Entrance guard community space problem assessment method

Abstract

The invention relates to an access control community space quality evaluation method and system based on multi-source image data, and the method comprises the steps: extracting residential AOI data from the AOI data of a city as a community boundary, and supplementing the community boundary data in combination with POI data and land utilization data; the method comprises the following steps: automatically acquiring community image data, position and related information data, and preprocessing the data; randomly selecting a certain number of cell image data as a data set, marking the image, training a target detection deep learning model by using the data set, and calculating the model performance; and applying a deep learning model in the cell image data to obtain a spatial quality evaluation result of each cell. Through a Yolo V8 deep learning model and in combination with community image data and street view data of housing resource platforms such as residential guests, deep analysis of access control community space quality is realized. The evaluation efficiency is improved, manpower and material resources are saved, and the unified standard and comparability of evaluation results can be better ensured. The whole process is more automatic and intelligent, the requirement for the professional level of an operator is lowered, and the system is easier to popularize.

Description

Technical Field

[0001] The present invention relates to the technical fields of deep learning image recognition and geographic information system, and in particular to a method and system for evaluating spatial problems in a gated community. Background Art

[0002] The traditional method of evaluating the spatial problems of gated communities is mainly based on large-scale questionnaires and interviews, and there is also a method of on-site scoring by professionals. Internationally, the American Housing Survey conducted by the U.S. Department of Housing and Urban Development, which has many years of evaluation experience, adopts a sampling survey method, collecting 100,000 samples at a time. The Chinese urban physical examination work launched in recent years uses on-site evaluation by professionals combined with interviews with community workers in the community (community) level physical examination. Both methods require a lot of manpower and material resources and are time-consuming.

[0003] Image recognition technology is a core technology in the field of computer vision and a practical application of deep learning algorithms. In recent years, models for urban image element recognition have been developed, and their recognition objects include street view images, remote sensing images, etc. Their recognition content includes spatial disorder elements, urban vacant land, green space, etc. However, the existing models do not use the internal image library of the community for targeted training, and the elements related to the community space problem indicators are not fully covered, and the recognition accuracy is not high.

[0004] The current methods used to evaluate the spatial quality of gated communities are inefficient, and alternative paths need to be explored. In addition, there is a lack of precedent for the collection of community image data, and it is necessary to explore multi-source acquisition methods and data processing methods for community spatial quality assessment. There is an urgent need to develop a highly efficient and accurate community quality assessment method that can cover all elements related to community quality indicators. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a method and system for evaluating the spatial quality of a gated community based on multi-source image data. By combining the Yolo V8 deep learning model with the community image data and street view data of housing source platforms such as Anjuke, an in-depth analysis of the spatial quality of the gated community is achieved.

[0006] In one aspect, the present invention provides a method for evaluating the spatial quality of a gated community based on multi-source image data, comprising the following steps:

[0007] S1 extracts residential AOI data from the city's AOI data as the community boundary, and supplements the community boundary data with POI data and land use data;

[0008] S2 automatically collects community image data, location and related information data, and pre-processes the data. The community image data comes from housing source platforms and street view data;

[0009] S3 randomly selects a certain number of cell image data as a data set, annotates the images, uses the data set to train a deep learning model for target detection, and calculates the model performance;

[0010] S4 applies a deep learning model to the cell image data to obtain the spatial quality assessment results for each cell.

[0011] In a preferred embodiment, step S1 comprises:

[0012] Extract the data of categories "residential area", "residential community" and "dormitory" in the AOI data;

[0013] Extract the data of the categories of "residential area", "residential community" and "dormitory" in the POI data;

[0014] Extract residential land data from land use data;

[0015] Combined with remote sensing images, the above data are deduplicated and integrated.

[0016] In a preferred embodiment, step S2 comprises:

[0017] Collect community video and image data;

[0018] De-duplicate and extract frames of video data to convert them into image data;

[0019] Convert the WGS84 coordinates corresponding to the cell into the BD09 coordinates in the Baidu coordinate system.

[0020] In a preferred implementation scheme, the community video and image data are derived from a housing platform that includes access control community video and image data, street view data, self-collected data, and a combination thereof, and the collected community-related information data includes the community's latitude and longitude, housing prices, rents, and whether there is an elevator.

[0021] Furthermore, the preferred housing source platform is Anjuke housing source platform.

[0022] Furthermore, preferably, collecting the community video and image data includes calling the JavaScript API of Baidu Map to obtain street view image data within and around the community.

