use type label common to the set of plots is assigned to the corresponding merged vector plot; Obtain the geometric area of ​​all the merged vector plots in the real geographic coordinate system, and remove the vector plots with a geometric area less than a preset area threshold. Then, integrate all the remaining vector plots to obtain the plot vector map.

[0013] Optionally, the traffic generation prediction method further includes filtering the traffic attributes of the aggregated vector land parcels before merging the land parcel set, including: Obtain traffic parameters for each land use type in the target planning area, and identify land use types in the target planning area that do not have traffic attributes; Read the multiple vector plots and corresponding land use type labels generated after reclassification and aggregation, and compare the read land use type labels one by one with land use types that do not have traffic attributes; Based on the comparison results, the vector land parcels that do not possess traffic attributes are identified and then removed.

[0014] Optionally, the traffic generation prediction method further includes calculating the future traffic generation volume of the target traffic zone by combining the traffic parameters and geometric area of ​​each vector plot in the target traffic zone, including: The target planning area is divided into several traffic zones, and a unique zone number is defined for each traffic zone to obtain a traffic zone division layer; Spatial overlay analysis is performed between the land parcel vector map and the traffic zone division layer to determine the traffic zone in which each vector land parcel in the land parcel vector map is located, and a corresponding zone number is added to each vector land parcel; Select all the vector plots that have the same community number as the target traffic community to obtain the internal plot set of the target traffic community, and obtain the travel rate and occurrence ratio of all the vector plots in the internal plot set during peak hours; Based on the travel rate and occurrence ratio of the vector plot during peak hours, and combined with the geometric area of ​​the vector plot in the real geographic coordinate system, the future traffic generation volume of the target traffic zone is calculated. The function expression for calculating the traffic generation volume is as follows:

[0015] In the formula, This represents the future traffic generation volume of the i-th traffic zone. Indicates the traffic zone number. This represents the geometric area of ​​the k-th vector plot. This represents the travel rate of the k-th vector plot during peak hours. This represents the proportion of the k-th vector plot occurring during peak hours. Indicates the index of the vector plot. This represents the total number of vector plots in the traffic zone.

[0016] Optionally, the traffic generation prediction method further includes calculating the future traffic attraction of the target traffic zone by combining the traffic parameters and geometric area of ​​each vector plot in the target traffic zone, including: Obtain the attraction ratio of all vector plots in the internal plot set of the target traffic zone during peak hours; Based on the peak-hour travel rate and attraction ratio of the vector plot, and combined with the geometric area of ​​the vector plot in the real geographic coordinate system, the future traffic attraction of the target traffic zone is calculated. The functional expression for calculating the traffic attraction is as follows:

[0017] In the formula, This represents the future traffic attraction of the i-th traffic zone. Indicates the traffic zone number. This represents the attraction ratio of the k-th vector plot during peak hours.

[0018] Optionally, the traffic generation prediction method further includes calculating and obtaining the future traffic generation and traffic attraction of the target traffic zone, and then verifying and iteratively optimizing the prediction results, including: The traffic generation and traffic attraction data obtained from each traffic zone are sorted according to the zone number and a travel generation matrix is ​​constructed. Based on the trip generation matrix and the road network topology of the target planning area, the predicted traffic flow of each road segment in the road network of the target planning area is obtained through traffic flow allocation calculation; Obtain the current traffic flow of each road segment within the target planning area, and compare the deviation between the predicted traffic flow and the current traffic flow. If the deviation value does not meet the preset deviation threshold, it is determined that the deviation originates from inaccurate traffic parameters or inaccurate land use type classification. If the deviation is determined to be due to inaccurate traffic parameters, the traffic parameters of each vector plot in the target traffic zone are reacquired until the deviation between the predicted traffic flow and the current traffic flow meets the preset deviation threshold. If the deviation is determined to be due to inaccurate land use type classification, the training sample set is reconstructed until the deviation between the predicted traffic flow and the current traffic flow meets the preset deviation threshold.

[0019] Compared with existing technologies, the traffic generation and prediction method based on GIS and image vectorization provided in this embodiment of the invention has the following advantages: The calibrated image is generated by georegistring a color raster image, laying a precise geolocation foundation for all spatial operations. A training sample set containing land use type labels provides accurate classification learning samples for the model. The image classification model is trained using the training sample set and the calibrated image is reclassified. Automated machine learning replaces manual interpretation, significantly improving the efficiency, accuracy, and generalization ability of land use identification, effectively overcoming image color difference and noise interference. By aggregating adjacent pixels of the same type in the reclassification results into independent vector plots, a plot vector map is directly generated, eliminating the need for additional manual vectorization steps. The geometric area of ​​each vector plot in the plot vector map is calculated, and all plots belonging to the target traffic zone are aggregated, providing accurate area data and correct spatial statistical units for calculation. Finally, the traffic generation and attraction of each plot within the target traffic zone are calculated based on traffic parameters, forming a complete automated method from planning images to quantitative prediction results. This fundamentally overcomes the dependence on traditional vector files and the bottlenecks of low efficiency and poor accuracy in manual processing. Attached Figure Description

