Construction monitoring method and system based on image recognition technology

Abstract

The application relates to the technical field of construction monitoring, in particular to a construction monitoring method and system based on image recognition technology, which comprises the following steps: acquiring construction videos of each construction area in a construction area cluster through multiple video monitoring devices, wherein the construction videos comprise at least one frame of construction image; performing color statistics on the construction image to obtain a color distribution matrix; determining a monitoring coefficient of the construction image according to the color distribution matrix, performing monitoring screening on the construction video according to the monitoring coefficient to obtain at least one frame of target construction image; and performing target detection on the target construction image according to a pre-trained target detection model. Through color statistics on the image in the construction video, the color distribution matrix is obtained, the monitoring coefficient is calculated, automatic screening of the construction area is realized, and it is ensured that only relevant construction images are monitored, so that unnecessary monitoring content is avoided, and the overall monitoring efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of construction monitoring technology, specifically to a construction monitoring method and system based on image recognition technology. Background Technology

[0002] Traditional methods often employ fixed monitoring strategies, processing all construction videos identically without intelligent filtering based on content. This results in a large amount of irrelevant monitoring data, increasing the monitoring burden and management costs, and reducing monitoring efficiency. Furthermore, traditional methods may rely on manual inspection or simple image processing techniques during video analysis, lacking precise target identification and analysis. This leads to inaccurate identification of construction elements, impacting on-site management and safety, and hindering the timely identification of potential safety issues or construction deviations, potentially causing delays in problem detection and resolution. Additionally, traditional methods often lack scientific optimization of the spatial layout of construction areas, with monitoring equipment deployment typically based on experience. This can lead to blind spots or over-monitored areas. The spatial distribution density and boundary information of construction areas are often not accurately extracted, resulting in inaccurate division of construction areas and affecting the rational allocation of resources and optimization of construction progress. Moreover, traditional methods are usually static and cannot be automatically adjusted according to actual changes on the construction site. This means that the system's monitoring strategy may no longer be applicable during construction and cannot respond to changes in construction activities in a timely manner, leading to a decline in monitoring effectiveness. Furthermore, traditional methods lack dynamic analysis based on historical data and spatial characteristics, making it impossible to provide timely warnings of potential construction problems or safety hazards. Due to insufficient analysis of the construction status, traditional methods are difficult to detect problems in the early stages and may miss the best intervention opportunity. Summary of the Invention

[0003] To achieve the above objectives, the present invention provides the following technical solution: a construction monitoring method based on image recognition technology, comprising: Construction videos of each construction area in the construction area cluster are acquired through multiple video surveillance devices, wherein each construction video includes at least one frame of construction image. Color statistics are performed on the construction images to obtain a color distribution matrix; monitoring coefficients of the construction images are determined based on the color distribution matrix; and the construction video is monitored and filtered based on the monitoring coefficients to obtain at least one target construction image. The target construction image is subjected to target detection based on a pre-trained target detection model to obtain at least one frame of construction element labels; wherein, the construction element labels include personnel location, equipment status and material stacking status; Based on the location information of the video surveillance equipment, determine the spatial feature information of the construction area corresponding to the construction area cluster; based on the spatial feature information of the construction area and the construction element tags, determine the area-element distribution map of the construction area cluster corresponding to each historical time node; Based on the region-element distribution map, the monitoring and early warning information of the construction area cluster is determined.

[0004] Preferably, determining the monitoring coefficient of the construction image based on the color distribution matrix includes: Based on the color distribution matrix, the percentage of pixels for each color within a preset color gamut is calculated to obtain a percentage sequence. The number of pixels within a preset color gamut interval corresponding to each color in the color distribution matrix is ​​taken as the number of colors; The maximum number of colors in the color distribution matrix corresponding to the construction images is counted to obtain the first monitoring indicator; the maximum number of colors in the color distribution matrix corresponding to all construction images in the construction video is counted to obtain the second monitoring indicator. The monitoring coefficient of the construction image is determined based on the ratio of the first monitoring indicator to the second monitoring indicator.