[0023] In a preferred embodiment, step S3 comprises:

[0024] Randomly select some cell image data as the data set;

[0025] Develop a labeling manual based on the indicators of community space quality issues and the characteristics of the deep learning model;

[0026] Annotate the data according to the annotation manual;

[0027] The images in the training set are input into the deep learning model. After multiple deep separable convolutional layers, the training stops and a cell space quality assessment model is generated.

[0028] The model was applied to the test set and the performance of the deep learning model was evaluated using the mAP50 indicator.

[0029] In a preferred embodiment, step S4 comprises:

[0030] Applying a cell space quality assessment model to each image;

[0031] The result of each indicator of each image is a binary result, that is, whether the image has the quality problem, if yes, it is 1, otherwise it is 0;

[0032] The spatial quality problem index result of each image is the sum of the results of all indicators;

[0033] The spatial quality problem index result for each cell is the average of the results of all images.

[0034] In a preferred embodiment, the deep learning model is a Yolov8 target detection model.

[0035] On the other hand, the present invention also provides a system for evaluating the spatial quality of a gated community based on multi-source image data, comprising a module for executing any of the above steps.

[0036] Compared with the existing evaluation methods, the method and system for evaluating the spatial quality of a gated community based on multi-source image data of the present invention improves the evaluation efficiency, saves manpower and material resources, and better ensures the unified standard and comparability of the evaluation results. First, by using big data and automatic data acquisition methods, the data acquisition efficiency is improved, and the time and cost of questionnaire collection, interviews, and field investigations are saved. Secondly, a deep learning model for identifying spatial evaluation indicators is trained, which greatly improves the data processing speed and is suitable for quantitative evaluation. Compared with questionnaire interviews, the teaching cost is saved, and the feedback differences caused by individual differences of the interviewees are avoided; compared with on-site scoring, the judgment differences caused by the differences in the standards of different scorers are eliminated. Through the Yolo V8 deep learning model combined with the community image data and street view data of housing source platforms such as Anjuke, the spatial quality of the gated community is deeply analyzed, which can cover all the relevant elements of the community quality indicators and obtain a high-accuracy community space quality evaluation model. The whole process is more automated and intelligent, which reduces the requirements for the professional level of the operator and makes it easier to promote. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 The present invention is a flow chart of the space problem assessment of a gated community according to a preferred embodiment of the present invention.

[0038] Figure 2 A flowchart of cell boundary generation according to a preferred embodiment of the present invention.

[0039] Figure 3 A cell boundary map generated for a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0040] The present invention provides a method for evaluating the spatial quality of a gated community based on multi-source image data, comprising the following steps:

[0041] S1 Access control community boundary extraction. Residential AOI data is extracted from the city's AOI data as the community boundary, and the community boundary data is supplemented by combining POI data, land use data, etc.

[0042] S2 Cell image data collection and processing: Automatically collect cell image data and location and other related information data, and pre-process the data.

[0043] S3 algorithm development. Randomly select a certain number of cell image data as a data set and annotate the images. Use the data set to train a deep learning model for target detection and calculate the model performance.

[0044] S4 Cell spatial quality assessment: Apply deep learning models to cell image data to obtain the spatial quality assessment results of each cell.

[0045] The index system for space problems in a gated community described in the present invention preferably includes the following indicators:

[0046]

[0047] Furthermore, the specific method of step S1 is preferably:

[0048] S11: Obtain urban AOI data, POI data and land use data;

[0049] S12: Obtaining AOI data, extracting data whose categories are “residential area”, “residential community”, and “dormitory”;

[0050] S13: Obtain POI data and extract data of categories "residential area", "residential community" and "dormitory";

[0051] S14: For land use data, residential land data is extracted;

[0052] S15: Combine remote sensing images to deduplicate and integrate these data.

[0053] Furthermore, the specific method of step S2 is preferably:

[0054] S21: Collecting community video and image data from the housing platform;

[0055] S22: performing deduplication and frame extraction processing on the video data to convert it into image data;

[0056] S23: Convert the WGS84 coordinates corresponding to the cell into the BD09 coordinates in the Baidu coordinate system;

[0057] S24: Calling the JavaScript API of Baidu Map to obtain street view image data within the community and its surroundings;

[0058] S25: Collect video and image data in the cell and perform frame extraction processing;

[0059] S26: Collecting the community-related information data such as the community's longitude and latitude, housing prices, rents, whether there is an elevator, etc.