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 A schematic block diagram illustrating the steps of the traffic generation and prediction method provided in this embodiment of the invention; Figure 2 The color raster image before boundary clipping provided in the embodiments of the present invention; Figure 3 This is a color raster image with cropped boundaries provided in an embodiment of the present invention; Figure 4 The vector map of the land parcels processed using an image classification model is provided for embodiments of the present invention. Detailed Implementation

[0021] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0022] This invention discloses a traffic generation and prediction method based on GIS and image vectorization, such as... Figure 1 As shown, it includes: S1. Obtain a color raster image of the target planning area, perform georegistration processing on the color raster image, and generate a calibrated image that maps the image pixel coordinate system to the real geographic coordinate system. S2. Select multiple color regions corresponding to useful land type labels from the calibrated image to construct a training sample set; S3. Based on GIS, establish an image classification model trained on a training sample set. Use the image classification model to reclassify the calibrated images and aggregate pixels with the same land use type label and spatially adjacent pixels in the classification results into independent vector plots to obtain plot vector maps. S4. Obtain the geometric area of ​​each vector plot in the plot vector map in the real geographic coordinate system, and collect all vector plots belonging to the target traffic zone from the plot vector map. S5. Obtain the traffic parameters of each vector plot in the target traffic zone, and calculate the future traffic generation and traffic attraction of the target traffic zone by combining the geometric area of ​​each vector plot in the target traffic zone.

[0023] Through the implementation of the above-described traffic generation and prediction method, a color raster image of the target planning area is first acquired and georegistration is performed on the image. This effectively solves the key problem that planning images generally lack geographic coordinate information, and generates a calibrated image that accurately maps the image pixel coordinate system to the real geographic coordinate system. This step provides an accurate and reliable geospatial benchmark for all subsequent spatial measurements and analyses, transforming the originally visually significant image into a data foundation for quantitative geographic analysis. The georegistration process preferably adopts a polynomial transformation model based on control points to achieve high-precision coordinate mapping.

[0024] The process of selecting multiple color regions corresponding to useful land type labels from the calibrated images to construct a training sample set provides crucial sample data with accurate semantic labels for subsequent supervised machine learning. The quality of the training sample set directly determines the upper limit of the model's learning. Its construction is usually accomplished by manually visually interpreting and delineating typical areas in GIS (Geographic Information System) software and assigning labels, ensuring the accurate injection of prior knowledge. An image classification model is built on the GIS platform and trained on this training sample set, such as a preferred algorithm like Support Vector Machine or Random Forest. This image classification model is then used to reclassify the calibrated images. Its core function is to completely replace the inefficient and subjective manual visual interpretation and delineation in traditional methods with automated and intelligent machine learning algorithms. This not only greatly improves processing speed but also effectively overcomes the classification difficulties caused by color difference, noise, gradation, or diverse legend styles in the original images through the complex features learned by the model from the samples, ensuring consistency and high accuracy in classification across different planning maps.

[0025] Pixels with the same land use type label and spatial adjacency in the reclassification results are aggregated into independent vector plots, directly generating plot vector maps. This innovative approach integrates pixel-level classification results with the generation process of vector geographic entities, completing the process in one step through GIS region growing and vectorization algorithms (such as raster-to-vector conversion based on connected component analysis). This eliminates the cumbersome intermediate steps of first outputting raster classification maps and then performing manual or semi-automatic vectorization in the traditional process, achieving a seamless and efficient conversion from classification information to structured vector data. The geometric area of ​​each vector plot in the plot vector map is obtained in the real geographic coordinate system. Utilizing GIS's precise spatial calculation capabilities for vector graphics, the geometric attributes of the graphics are transformed into the core quantitative indicator—area. This area value is one of the most fundamental and critical input parameters in the traffic generation prediction model, and its calculation accuracy directly determines the reliability of the final prediction results. The calculation process is preferably performed in a projected coordinate system to ensure the accuracy of area measurement.