[0005] Preferably, the target construction image is subjected to target detection based on a pre-trained target detection model to obtain at least one frame of construction element labels, including: Each frame of the target construction image is input into the pre-trained target detection model for target detection. After the target detection of each frame of the target construction image is completed, the corresponding construction element label is output according to the detection result. If the target detection model detects a target, it stops detecting the target construction image in the remaining frames; If no target is detected in any frame of target construction images, then each frame of target construction images will be divided into multiple image regions with priority. Feature extraction is performed on each image region in each frame of the target construction image in order of priority from high to low. After feature extraction of each image region is completed, target detection is performed based on the feature extraction results. If the feature extraction detects a target, then feature extraction for the remaining image region is stopped; If no target is detected in any image region, target detection is performed based on the image region of the target construction image in the next frame.

[0006] Preferably, each frame of the target construction image is divided into multiple image regions with priority, including: The image region where the target was most recently detected is obtained as the central image region. In the target construction image, the central image region and the image region of its surrounding first preset range are used to obtain the first priority image region. Determine the remaining image regions in the target construction image other than the highest priority image region, and take the image regions within a second preset range surrounding the first priority image region in the remaining image regions as the second priority image regions; Specific image regions are determined from each frame of target construction images based on the preset monitoring range.

[0007] Preferably, feature extraction is performed on each image region in each frame of the target construction image in descending order of priority. After feature extraction of each image region is completed, target detection is performed based on the feature extraction results, including: The image regions in each frame of the target construction image are sorted from high to low priority to obtain the image region sequence; According to the image region sequence, feature extraction is performed on each current image region to obtain a region feature map of the current image region; wherein, the feature extraction of the current image region to obtain a region feature map of the current image region includes: The current image region is convolved by a pre-trained multi-scale convolutional network to obtain multi-scale initial features. The multi-scale initial features are fused to obtain a fused initial feature map; The fused initial feature map is subjected to channel attention enhancement processing to highlight key channel features, resulting in a region feature map of the current image area. After feature extraction of each current image region is completed, target detection is immediately performed based on the region feature map of the current image region to obtain the target detection result of the current image region; wherein, the target detection based on the region feature map of the current image region to obtain the target detection result of the current image region includes: Target classification and location regression are performed on the region feature map of the current image region to obtain the target detection result of the current image region; The target detection result of the current image region is used as the input data for feature extraction of the next image region. Feature extraction and target detection are then performed on the next image region until all image regions have been processed.

[0008] Preferably, based on the location information of the video surveillance equipment, the spatial characteristic information of the construction area corresponding to the construction area cluster is determined, including: Spatial clustering analysis is performed on the location information of the video surveillance equipment. A clustering threshold is set according to the Euclidean distance between the video surveillance equipment, and adjacent video surveillance equipment are classified into the same construction area to obtain the initial construction area division result. Based on the initial construction area division results, the geometric center coordinates of each construction area are determined, and the representative points of the construction area are obtained. Based on the geometric center coordinates, a boundary expansion analysis is performed on each construction area. Taking a representative point as the center, the area is expanded outward according to a preset radius. The distribution density of video surveillance equipment in the expanded area is statistically analyzed. The expansion radius is adjusted until the equipment distribution density is stable, and the optimized construction area boundary is obtained. Extract the location information of all video surveillance devices within the optimized construction area boundary to obtain a spatial point set of the construction area; perform spatial interpolation processing on the spatial point set of the construction area to obtain a spatial distribution density map of the construction area.

