CN-121982554-A - Urban micro-update achievement comparison method and device based on time sequence streetscape semantic analysis

CN121982554ACN 121982554 ACN121982554 ACN 121982554ACN-121982554-A

Abstract

The invention discloses a city micro-update achievement comparison method and device based on time sequence streetscape semantic analysis, wherein the method comprises the steps of effectively comparing sample images of streetscapes with variation amplitude lower than a standard threshold value in a sample set to serve as stable anchor point samples; comparing vegetation of stable anchor point samples in a time period to be calibrated and a reference time period, taking the calculated deviation value as a calibration coefficient, respectively calculating indexes of a plurality of comparison indexes for streetscapes after the streetscape images are calibrated through the calibration coefficient, forming an index polygon based on the indexes of each comparison index, and obtaining a comparison result based on the comparison of the areas of the index polygons. By adopting the technical scheme, the street view images are screened by utilizing the geometric invariance of the building in time sequence change, the calibration coefficients of the vegetation elements are calculated in an inversion mode, the noise caused by season replacement is eliminated, a nonlinear comprehensive measurement model based on a polygon graph index method is constructed, and the coordination and restriction relation of all the environmental elements is captured in a sharp mode.

Inventors

LIANG CHIHAO
LIANG HANWEI
ZHANG LONGCHENG
LI DEMIN
Xue Tianyue

Assignees

南京信息工程大学

Dates

Publication Date: 20260505
Application Date: 20260409

Claims (10)