[0026] The process of collecting all vector plots belonging to the target traffic zone from the plot vector map is achieved through GIS spatial overlay analysis. This establishes the spatial affiliation between micro-plots and macro-level traffic analysis statistical units, ensuring that subsequent traffic volume calculations can be correctly attributed to the corresponding traffic zones. It serves as a spatial bridge connecting land use and traffic demand forecasting. Finally, traffic parameters for each plot within the target traffic zone (such as peak-hour travel rates, occurrence ratios, and attraction ratios for various land uses determined based on local traffic surveys or standards) are obtained. Combined with the precisely calculated geometric areas of each plot, a traffic generation model is used to calculate the future traffic generation and attraction volumes for the target traffic zone. This completes the technical process from a raw, geographically uninformed planning image to a complete set of quantitative traffic demand forecasting data. It eliminates the deep dependence on traditional GIS vector planning files, solves the traffic analysis bottleneck caused by missing or untimely data acquisition in the early stages of projects, and provides strong technical support for rapid, objective, and scientific traffic planning decisions.

[0027] Furthermore, georegistration processing is performed on the color raster image to generate a calibrated image that maps the image pixel coordinate system to the real geographic coordinate system, including: The color raster image is cropped at the boundary to obtain the image to be registered, which is focused on the target planning area; Select multiple pixels with known real geographic coordinates in the image to be registered, and obtain coordinate transformation parameters based on the mapping relationship between the image pixel coordinates and the known real geographic coordinates of each selected pixel in the image to be registered. The coordinate transformation parameters are used to convert the image pixel coordinates of all pixels in the image to be registered into real geographic coordinates to obtain a preliminary calibration image. The geolocation accuracy of each pixel in the image to be registered is verified until the verification result meets the preset accuracy conditions, and the calibrated image is obtained.

[0028] Through the implementation of the above-described traffic generation and prediction method embodiments, such as Figure 2 and Figure 3 As shown, boundary cropping is performed on the color raster image to obtain a registration image focused on the target planning area. This operation effectively removes redundant information around the image that is irrelevant to the planning area, improving the computational efficiency and positioning focus of subsequent processing. By selecting multiple pixels with known real geographic coordinates in the registration image and obtaining coordinate transformation parameters based on the mapping relationship between the image pixel coordinates of each selected pixel in the registration image and the known real geographic coordinates, a precise mathematical transformation model from image space to real geographic space is established, which is the core to ensure the accuracy of the entire registration process.

[0029] By using the acquired coordinate transformation parameters, the image pixel coordinates of all pixels in the image to be registered are converted into real geographic coordinates, thus obtaining a preliminary calibration image. This process fundamentally transforms the entire image from a pixel array without geographic reference to spatial data with actual geographic significance. Finally, the geographic positioning accuracy of each pixel in the preliminary calibration image is verified, and iterative adjustments are made until the verification results meet the preset accuracy conditions. The final calibrated image is then obtained, ensuring the credibility and reliability of the registration results. This allows all subsequent spatial measurements, plot extraction, and area calculations based on this image to be established on an accurate geographic foundation, thereby fundamentally solving the problem that planning images lack coordinate information and cannot be directly used for quantitative spatial analysis.

[0030] Furthermore, a training sample set is constructed by selecting multiple color regions corresponding to useful land type labels from the calibrated image, including: Identify and delineate color regions representing different land use types in the calibrated image; Based on the land use type represented by different colors, add a corresponding and unique land use type label to each color area; Select a portion of the color region from the calibrated image and extract color feature data from the selected color region; The extracted color feature data is associated with the land use type labels of the corresponding color areas to obtain training samples; All training samples are collected and output as structured data files for recognition and reading by the GIS platform, thus obtaining the training sample set.

[0031] Furthermore, an image classification model trained on a training sample set is established based on GIS, including: Read the training sample set in the GIS platform and extract the color feature data from each training sample; Color feature data is quantized into feature vectors, and the land use type label of the corresponding color area is used as the classification supervision signal for the feature vectors. All feature vectors and their corresponding classification supervision signals are taken as input, and a pre-set supervised classification algorithm is called to perform the model training process to generate an initial classification model; Cross-validation was used to optimize and adjust the parameters of the initial classification model, resulting in a trained image classification model.

[0032] Through the implementation of the traffic generation prediction method described above, color regions representing different land use types are identified and delineated in the calibrated image. This provides an accurate and representative visual sample basis for subsequent machine learning classification, ensuring the quality and coverage of the model's learning source. A corresponding and unique land use type label is added to each color region according to the land use type represented by different colors, giving each sample a clear category affiliation. This constitutes an indispensable ground truth label in supervised learning and is a fundamental prerequisite for effective error feedback and parameter optimization during the training process.

[0033] By selecting specific color regions from the calibrated image and extracting their color feature data, the intuitive color information in the image is transformed into quantifiable and computable feature values, providing the model with a digital basis for identifying and distinguishing different land use types. The extracted color feature data is associated with the corresponding land use type labels of the color regions to obtain training samples, establishing a one-to-one correspondence between "features" and "labels," forming standardized training units that can be directly used for machine learning. All training samples are collected and output as structured data files for recognition and reading by the GIS platform, ensuring high compatibility between the entire training data system and the GIS environment. This provides a standardized data interface for seamlessly integrating model training and application within the GIS platform.