[0009] Preferably, determining the spatial characteristic information of the construction area corresponding to the construction area cluster based on the location information of the video surveillance equipment further includes: The spatial distribution density map of the construction area is analyzed using a connected component analysis algorithm to identify connected components as hotspots for construction activities, thus obtaining a set of construction hotspots. Calculate the geometric features of the set of construction hotspot areas to obtain basic spatial feature information; wherein, the basic spatial feature information includes area, perimeter, and aspect ratio; Based on the aforementioned basic spatial characteristic information, the spatial clustering degree of construction activity hotspot areas is calculated using spatial autocorrelation analysis to obtain spatial clustering characteristics; Based on the spatial clustering characteristics and basic spatial characteristics of the construction activity hotspots, multi-scale analysis is performed to identify the spatial distribution patterns of construction areas at different scales, thereby obtaining the spatial characteristics of the construction area clusters.

[0010] Preferably, based on the spatial feature information of the construction area and the construction element tags, a region-element distribution map corresponding to each historical time node of the construction area cluster is determined, including: A spatial map of the construction area is determined based on the spatial feature information of the construction area; wherein, the spatial map of the construction area includes construction area nodes and edges, and the node attributes of the construction area nodes include the area coordinates and boundary information of the corresponding construction area; Based on the regional coordinates and boundary information of the construction area, the construction element labels are mapped to the spatial map of the construction area to obtain a region-element distribution map.

[0011] Preferably, the monitoring and early warning information for the construction area cluster is determined based on the region-element distribution map, including: Analyze the status of construction elements in each construction area at each historical time point in the region-element distribution map; Based on the preset area-element standard table, determine whether the status of construction elements in each construction area meets the early warning conditions; The early warning information corresponding to the status of construction elements that meet the early warning conditions will be used as the monitoring and early warning information for the construction area cluster.

[0012] The construction monitoring system based on image recognition technology, applicable to the aforementioned construction monitoring method based on image recognition technology, includes: A construction monitoring unit is used to acquire construction videos of each construction area in a construction area cluster through multiple video monitoring devices, wherein the construction video includes at least one frame of construction image. The monitoring and filtering unit is used to perform color statistics on the construction images to obtain a color distribution matrix; determine the monitoring coefficient of the construction images based on the color distribution matrix; and perform monitoring and filtering on the construction video based on the monitoring coefficient to obtain at least one frame of target construction image. The target detection unit is used to perform target detection on the target construction image according to the pre-trained target detection model to obtain at least one frame of construction element labels; wherein, the construction element labels include personnel location, equipment status and material stacking status; The distribution determination unit is used to determine the spatial feature information of the construction area corresponding to the construction area cluster based on the location information of the video surveillance equipment; and to determine the area-element distribution map of the construction area cluster corresponding to each historical time node based on the spatial feature information of the construction area and the construction element tags. The monitoring and early warning unit is used to determine the monitoring and early warning information of the construction area cluster based on the area-element distribution map.

[0013] Compared with the prior art, the beneficial effects of the present invention are: (1) This invention achieves automatic screening of construction areas by performing color statistics on images in construction videos, obtaining a color distribution matrix, and calculating monitoring coefficients. This ensures that only relevant construction images are monitored, avoiding unnecessary monitoring content and improving overall monitoring efficiency. Furthermore, the invention uses a target detection model to accurately analyze construction images, identify construction elements, and provide real-time monitoring feedback. This not only improves on-site safety and management efficiency but also enables the timely detection of potential construction problems. (2) This invention optimizes the spatial layout of the construction area through spatial clustering analysis and spatial feature extraction, providing accurate boundary and spatial distribution density information of the construction area, which helps to achieve more scientific division and optimized management of the construction area; and through construction element tags and spatial feature information of the construction area, it can generate a region-element distribution map, analyze the construction status of historical time nodes based on this map, and thus provide early warning of potential problems. This predictive ability can significantly improve construction safety and avoid accidents. (3) This invention can identify hot spots in construction activities and study the spatial distribution patterns of these areas through multi-scale analysis methods. This can help managers understand the actual situation of construction at different scales, facilitate the rational allocation of resources, and optimize the construction process. Moreover, based on the dynamic changes of different time nodes and spatial characteristics, the system can automatically adapt to changes in the construction site and adjust the monitoring strategy. This adaptability enhances the intelligence and flexibility of the monitoring system, enabling it to maintain efficient monitoring at different construction stages. Attached Figure Description

[0014] Figure 1 This is a schematic flowchart of the overall method in one embodiment of the present invention; Figure 2 This is a schematic diagram of the overall system architecture in one embodiment of the present invention.