[0060] The housing listing platform may select any housing listing publishing platform including community videos and image data, preferably Anjuke, Beike, Fangtianxia, ​​Juepaifang, Lianjia and other housing listing platforms, and Anjuke is more preferred.

[0061] Furthermore, the specific method of step S3 is preferably:

[0062] S31: Randomly select some cell image data as a data set;

[0063] S32: Develop a labeling manual based on the indicators of the community space quality issues and the characteristics of the deep learning model to be adopted;

[0064] S33: Annotate the data according to the annotation manual;

[0065] S34: input the images in the training set into the deep learning model, and after multiple depth-separable convolutional layers, the training stops, and a cell space quality assessment model is generated;

[0066] S35: Apply the model to the test set and use the mAP50 indicator to evaluate the performance of the deep learning model.

[0067] Furthermore, the specific method of step S4 is preferably:

[0068] S41: Applying the cell spatial quality assessment model to each image

[0069] S42: The result of each indicator of each image is a binary result, that is, whether the image has the quality problem, if yes, it is 1, otherwise it is 0.

[0070] S43: The spatial quality problem index result of each image is the sum of the results of all indicators.

[0071] S44: The spatial quality problem index result of each cell is the average value of the results of all images.

[0072] Further, the preferred deep learning model is the Yolo v8 target detection model. While inheriting the advantages of the YOLO series, the Yolo v8 deep learning model introduces a new backbone network, Ancher-Free detection head and loss function to improve the performance and flexibility of the model. The new backbone network can extract image features more effectively, and the Ancher-Free detection head reduces the number of hyperparameters, making the model easier to train. In addition, the new loss function further optimizes the training process of the model and improves the detection accuracy. Therefore, it is very suitable for this technical solution. The indicator mAP50 is an important indicator for evaluating the performance of the YOLOv8 model. It represents the mean average precision when the IoU (intersection over union) threshold is 0.5. Specifically: IoU (Intersection over Union): an indicator that measures the degree of overlap between the predicted box and the true box. When IoU is greater than 0.5, the prediction is considered correct. AP (Average Precision): Calculate the precision (Precision) and recall (Recall) at different confidence thresholds, and then take the average. The higher the AP, the stronger the detection ability of the model. mAP: is the average of AP of all categories. In the case of mAP50, only the case with an IoU threshold of 0.5 is considered. The level of mAP50 directly reflects the accuracy of the model in the detection task and is usually used to compare the performance of different models or different training configurations. A high mAP50 means that the model performs well in identifying objects.

[0073] The technical solution of the present invention is clearly and completely described below in conjunction with specific embodiments. The embodiments given are only for better illustrating the present invention rather than for limiting the scope of the present invention.

[0074] Example 1

[0075] This embodiment provides a method for evaluating the spatial quality of a cell based on Anjuke image data, the method comprising the following steps:

[0076] S1 extracts three types of data, namely "residential area", "residential community" and "dormitory", from the city's AOI data as the community boundaries, and uses the three types of POI data and residential land data of "residential area", "residential community" and "dormitory" in combination with high-resolution remote sensing image data to supplement and deduplicate the community boundary data.

[0077] S2 automatically collects image data and location and other related information data from housing listing platforms such as Anjuke, and processes the data. First, the video data is deduplicated according to the standard of retaining only one video content of the same length and size, and the deduplicated video data is extracted as image data. The image data after frame extraction is tested for image similarity using indicators such as image color contrast and brightness, and only one image with a similarity of more than 90% is retained.

[0078] S3 algorithm development. Randomly select 20,000 cell image data as the data set, and annotate the images using rectangular boxes and try to fit the feature boundaries under the guidance of the annotation manual. Split the training set, test set, and validation set in a ratio of 8:1:1, and use the training set to train the target detection deep learning model and calculate the model performance.

[0079] S4 Cell spatial quality assessment: Apply deep learning models to cell image data to obtain the spatial quality assessment results of each cell.

[0080] Among them, the labeling platform for step S3 is labelimg software. In order to ensure the consistency of the audit standards for each indicator, each labeler is responsible for 4-5 indicators.

[0081] Different cells will have different numbers of coverage indicators due to different numbers of images. Generally speaking, cells with large data volumes have more comprehensive coverage indicators, while cells with small data volumes have fewer coverage indicators. Therefore, in step S4, cells with image data volumes less than 20 will be excluded to reduce the problem of incomplete coverage of indicators caused by too little data.