[0034] The training sample set typically contains data as shown in Table 1: Table 1 Training Sample Data Table

[0035] The training sample set is read from a GIS platform, and color feature data is extracted from each training sample. This process transforms visual information into digital information that can be processed by the algorithm. It also fully leverages the advantages of the GIS environment in the integrated management of geospatial and attribute data, ensuring the consistency of spatial reference between the training data and the subsequent images to be classified. The color feature data is quantized into feature vectors, and the land use type label of the corresponding color area is used as the classification supervision signal for the feature vector. This realizes the transformation from raw features to standard input for machine learning. The quantization process is usually based on statistics (such as calculating the color mean and variance) in color spaces such as RGB or HSV. The explicit provision of the classification supervision signal is fundamental to the implementation of supervised learning, ensuring that the model's learning target direction is correct.

[0036] Using all feature vectors and corresponding classification supervision signals as input, a pre-defined supervised classification algorithm (such as a preferred support vector machine or random forest) is invoked to perform the model training process to generate an initial classification model. The core function of this step is to automatically discover and construct the complex nonlinear mapping relationship between color features and land use type labels, replacing the traditional method's reliance on manual experience to establish fixed color rules. This significantly improves the model's adaptability and classification accuracy to common color differences, uneven lighting, and legend color variations in planning images. Cross-validation is used to optimize and adjust the parameters of the initial classification model, ultimately obtaining the trained image classification model. This optimization process, by dividing the sample data into training and validation sets and iterating multiple times, effectively evaluates the model's generalization performance and prevents overfitting. By adjusting key parameters such as the kernel function parameters of the support vector machine or the tree depth of the random forest, the model maintains a high recognition rate for training samples while also exhibiting stronger robustness and reliability in classifying unseen image regions. The entire process is completed within the GIS platform, ensuring a smooth transition from training data preparation and model training to subsequent applications, thus forming an efficient, reliable, and reusable automated classification model construction solution.

[0037] Furthermore, the calibrated images are reclassified using an image classification model, and pixels with the same land use type label and spatially adjacent pixels in the classification results are aggregated into independent vector plots to obtain a plot vector map, including: Each pixel of the calibrated image is input into the image classification model for reclassification, and the land use type label of each pixel is output. In the reclassified and calibrated image, all pixels with the same land use type label and spatially adjacent pixels are aggregated into independent polygonal vector plots, and the shared land use type label of the aggregated plots is assigned to the corresponding vector plots. All vector plots with land use type labels are integrated into a single vector layer, and the resulting vector plot is output.

[0038] Furthermore, the traffic generation prediction method also includes merging and eliminating multiple aggregated vector plots before obtaining the plot vector map, including: Read the multiple vector plots and their corresponding land use type labels generated after reclassification and aggregation, and identify all vector plots with the same land use type label and adjacent boundaries to obtain a set of plots; Merge all vector plots in each plot set into a continuous and single vector plot, and assign the land use type label common to the plot sets to the corresponding merged vector plot; Obtain the geometric area of ​​all merged vector plots in the real geographic coordinate system, remove vector plots with a geometric area smaller than a preset area threshold, and then integrate all remaining vector plots to obtain a plot vector map.

[0039] By implementing the above-described traffic generation prediction method, each pixel of the calibrated image is input into the trained image classification model for reclassification, and the land use type label output by each pixel is obtained. This achieves automated and refined interpretation of the entire planning image. The consistency and objectivity of the machine learning model completely replace the human error and efficiency bottleneck caused by the traditional reliance on manual visual color differentiation, ensuring the high efficiency and accuracy of classification of massive pixels.

[0040] After reclassification, all spatially adjacent pixels with the same land use type label in the image are aggregated into independent polygonal vector plots. This step uses spatial adjacency (usually determined by four-connectivity or eight-connectivity rules) to reorganize the discrete classification results into meaningful geographic entities. Aggregation and vector generation are completed in one go using the regional grouping and vectorization functions in GIS (such as the "Raster to Polygon" tool in ArcGIS), which significantly improves data conversion efficiency. At the same time, the land use type label shared by the aggregation is assigned to the corresponding vector plots, realizing the lossless transfer and inheritance from pixel-level classification information to vector geographic feature attributes, ensuring the traceability of plot properties in subsequent analysis.

[0041] Combine all vector plots with land use type labels into a single vector layer and output the resulting plot vector map (see reference). Figure 4 This means that the final transformation from unstructured raster images to structured, attributed GIS vector data is complete. This plot vector map is the core data foundation for subsequent spatial analysis and calculation.