[0015] In the diagram: 1. Construction monitoring unit; 2. Monitoring screening unit; 3. Target detection unit; 4. Distribution determination unit; 5. Monitoring and early warning unit. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Example 1, please refer to Figure 1 This invention provides a technical solution: a construction monitoring method based on image recognition technology, comprising: S1. Obtain construction videos of each construction area in the construction area cluster through multiple video surveillance devices, wherein the construction video includes at least one frame of construction image; S2. Perform color statistics on the construction images to obtain a color distribution matrix; determine the monitoring coefficients of the construction images based on the color distribution matrix; filter the construction videos based on the monitoring coefficients to obtain at least one target construction image. S3. Perform target detection on the target construction image based on the pre-trained target detection model to obtain at least one frame of construction element labels; among which, the construction element labels include personnel location, equipment status and material stacking status; S4. Based on the location information of the video surveillance equipment, determine the spatial characteristic information of the construction area cluster; based on the spatial characteristic information of the construction area and the construction element tags, determine the area-element distribution map of the construction area cluster corresponding to each historical time node. S5. Determine monitoring and early warning information for construction area clusters based on the regional-element distribution map.

[0018] It should be noted that construction videos from various construction areas within the construction area cluster are acquired through multiple video surveillance devices; each video contains at least one frame of construction image. The goal of this stage is to collect real-time video data from the construction site to provide raw material for subsequent analysis. The video surveillance devices provide visual data by capturing images of the construction area in real time, with monitoring points covering different areas of the construction site. Color statistics are performed on each frame of construction image to obtain a color distribution matrix. The color distribution matrix identifies the main color regions in the image by analyzing the color features. Then, a monitoring coefficient is calculated based on the color distribution matrix. This coefficient is used to measure whether the construction image has the construction features of interest. The construction videos are then filtered and extracted based on this monitoring coefficient. At least one frame of the target construction image is generated; the color distribution matrix can reflect the characteristics of different areas of the construction site, such as the color of different materials, the status of equipment, and the clothing of personnel; by calculating the monitoring coefficient, irrelevant images can be filtered out, and those containing key construction elements can be retained; the target construction image is analyzed by a pre-trained target detection model to extract construction element labels; these labels include: personnel location: identifying the specific location of personnel in the image and determining whether workers are working in the designated area; equipment status: checking the operating status of equipment, whether it is in working or idle state; material stacking status: checking the stacking of construction materials to ensure compliance with safety regulations and construction procedures; target detection models (such as YOLO, Faster) are used. Deep learning technologies such as R-CNN are used to automatically identify targets in construction images and generate labels and locations for each target. Based on the location information of video surveillance equipment, the spatial characteristics of each construction area cluster are determined. This information helps to understand the geographical location, size, and structure of each construction area. Then, by combining the spatial characteristics of the construction area with the construction element labels (labels obtained from the target detection step), a region-element distribution map for each historical time point is determined. The spatial characteristics of the construction area may include the location of the monitoring equipment, the boundaries and layout of the construction area, etc. The region-element distribution map shows the location and distribution of different elements (such as personnel, equipment, and materials) in each construction area at different time points. Based on the generated region-element distribution map and the monitoring status of the construction area, the system can automatically determine whether there are potential safety hazards or construction problems, and then generate monitoring and early warning information. This early warning information helps construction managers respond to potential problems in a timely manner and avoid accidents.