[0082] Example 2

[0083] This embodiment provides a method for evaluating the quality of a cell space based on street view image data. The method is basically the same as that of Embodiment 1, except that the specific method of step S2 is:

[0084] S21: Convert the WGS84 coordinates corresponding to the cell to the BD09 coordinates in the Baidu coordinate system;

[0085] S22: Obtain the roads within the cell and its surrounding areas and generate 50m sampling points.

[0086] S23: Calling the JavaScript API of Baidu Map;

[0087] S24: Obtain street view image data of roads within the community and its surrounding areas.

[0088] Among them, according to the positional relationship between the street view image and the community boundary, the street view image of the corresponding perspective is selected. When the street view point is completely within the community boundary, the street view images in four directions (0°, 90°, 180°, 270°) are extracted; when the street view point is completely outside the community boundary, the street view image data facing the community boundary is extracted.

[0089] Example 3

[0090] This embodiment provides a method for evaluating the spatial quality of a cell based on self-collected image data. The method is basically the same as that of Embodiment 1, except that the specific method of step S2 is:

[0091] S21: Equip a Gopro or other camera device on a bicycle / electric bicycle;

[0092] S22: Install track recording software such as “Two-Step Road” on the smartphone;

[0093] S23: The data collector drives a bicycle / electric bicycle to collect image data in the community;

[0094] S24: Match the cell image data and GPS track points, and extract image data at a sampling distance of 50m.

[0095] Among them, GoPro and other camera equipment are set to wide-angle mode, and the shooting direction is straight ahead. The community image data and GPS data will be matched according to the time tags of the two.

[0096] Data from different sources do not conflict with each other and can be used in combination if all data are available.

Claims

1. A method for evaluating the spatial quality of a gated community based on multi-source image data, comprising the following steps: S1 extracts residential AOI data from the city's AOI data as the community boundary, and supplements the community boundary data with POI data and land use data; S2 automatically collects community videos, image data, location and related information data, and pre-processes the data. The community image data sources include housing source platforms and street view data; S3 randomly selects a certain number of cell image data as a data set, annotates the images, uses the data set to train a deep learning model for target detection, and calculates the model performance; S4 applies a deep learning model to the cell image data to obtain the spatial quality assessment results for each cell.

2. The method for evaluating the spatial quality of a gated community according to claim 1, characterized in that: The S1 step includes: Extract the data of "residential area", "residential community" and "dormitory" in the AOI data; Extract the data of "residential area", "residential community" and "dormitory" in the POI data; Extract residential land data from land use data; Combined with remote sensing images, the above data are deduplicated and integrated.

3. The method for evaluating the spatial quality of a gated community according to claim 1, characterized in that S2 The steps include: Collect community video and image data; De-duplicate and extract frames of video data to convert them into image data; Convert the WGS84 coordinates corresponding to the cell into the BD09 coordinates in the Baidu coordinate system.

4. The method for evaluating the spatial quality of a gated community according to claim 1, characterized in that The community video and image data come from housing source platforms that include access control community video and image data, street view data, self-collected data and their combinations. The collected community-related information data includes the community’s latitude and longitude, housing prices, rents, and whether there are elevators.

5. The method for evaluating the spatial quality of a gated community according to claim 4 is characterized in that The housing listing platform is selected from Anjuke housing listing platform.

6. The method for evaluating the spatial quality of a gated community according to claim 4 is characterized in that Collecting community video and image data includes calling the JavaScript API of Baidu Map to obtain street view image data within and around the community.

7. The method for evaluating the spatial quality of a gated community according to claim 1, characterized in that The S3 steps include: Randomly select some cell image data as the data set; Develop a labeling manual based on the indicators of community space quality issues and the characteristics of the deep learning model; Annotate the data according to the annotation manual; The images in the training set are input into the deep learning model. After multiple deep separable convolutional layers, the training stops and a cell space quality assessment model is generated. The model was applied to the test set and the performance of the deep learning model was evaluated using the mAP50 indicator.

8. The method for evaluating the spatial quality of a gated community according to claim 1, characterized in that The S4 step includes: Applying a cell space quality assessment model to each image; The result of each indicator of each image is a binary result, that is, whether the image has the quality problem, if yes, it is 1, otherwise, it is 0; The spatial quality problem index result of each image is the sum of the results of all indicators; The spatial quality problem index result for each cell is the average of the results of all images.

9. The method for evaluating the spatial quality of a gated community according to any one of claims 1 to 8, characterized in that: The deep learning model is the Yolov8 target detection model.

10. A system for assessing the spatial quality of a gated community based on multi-source image data, comprising a module for executing the steps of any one of the methods of claims 1-9.

Images