[0042] Building upon this foundation, data quality is further optimized by fusing and eliminating the initially aggregated vector plots before obtaining the final plot vector map. Specifically, multiple initially generated vector plots and their corresponding land use type labels are read, and all plots with the same label and adjacent boundaries are identified to form a plot set. This identification process accurately locates fragmented patches of the same land use type caused by minor deviations in classification boundaries or image noise. All vector plots in each plot set are merged into a continuous and single vector plot, and the land use type labels common to the original sets are assigned to the merged new plot. This fusion operation (e.g., using the "fusion" or "aggregation" tools in GIS) effectively restores the true and complete outline of the land use, eliminates redundant internal boundaries, and significantly improves the visual representation and topological quality of the data. Subsequently, the geometric area of ​​all merged vector plots in the real geographic coordinate system is obtained, and vector plots with a geometric area smaller than a preset area threshold (this threshold is set according to the planning map scale and minimum map area specifications) are eliminated. This filtering operation removes classification noise or overly fragmented patches that have no practical statistical significance.

[0043] Finally, by integrating all the remaining vector plots after fusion and area filtering, an optimized plot vector map with clear structure, accurate attributes, and complete geometric representation is obtained. This series of post-processing operations significantly improves the practicality and reliability of the original classification and aggregation results, enabling them to more accurately serve subsequent area calculations and traffic parameter correlations, thereby ensuring the scientific validity and accuracy of the final traffic generation prediction results.

[0044] Furthermore, the traffic generation prediction method also includes filtering the traffic attributes of the aggregated vector plots before merging the plot sets, including: Obtain traffic parameters for each land use type in the target planning area, and identify land use types in the target planning area that do not have traffic attributes; Read the multiple vector plots and their corresponding land use type labels generated after reclassification and aggregation, and compare each of the read land use type labels with land use types that do not have transportation attributes; Based on the comparison results, vector plots that do not have traffic attributes were identified and then removed.

[0045] By implementing the above-described traffic generation prediction method, traffic parameters for each land use type in the target planning area are obtained. Based on these parameters (usually derived from traffic engineering specifications or local traffic surveys), land use types in the target planning area that do not possess traffic generation and attraction attributes, such as green spaces, water bodies, or plazas, are clearly defined. This step establishes an objective and accurate discrimination standard for subsequent screening operations, ensuring that the basis for data processing is sufficient and in line with industry practices.

[0046] By reading multiple vector plots and their corresponding land use type labels generated after reclassification and aggregation, and comparing each land use type label with the aforementioned set of land use types that do not have traffic attributes, this process achieves automated and batch identification of the traffic function attributes of massive plots, completely replacing the inefficient mode of manual judgment in traditional methods, and significantly improving processing speed and consistency.

[0047] Based on the comparison results, all vector plots without traffic attributes were accurately identified and removed from the current dataset. This operation cleaned up the dataset before data fusion and area calculation, effectively preventing plots such as green spaces and water bodies that do not contribute to traffic from being incorrectly included in the subsequent traffic parameter association and area accumulation process. This ensured the purity and accuracy of the traffic generation calculation results at the source, avoided interference from irrelevant data to the prediction model, and improved the efficiency of the overall analysis process and the reliability of the prediction conclusions.

[0048] Furthermore, the traffic generation prediction method also includes calculating the future traffic generation volume of the target traffic zone by combining the traffic parameters and geometric area of ​​each vector plot in the target traffic zone, including: The target planning area is divided into several traffic zones, and a unique zone number is defined for each traffic zone to obtain a traffic zone division layer; Spatial overlay analysis was performed between the land parcel vector map and the traffic zone division layer to determine the traffic zone in which each vector land parcel in the land parcel vector map is located, and a corresponding zone number was added to each vector land parcel. Select all vector plots with the same neighborhood number as the target traffic zone to obtain the internal plot set of the target traffic zone, and obtain the travel rate and occurrence ratio of all vector plots in the internal plot set during peak hours; Based on the peak hour travel rate and occurrence ratio of the vector plot, and combined with the geometric area of ​​the vector plot in the real geographic coordinate system, the future traffic generation volume of the target traffic zone is calculated. The function expression for calculating the traffic generation volume is:

[0049] In the formula, This represents the future traffic generation volume of the i-th traffic zone. Indicates the traffic zone number. This represents the geometric area of ​​the k-th vector plot. This represents the travel rate of the k-th vector plot during peak hours. This represents the proportion of the k-th vector plot occurring during peak hours. Indicates the index of the vector plot. This represents the total number of vector plots in the traffic zone.