[0019] In an optional embodiment, determining the monitoring coefficients of the construction image based on the color distribution matrix includes: The percentage of pixels for each color within a preset color gamut is calculated based on the color distribution matrix to obtain a percentage sequence. The number of pixels in the preset color gamut interval corresponding to each color in the color distribution matrix is ​​taken as the number of colors; The first monitoring indicator is obtained by counting the maximum number of colors in the color distribution matrix corresponding to the construction images; the second monitoring indicator is obtained by counting the maximum number of colors in the color distribution matrix corresponding to all construction images in the construction video. The monitoring coefficient of the construction images is determined based on the ratio of the first monitoring indicator to the second monitoring indicator.

[0020] It's important to note that a color distribution matrix is ​​a matrix that reflects the color information in an image. Each row or column of the matrix represents the distribution of different colors (or color gamut intervals) in the image. A preset color gamut interval is usually a predefined range of colors, such as the red, green, and blue intervals in the RGB color gamut. These intervals are used to classify the color information in the image. The proportion of each color in the image is calculated by counting the number of pixels of each color within the preset color gamut interval. For example, if there are 500 red pixels in an image with a total of 5000 pixels, then the proportion of red pixels is 10%. Based on the image's color distribution matrix, the proportion of each color in the preset color gamut interval is calculated. The process involves comparing the proportions of different colors within the image, resulting in a sequence where each element represents the percentage of a particular color in the entire image. The number of pixels (i.e., the number of pixels) of each color in the image is then counted. For example, in an image, one color might occupy 300 pixels, while another might occupy 500 pixels. The number of colors is then compared to a preset color gamut: the number of colors is based on a preset color gamut, meaning only the number of pixels within certain predefined color ranges is considered. For each color in the color distribution matrix, the number of pixels within the preset color gamut is counted to obtain the total number of each color. Finally, the maximum number of colors in the color distribution matrix within a single construction image is calculated. The maximum value represents the richest number of color types appearing in the image. For example, if a construction image contains 5, 6, or 7 colors, the maximum value is 7. This value reflects the color complexity of the image; a higher number of colors in an image indicates richer and more complex image information, potentially corresponding to a more complex construction site. The maximum number of colors in the color distribution matrix is ​​calculated across all images in the entire construction video. Here, the maximum number of color types across all images is evaluated as a whole. This value reflects the maximum color complexity of the entire construction video and can help identify whether color usage exists at the construction site. Excessive or abnormal changes; a monitoring coefficient is obtained by calculating the ratio of the first monitoring indicator to the second monitoring indicator; the ratio reflects the relative relationship between the color complexity in a single construction image and the color complexity in the entire construction video; if the number of colors in a single image is higher than the maximum number of colors in the entire video, it may mean that the image has special monitoring value and may require additional attention; based on the monitoring coefficient, construction images can be filtered; images with higher monitoring coefficients may be important monitoring targets because they represent abnormal color changes at the construction site or the special nature of the construction situation; in this way, the construction monitoring system can automatically filter out images that need to be monitored in detail.

[0021] In an optional embodiment, target detection is performed on the target construction image based on a pre-trained target detection model to obtain at least one frame of construction element labels, including: Each frame of the target construction image is input into the pre-trained target detection model for target detection. After the target detection of each frame of the target construction image is completed, the corresponding construction element label is output according to the detection result. If the target detection model detects a target, it stops detecting the target construction image in the remaining frames; If no target is detected in any frame of target construction images, then each frame of target construction images will be divided into multiple image regions with priority. Feature extraction is performed on each image region in each frame of the target construction image in order of priority from high to low. After feature extraction of each image region is completed, target detection is performed based on the feature extraction results. If the feature extraction detects a target, then feature extraction for the remaining image region is stopped; If no target is detected in any image region, target detection is performed based on the image region of the target construction image in the next frame.