1. A city micro-update effect comparison method based on time sequence street view semantic analysis is characterized by comprising the following steps: obtaining a plurality of historical street view images and updated street view images, filtering images which have excessive interference factors and cannot effectively represent streets, respectively matching semantic information in the historical street view images and the updated street view images by utilizing a feature extraction model, determining matched semantic information and corresponding historical street view images and updated street view images, respectively taking the screened historical street view images and updated street view images as historical street view samples and updated street view samples, and combining the historical street view images and updated street view images into an effective comparison sample set; The method comprises the steps of effectively comparing a sample set, inquiring a time period of sparse vegetation and exuberant vegetation and respectively serving as a time period to be calibrated and a reference time period, comparing the time period to be calibrated with the stable anchor sample in the reference time period, and taking a calculated deviation value as a calibration coefficient, wherein a sample image of a street view with a variation amplitude lower than a standard threshold value is taken as a stable anchor sample; After the micro-updated foreground image and the micro-updated background image are calibrated through the calibration coefficients, respectively calculating indexes of a plurality of comparison indexes for the micro-updated foreground and the micro-updated background; setting a central point, obtaining vertexes from the central point based on the index of each comparison index, connecting the vertexes to form an index polygon, and obtaining a comparison result based on the comparison of the areas of the index polygons of the micro-updated foreground and the micro-updated background.
2. The method for comparing urban micro-update success based on semantic analysis of time series streetscape according to claim 1, wherein said filtering out the images in which too many interference factors exist and streetscape cannot be effectively characterized comprises: In the historical street view image and the updated street view image, if the proportion of the number of pixels corresponding to vehicles and pedestrians to the number of pixels of the image is higher than a first limit proportion, the corresponding historical street view image or the updated street view image is filtered, and if the proportion of the number of pixels corresponding to sky and vegetation to the number of pixels of the image is lower than a second limit proportion, the corresponding historical street view image or the updated street view image is filtered.
3. The urban micro-update success-comparison method based on time-series street view semantic analysis according to claim 1, wherein the step of taking a sample image of a street view with a variation amplitude lower than a standard threshold as a stable anchor point sample comprises the steps of: and if the deviation between the building proportions of the historical street view sample and the updated street view sample is smaller than an ideal threshold, determining that the variation amplitude of the street view in the image is lower than the standard threshold, taking the street view in the image as an anchoring street view, and taking the sample image comprising the anchoring street view in the effective comparison sample set as a stable anchor point sample.
4. The urban micro-update success-comparison method based on time-series street view semantic analysis according to claim 3, wherein the period of inquiring vegetation sparsity and vegetation flourishing comprises: The method comprises the steps of calculating the proportion of the number of vegetation pixels to the number of all pixels of an image, taking a shooting time period of the image with the vegetation proportion lower than a sparse threshold value as a time period to be calibrated, and taking the shooting time period of the image with the vegetation proportion higher than a flourishing threshold value as a reference time period.
5. The urban micro-update success-comparison method based on time-series street view semantic analysis according to claim 3, wherein comparing the time period to be calibrated with the stable anchor point sample in the reference time period comprises: calculating the ratio of vegetation pixels to all pixels of the image as greening ratio for a sample image comprising the same anchoring street view in a stable anchor point sample, calculating the ratio of greening ratio between a reference time period and the stable anchor point sample in a time period to be calibrated as gain ratio, obtaining the median of gain ratio distribution, and taking the median as a calibration coefficient.
6. The urban micro-update success comparison method based on time sequence street view semantic analysis according to claim 5, wherein the calibrating the micro-updated foreground image and the micro-updated street view image by the calibration coefficient comprises: and if the micro-updated foreground image and the micro-updated background image are in the time zone to be calibrated, and the other shooting time is in the reference time zone, calibrating the image in the time zone to be calibrated through the calibration coefficient.
7. The urban micro-update success comparison method based on time sequence street view semantic analysis according to claim 6, wherein the calibrating the micro-updated foreground image and the micro-updated street view image by the calibration coefficient comprises: And calibrating the vegetation characteristic value based on the calibration coefficient for the image in the time period to be calibrated.
8. The urban micro-update achievement comparison method based on time-series street view semantic analysis according to claim 1, wherein the comparison indexes comprise at least three of dynamic target pixel density, static building interface pixel ratio, slow road geometry ratio, visual field non-motorized cleanliness, vegetation semantic pixel abundance, visual field openness, street canyon aspect ratio, scene visual semantic richness and street boundary flexibility degree.
9. The method for comparing urban micro-update success based on time-series street view semantic analysis according to claim 1, wherein said connecting a plurality of vertices to form an exponential polygon comprises: Carrying out normalization calculation on the index of each comparison index, setting a central point, and extending from the central point to obtain vertexes based on the index of the comparison index after normalization, connecting the vertexes to form an equilateral n-polygon and taking the equilateral n-polygon as an index polygon, wherein n represents the number of the comparison indexes in practical application; the comparing of the areas of the index polygons based on the micro-updated foreground and the micro-updated background to obtain a comparison result comprises the following steps: And constructing a unit circle with the radius of 1, calculating the area ratio between the index polygon and the unit circle, calculating the difference value of the area ratio between the micro-updated street view and the micro-updated street view, and obtaining a comparison result according to the positive and negative deviation degree of the calculated difference value.
10. A city micro-update effect comparison device based on time sequence streetscape semantic analysis is characterized by comprising a screening unit, a calibration coefficient calculation unit, an index calculation unit and a comparison unit, wherein: The screening unit is used for acquiring a plurality of historical street view images and updated street view images, filtering images which have excessive interference factors and cannot effectively represent the street view, respectively utilizing a feature extraction model to match semantic information in the historical street view images and the updated street view images, determining matched semantic information and corresponding historical street view images and updated street view images, respectively taking the screened historical street view images and updated street view images as historical street view samples and updated street view samples, and combining the historical street view images and the updated street view images into an effective comparison sample set; The calibration coefficient calculation unit is used for effectively comparing sample sets, taking a sample image of a street view with the variation amplitude lower than a standard threshold value as a stable anchor point sample, inquiring a time period with sparse vegetation and exuberant vegetation and respectively taking the time period to be calibrated and a reference time period as a time period to be calibrated; the index calculation unit is used for calculating indexes of a plurality of comparison indexes for the micro-updated foreground and the micro-updated background respectively after the micro-updated foreground and the micro-updated background are calibrated through the calibration coefficient; The comparing unit is used for setting a center point, obtaining vertexes based on the index extension of each comparison index from the center point, connecting the vertexes to form an index polygon, and obtaining a comparison result based on the comparison of the areas of the index polygons of the micro-updated foreground and the micro-updated background.