[0050] Furthermore, the traffic generation prediction method also includes calculating the future traffic attraction of the target traffic zone by combining the traffic parameters and geometric area of ​​each vector plot in the target traffic zone, including: Obtain the attraction ratio of all vector plots in the internal plot set of the target traffic zone during peak hours; Based on the peak-hour travel rate and attraction ratio of the vector plot, and combined with the geometric area of ​​the vector plot in the real geographic coordinate system, the future traffic attraction of the target traffic zone is calculated. The functional expression for calculating the traffic attraction is:

[0051] In the formula, This represents the future traffic attraction of the i-th traffic zone. Indicates the traffic zone number. This represents the attraction ratio of the k-th vector plot during peak hours.

[0052] By implementing the above-described traffic generation and prediction method, the target planning area is divided into several traffic zones, and a unique zone number is defined for each traffic zone, resulting in a traffic zone division layer. This step establishes a clear regional framework for subsequent traffic statistics and analysis based on spatial units.

[0053] Spatial overlay analysis of the land parcel vector map and the traffic zone division layer determines the traffic zone where each vector land parcel in the land parcel vector map is located, and adds a corresponding zone number to each vector land parcel. This operation accurately establishes the attribution relationship between land use spatial entities (land parcels) and traffic analysis and statistical units (zones), which is the key to ensuring that traffic volume can be correctly collected into the corresponding spatial units.

[0054] All vector plots with the same neighborhood number as the target traffic zone are selected to obtain the internal plot set of the target traffic zone. The trip rate and occurrence ratio of all vector plots in this set during peak hours are then obtained. This process extracts core parameters directly related to traffic generation from the purified plot data, preparing the data for quantitative calculations. Based on the trip rate and occurrence ratio of the vector plots during peak hours, combined with their geometric area calculated in the real geographic coordinate system, the future traffic generation volume of the target traffic zone is calculated. This calculation model considers the area scale of the plots. Land use intensity attributes and travel direction characteristics The organic combination of these three elements enables scientific and quantitative prediction of traffic generation from micro-level land parcel attributes to macro-level community traffic volume.

[0055] Similarly, by obtaining the attraction ratio of all vector plots in the internal plot set during peak hours and calculating the future traffic attraction of the target traffic zone, this process fully depicts the spatial distribution of traffic demand and ultimately realizes the paired prediction of traffic generation and traffic attraction for each traffic zone, providing an accurate travel generation matrix data foundation for the subsequent traffic distribution and allocation stages.

[0056] Furthermore, the traffic generation prediction method also includes calculating and obtaining the future traffic generation and traffic attraction volumes of the target traffic zone, and then verifying and iteratively optimizing the prediction results, including: The traffic generation and traffic attraction data obtained from each traffic zone are sorted by zone number and a trip generation matrix is ​​constructed. Based on the trip generation matrix and the road network topology of the target planning area, the predicted traffic flow of each road segment in the road network of the target planning area is obtained through traffic flow assignment calculation; Obtain the current traffic flow of each road segment within the target planning area, and compare the deviation between the predicted traffic flow and the current traffic flow. If the deviation value does not meet the preset deviation threshold, the deviation is determined to be caused by inaccurate traffic parameters or inaccurate land use type classification. If the deviation is determined to be due to inaccurate traffic parameters, the traffic parameters of each vector plot in the target traffic zone are reacquired until the deviation between the predicted traffic flow and the current traffic flow meets the preset deviation threshold. If the deviation is due to inaccurate land use classification, the training sample set is reconstructed until the deviation between the predicted traffic flow and the current traffic flow meets the preset deviation threshold.

[0057] By implementing the above-described traffic generation prediction method, the traffic generation and traffic attraction data obtained from each traffic zone are sorted according to the zone number and a travel generation matrix is ​​constructed. This operation integrates the discrete zone prediction results into a complete matrix representing the spatial distribution of traffic demand in the entire planning area, providing standardized input data for subsequent traffic network simulation.

[0058] Based on the travel generation matrix and the road network topology of the target planning area, the predicted traffic flow of each road segment in the target planning area is obtained through traffic flow allocation calculation, such as by using a user equilibrium allocation method. This step transforms the macroscopic traffic demand into the microscopic road network carrying capacity, realizing the physical mapping from "generation volume" to "observable road segment flow", so that the prediction results have a form that is directly comparable to the actual traffic conditions.

[0059] The system obtains the current traffic flow of each road segment within the target planning area and compares it with the calculated predicted traffic flow to obtain the deviation value. This comparison process enables an objective and quantitative assessment of the model's prediction accuracy and is a key step in determining the reliability of the prediction results. If the deviation value does not meet the preset deviation threshold, it further determines whether the deviation originates from inaccurate traffic parameters or inaccurate land use classification. This diagnostic mechanism can pinpoint the main source of error and provide direction for subsequent precise optimization.