[0022] In an optional embodiment, each frame of the target construction image is divided into multiple image regions with priority, including: The image region where the target was most recently detected is obtained as the central image region. The central image region and the image region of its surrounding first preset range are then used to obtain the first priority image region in the target construction image. Identify the remaining image regions in the target construction image, excluding the highest priority image region, and designate the image regions within a second preset range surrounding the first priority image region in the remaining image regions as the second priority image regions; Specific image regions are determined from each frame of target construction images based on the preset monitoring range.

[0023] It should be noted that object detection refers to identifying target objects (such as equipment or workers in construction) from image or video frames. Each object detection provides the target's location, category, and other relevant information. In the most recent object detection, the image region where the target is located is determined, usually represented by a rectangular or polygonal bounding box. This region is the currently interested region, also known as the "center image region." The result of object detection is the coordinates and size of this center region. The first preset range is usually centered on the target region, extending outwards by a certain distance (e.g., 5 pixels, 10 pixels, etc.), or it can be a fixed-size rectangular or circular region. The first priority image region is the region obtained by extending outwards from the center image region by the first preset range; this region is considered to be the image region that needs priority attention. This is usually because the content within this region is most relevant to the target and therefore given the highest priority. Other regions besides the first priority region: In a target construction image, the remaining regions besides the first priority image region are the second priority region and its surrounding areas. That is, first determine a maximum priority region... The first priority area is selected by excluding the first priority area, leaving a second priority image area. The second preset range is similar to the first preset range, but it's a new range, typically larger or smaller than the first. For example, the first range might be 5 pixels, while the second range might be 10 pixels. The second priority image area is the remaining area outside the first priority area, specifically the portion closest to the first priority area (i.e., the second preset range surrounding the first priority area). This area is considered the second priority area, representing a relatively important but less urgent or critical image area than the first priority area. The preset monitoring range is a pre-defined area range, possibly set based on specific needs during construction. For example, certain equipment areas or critical construction locations may have higher monitoring requirements. Based on the targets in each frame and the set monitoring range, the image areas that need to be monitored are determined. These areas typically include higher priority areas (such as first and second priority areas) and some areas with special monitoring needs (such as high-risk areas).

[0024] In an optional embodiment, feature extraction is performed on each image region in each frame of the target construction image in descending order of priority. After feature extraction of each image region is completed, target detection is performed based on the feature extraction results, including: The image regions in each frame of the target construction image are sorted from high to low priority to obtain the image region sequence; Based on the image region sequence, feature extraction is performed on each current image region sequentially to obtain the region feature map of the current image region; wherein, feature extraction of the current image region to obtain the region feature map of the current image region includes: The current image region is convolved by a pre-trained multi-scale convolutional network to obtain multi-scale initial features. Multi-scale initial features are fused to obtain a fused initial feature map; Channel attention enhancement is applied to the fused initial feature map to highlight key channel features, resulting in a region feature map of the current image area. After each feature extraction of the current image region is completed, target detection is immediately performed based on the region feature map of the current image region to obtain the target detection result of the current image region; wherein, the target detection based on the region feature map of the current image region to obtain the target detection result of the current image region includes: Target classification and location regression are performed on the region feature map of the current image region to obtain the target detection result of the current image region; The target detection result of the current image region is used as the input data for feature extraction of the next image region. Feature extraction and target detection are then performed on the next image region until all image regions have been processed.