Description

Urban micro-update achievement comparison method and device based on time sequence streetscape semantic analysis Technical Field The invention relates to the technical field of computer vision and digital image processing, in particular to a city micro-update effect comparison method and device based on time sequence street view semantic analysis. Background As the global urbanization progresses from incremental expansion to stock upgrading, urban micro-updates have become the core path for improving the quality of the living environment. Unlike the development model of traditional large-scale construction, micro-updates emphasize activation of urban vitality by low-interference interventions such as the remediation of street interfaces, the placement of pocket parks, and the optimization of slow-going systems. However, this high frequency, small scale, decentralized update pattern presents a significant micro-update quality assessment challenge to the city manager. In recent years, street view images (STREET VIEW IMAGERY, SVI) are widely used as an emerging city perception data source for measuring the built environment of a city due to the characteristics of high resolution, wide coverage and human visual angle. Semantic information is extracted from massive streetscapes by using a deep learning technology, so that the method becomes a main paradigm of city science calculation. However, two unresolved key technical bottlenecks remain in handling long-time-series micro-update evaluation tasks. First, in urban micro-update comparisons, fluctuations in visual characteristics caused by natural alternations constitute a large systematic error. For example, if the reference image of the previous year is taken in summer and the updated acceptance image is taken in winter, direct comparison may lead to false conclusions about the reverse of greening, thereby misleading decisions. The prior solution is mostly attempted to perform macroscopic correction through satellite remote sensing data, but the cross-mode correction effect is poor due to geometrical differences of a satellite image overlooking view angle and a street view head-up view angle and dislocation of resolution scale. Second, in conventional urban quality comparison and assessment, a linear weighted sum model is often used to calculate the results. This logic results in that the various indicators in the urban environment can compensate each other. For example, extremely high greening levels can offset extremely poor walking space or safety hazards. However, the quality of urban public spaces often depends on the streets with short plates, green trees shadow but sidewalks are completely occupied by motor vehicles, the experience is very poor for pedestrians, the comprehensive quality should be judged to be low, the middle and high scores are obtained through non-linear weighting, and a traditional linear model cannot capture the non-linear constraint and coupling relation among the elements, so that the evaluation result is deviated from the actual perception of residents. Disclosure of Invention The invention aims to provide a city micro-update achievement comparison method and device based on time sequence street view semantic analysis, and aims to solve the technical problem that in city micro-update evaluation in the prior art, the accuracy of city micro-update comparison results is low due to vegetation change and linear weighting result calculation methods. The invention provides a city micro-update effect comparison method based on sequential street view semantic analysis, which comprises the steps of obtaining a plurality of historical street view images and update street view images, filtering images with excessive interference factors and incapable of effectively representing street views, utilizing a feature extraction model to respectively match semantic information in the historical street view images and the update street view images, determining matched semantic information and corresponding historical street view images and update street view images, combining the historical street view images and the update street view images obtained through screening as historical street view samples and update street view samples respectively to form an effective comparison sample set, wherein the effective comparison sample set takes sample images of streets with variation amplitude lower than a standard threshold value as stable anchor point samples, inquiring vegetation sparsity and vegetation flourishing time periods and respectively as a time period to be calibrated and a reference time period, comparing the vegetation values of the stable anchor point samples in the time period to be calibrated with the reference time period, taking calculated deviation values as calibration coefficients, calibrating the micro-update street view images and the corresponding historical street view images respectively as the historical street view samples, comp