[0060] If the judgment deviation stems from inaccurate traffic parameters, the traffic parameters (such as trip rate, occurrence ratio, and attraction ratio) of each vector plot in the target traffic zone are reacquired, and recalculated based on the new parameters until the deviation between the predicted traffic flow and the current traffic flow meets the preset deviation threshold. This feedback loop achieves dynamic calibration of the core input parameters, effectively absorbing errors caused by changes in traffic behavior or inaccurate initial parameter estimates. If the judgment deviation stems from inaccurate land use classification, the process returns to the step of reconstructing the training sample set by supplementing or correcting samples, then updates the image classification model and re-executes the subsequent entire process until the predicted traffic flow deviation meets the preset threshold. This mechanism ensures error correction and optimization from the source of image recognition, fundamentally improving the final prediction result by enhancing the accuracy of land use information extraction. This entire set of verification and iterative optimization processes constitutes a closed-loop system with self-diagnosis and correction capabilities, significantly improving the adaptability of the entire traffic generation prediction method to complex real-world situations and the reliability and accuracy of the final output results.

[0061] It should be understood that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Those skilled in the art can modify the technical solutions described in the above embodiments, or make equivalent substitutions for some of the technical features; and all such modifications and substitutions should fall within the protection scope of the appended claims of the present invention.

Claims

1. A traffic generation and prediction method based on GIS and image vectorization, characterized in that, The traffic generation prediction method includes: A color raster image of the target planning area is acquired, and georegistration processing is performed on the color raster image to generate a calibrated image that maps the image pixel coordinate system to the real geographic coordinate system. A training sample set is constructed by selecting multiple color regions corresponding to useful land type labels from the calibrated image; An image classification model trained on the training sample set is established based on GIS. The image classification model is used to reclassify the calibrated image, and pixels with the same land use type label and spatially adjacent pixels in the classification results are aggregated into independent vector plots to obtain a plot vector map. Obtain the geometric area of ​​each vector plot in the real geographic coordinate system from the plot vector map, and collect all the vector plots belonging to the target traffic zone from the plot vector map; Obtain the traffic parameters of each vector plot in the target traffic zone, and calculate the future traffic generation and traffic attraction of the target traffic zone by combining the geometric area of ​​each vector plot in the target traffic zone.

2. The traffic generation and prediction method based on GIS and image vectorization according to claim 1, characterized in that, The georegistration process of the color raster image to generate a calibrated image that maps the image pixel coordinate system to the real geographic coordinate system includes: The color raster image is cropped at the boundary to obtain a registration image focused on the target planning region; Multiple pixels with known real geographic coordinates are selected in the image to be registered, and coordinate transformation parameters are obtained according to the mapping relationship between the image pixel coordinates and the known real geographic coordinates of each selected pixel in the image to be registered. The coordinate transformation parameters are used to convert the image pixel coordinates of all pixels in the image to be registered into real geographic coordinates to obtain a preliminary calibration image. The geolocation accuracy of each pixel in the image to be registered is verified until the verification result meets the preset accuracy condition, and the calibrated image is obtained.

3. The traffic generation and prediction method based on GIS and image vectorization according to claim 1, characterized in that, The step of selecting multiple color regions corresponding to useful land type labels from the calibrated image to construct a training sample set includes: In the calibrated image, color regions representing different land use types are identified and delineated. According to the land use type represented by different colors, add a corresponding and unique land use type label to each of the color areas; Select a portion of the color region from the calibrated image, and extract color feature data from the selected color region; The extracted color feature data is associated with the land use type label corresponding to the color region to obtain training samples; All the training samples are collected and output as a structured data file for recognition and reading by the GIS platform, thus obtaining the training sample set.

4. The traffic generation and prediction method based on GIS and image vectorization according to claim 3, characterized in that, The establishment of the image classification model based on GIS and trained on the training sample set includes: The training sample set is read from the GIS platform, and the color feature data of each training sample is extracted. The color feature data is quantized into a feature vector, and the land use type label corresponding to the color region is used as a classification supervision signal for the feature vector. All the feature vectors and the corresponding classification supervision signals are taken as input, and a preset supervised classification algorithm is called to perform the model training process to generate an initial classification model; Cross-validation is used to optimize and adjust the parameters of the initial classification model to obtain the trained image classification model.

5. The traffic generation and prediction method based on GIS and image vectorization according to claim 1, characterized in that, The step of reclassifying the calibrated image using the image classification model and aggregating spatially adjacent pixels with the same land use type label in the classification results into independent vector plots to obtain a plot vector map includes: Each pixel of the calibrated image is input into the image classification model for reclassification, and the land use type label of each pixel is output. In the reclassified and calibrated image, all pixels with the same land use type label and spatially adjacent pixels are aggregated into independent polygonal vector plots, and the shared land use type label is assigned to the corresponding vector plot. All vector plots with the aforementioned land use type label are integrated into a single vector map layer, and the resulting vector map of the plots is output.