[0025] It should be noted that in object detection, an image is typically segmented into multiple regions, each of which may contain a target or has the potential to contain a target. These regions are prioritized according to certain predefined rules (e.g., target size, proximity to the central region, target importance, etc.). The image region sequence is a list of image regions, ranked from highest to lowest priority. After sorting, the image regions are sequentially subjected to feature extraction and object detection. In object detection, feature extraction refers to extracting key information from the image that helps identify the target. Each image region is processed individually, and the features of each region are extracted for subsequent object detection. Each image region generates a feature map, which is a comprehensive representation of all information within that region, and may include texture... Features such as texture, shape, color, and edges are extracted; multi-scale networks are used to extract features from different scales (i.e., different viewpoints or sizes); targets in an image may have different sizes and shapes, so it is necessary to capture more detailed information through multiple scales; by processing image regions through multi-layer convolutional neural networks (CNNs), the convolutional network automatically extracts multi-level features from the image; each convolutional layer learns the details of the image from different scale viewpoints, forming multi-scale initial features; multi-scale initial features usually contain information from different scales, which are merged through feature fusion techniques to obtain a comprehensive feature map; this fusion method can weight or merge information from different scales, enabling the network to simultaneously capture details in the image and Global information; the initial feature map after fusion is a feature representation obtained through feature fusion, which contains multi-scale information and can better describe the target features in the image region; in convolutional networks, different channels of the image (different dimensions of the feature map) contain different information; through the attention mechanism, more attention can be paid to important channel features to highlight key information in the image region; this step automatically enhances the network's attention to key channel features through the "channel attention" mechanism; after channel attention enhancement, the final region feature map will highlight the key features in the image, making target detection more accurate; target detection determines whether the image region contains a target based on the region feature map, and if a target is found, determines the target's location and category. Detection is based on region feature maps: After each feature extraction, the extracted region feature maps are used for target classification (e.g., determining whether it is construction equipment, workers, etc.) and location regression (determining the target's bounding box). Target classification is the process of classifying targets in an image region; for example, determining whether the current region contains construction equipment, workers, or other objects. Regression accurately locates the target by predicting its bounding box position (coordinate values); the regression result is a bounding box enclosing the target's position. Finally, the results of target classification and location regression provide the target category and location information detected in the current image region. The target detection results of the current image region are used as input data for feature extraction and target detection in the next image region.In this way, all image regions are processed one by one; each image region undergoes feature extraction and object detection sequentially according to the above steps, until all image regions have been processed, ensuring that every region in the image has been detected and analyzed.

[0026] In an optional embodiment, the spatial characteristic information of the construction area of ​​the corresponding construction area cluster is determined based on the location information of the video surveillance equipment, including: Spatial clustering analysis is performed on the location information of video surveillance equipment. Clustering thresholds are set according to the Euclidean distance between video surveillance equipment to classify adjacent video surveillance equipment into the same construction area, thus obtaining the initial construction area division results. Based on the initial construction area division results, the geometric center coordinates of each construction area are determined, and the representative points of the construction area are obtained; Based on the geometric center coordinates, a boundary expansion analysis is performed on each construction area. Taking a representative point as the center, the area is expanded outward according to a preset radius. The distribution density of video surveillance equipment in the expanded area is statistically analyzed. The expansion radius is adjusted until the equipment distribution density is stable, and the optimized construction area boundary is obtained. The location information of all video surveillance devices within the optimized construction area boundary is extracted to obtain the spatial point set of the construction area; spatial interpolation processing is performed on the spatial point set of the construction area to obtain the spatial distribution density map of the construction area.