6. The traffic generation and prediction method based on GIS and image vectorization according to claim 1, characterized in that, The traffic generation prediction method further includes, before obtaining the land parcel vector map, performing fusion and elimination processing on the aggregated multiple vector land parcels, including: Read the multiple vector land parcels and their corresponding land use type labels generated after reclassification and aggregation, and identify all vector land parcels with the same land use type label and adjacent boundaries to obtain a land parcel set; All vector plots in each set of plots are merged into a continuous and single vector plot, and the land use type label common to the set of plots is assigned to the corresponding merged vector plot; Obtain the geometric area of ​​all the merged vector plots in the real geographic coordinate system, and remove the vector plots with a geometric area less than a preset area threshold. Then, integrate all the remaining vector plots to obtain the plot vector map.

7. The traffic generation and prediction method based on GIS and image vectorization according to claim 6, characterized in that, The traffic generation prediction method further includes filtering the traffic attributes of the aggregated vector land parcels before merging the land parcel set, including: Obtain traffic parameters for each land use type in the target planning area, and identify land use types in the target planning area that do not have traffic attributes; Read the multiple vector plots and corresponding land use type labels generated after reclassification and aggregation, and compare the read land use type labels one by one with land use types that do not have traffic attributes; Based on the comparison results, the vector land parcels that do not possess traffic attributes are identified and then removed.

8. The traffic generation and prediction method based on GIS and image vectorization according to claim 1, characterized in that, The traffic generation prediction method further includes calculating the future traffic generation volume of the target traffic zone by combining the traffic parameters and geometric area of ​​each vector plot in the target traffic zone, including: The target planning area is divided into several traffic zones, and a unique zone number is defined for each traffic zone to obtain a traffic zone division layer; Spatial overlay analysis is performed between the land parcel vector map and the traffic zone division layer to determine the traffic zone in which each vector land parcel in the land parcel vector map is located, and a corresponding zone number is added to each vector land parcel; Select all the vector plots that have the same community number as the target traffic community to obtain the internal plot set of the target traffic community, and obtain the travel rate and occurrence ratio of all the vector plots in the internal plot set during peak hours; Based on the travel rate and occurrence ratio of the vector plot during peak hours, and combined with the geometric area of ​​the vector plot in the real geographic coordinate system, the future traffic generation volume of the target traffic zone is calculated. The function expression for calculating the traffic generation volume is as follows: In the formula, This represents the future traffic generation volume of the i-th traffic zone. Indicates the traffic zone number. This represents the geometric area of ​​the k-th vector plot. This represents the travel rate of the k-th vector plot during peak hours. This represents the proportion of the k-th vector plot occurring during peak hours. Indicates the index of the vector plot. This represents the total number of vector plots in the traffic zone.

9. The traffic generation and prediction method based on GIS and image vectorization according to claim 8, characterized in that, The traffic generation prediction method further includes calculating the future traffic attraction of the target traffic zone by combining the traffic parameters and geometric area of ​​each vector plot in the target traffic zone, including: Obtain the attraction ratio of all vector plots in the internal plot set of the target traffic zone during peak hours; Based on the peak-hour travel rate and attraction ratio of the vector plot, and combined with the geometric area of ​​the vector plot in the real geographic coordinate system, the future traffic attraction of the target traffic zone is calculated. The functional expression for calculating the traffic attraction is as follows: In the formula, This represents the future traffic attraction of the i-th traffic zone. Indicates the traffic zone number. This represents the attraction ratio of the k-th vector plot during peak hours.

10. The traffic generation and prediction method based on GIS and image vectorization according to claim 1, characterized in that, The traffic generation prediction method further includes calculating and obtaining the future traffic generation and traffic attraction of the target traffic zone, and then verifying and iteratively optimizing the prediction results, including: The traffic generation and traffic attraction data obtained from each traffic zone are sorted according to the zone number and a travel generation matrix is ​​constructed. Based on the trip generation matrix and the road network topology of the target planning area, the predicted traffic flow of each road segment in the road network of the target planning area is obtained through traffic flow allocation calculation; Obtain the current traffic flow of each road segment within the target planning area, and compare the deviation between the predicted traffic flow and the current traffic flow. If the deviation value does not meet the preset deviation threshold, it is determined that the deviation originates from inaccurate traffic parameters or inaccurate land use type classification. If the deviation is determined to be due to inaccurate traffic parameters, the traffic parameters of each vector plot in the target traffic zone are reacquired until the deviation between the predicted traffic flow and the current traffic flow meets the preset deviation threshold. If the deviation is determined to be due to inaccurate land use type classification, the training sample set is reconstructed until the deviation between the predicted traffic flow and the current traffic flow meets the preset deviation threshold.

Images