[0027] It should be noted that the location of each video surveillance device can be represented by two-dimensional coordinates (such as longitude and latitude). Using this location information, we can classify the devices in space. Euclidean distance is a common method for measuring the straight-line distance between two points. After setting a clustering threshold, devices whose distance falls within the threshold range can be considered as a cluster. That is, when the Euclidean distance between adjacent devices is less than the set threshold, they are grouped into the same construction area. In this way, all adjacent devices are divided into a preliminary construction area. Through this clustering process, we obtain a preliminary construction area division map, with devices within each area close to each other. For each construction area, we can calculate its geometric center, usually called the "centroid" or "gravimeter." The geometric center is a representative point of the area, usually calculated by a weighted average of the locations of all devices within the area. This geometric center coordinate becomes the representative point of the construction area and can serve as the basis for further analysis and expansion. Using the geometric center as the center, we can expand outwards by a radius, forming a circular or other shaped expansion area. The initial expansion radius is determined based on a preset value. Within the extended area, the distribution density of video surveillance equipment is statistically analyzed; that is, the number of devices or the uniformity of their distribution within each extended area. The device density may change as the extension radius increases or decreases. The extension radius is adjusted until the device distribution density stabilizes, resulting in an optimized construction area boundary. Through this process, we obtain an optimized construction area boundary that conforms to the device distribution; this boundary more accurately reflects the actual distribution of the construction area. The location information of all video surveillance devices within the optimized construction area boundary is extracted to form a "spatial point set." This point set contains the locations of all devices within the area, providing data support for subsequent spatial analysis. Spatial interpolation is a method used to estimate data at unknown locations; here, it is used to infer and estimate the device distribution density at other locations within the construction area based on existing device location data. Through interpolation, we can generate a smoother and more coherent spatial distribution map of the construction area. The interpolated data is visualized as a spatial distribution density map, intuitively showing the distribution of video surveillance equipment in different areas and helping us understand the device density of the construction area.

[0028] In an optional embodiment, determining the spatial characteristic information of the construction area corresponding to the construction area cluster based on the location information of the video surveillance equipment further includes: The spatial distribution density map of the construction area is analyzed based on the connected component analysis algorithm to identify connected components as hotspots for construction activities, thus obtaining a set of construction hotspots. Calculate the geometric features of the construction hotspot area set to obtain the foundation spatial feature information; the foundation spatial feature information includes area, perimeter and aspect ratio; Based on the basic spatial characteristics information, the spatial clustering degree of the hot spots of construction activities is calculated by the spatial autocorrelation analysis method to obtain the spatial clustering characteristics; Multi-scale analysis is conducted based on the spatial clustering characteristics and basic spatial characteristics of construction activity hotspots to identify spatial distribution patterns of construction areas at different scales, thereby obtaining spatial characteristic information of construction area clusters.

[0029] It should be noted that connected component analysis is used to analyze the spatial distribution density map of the construction area. Connected component analysis helps us identify which areas are connected and densely packed; these areas typically represent hotspots of construction activity. The connected component analysis results will provide all construction hotspots, which usually have high equipment density, indicating areas with frequent construction activity. For each construction hotspot, we can calculate some geometric features, including: area: the total area of ​​the hotspot, representing the size of the area; perimeter: the boundary length of the hotspot; aspect ratio: the length-to-width ratio of the hotspot, used to understand the shape of the area, whether it is close to a square or a rectangle, etc. Based on the basic spatial feature information (such as area, perimeter, and aspect ratio), spatial autocorrelation analysis is used to calculate the spatial clustering degree of construction activity hotspots; spatial... Autocorrelation analysis, by detecting patterns in spatial data, can measure whether there is a clustering trend in equipment distribution in hotspot areas. Spatial clustering characteristics can help us understand whether construction activities tend to cluster in certain areas. Areas with high spatial clustering may be places with frequent construction activities and dense equipment, while those with low spatial clustering indicate more dispersed construction activities. Analyzing at different scales can identify the spatial distribution patterns of construction areas. For example, some areas may appear as hotspots at a small scale but may no longer be obvious at a large scale. Multi-scale analysis helps us identify the characteristics of construction areas at different scales. By analyzing the spatial distribution at multiple scales, we can identify the spatial distribution patterns of construction areas at different scales. These patterns can provide information about the spatial distribution characteristics of construction activities, such as the clustering and distribution trends of construction areas.

[0030] In an optional embodiment, based on the spatial feature information of the construction area and the labels of construction elements, a region-element distribution map corresponding to each historical time node of the construction area cluster is determined, including: The spatial map of the construction area is determined based on the spatial characteristics of the construction area; the spatial map of the construction area includes construction area nodes and edges, and the node attributes of the construction area nodes include the area coordinates and boundary information of the corresponding construction area. Based on the regional coordinates and boundary information of the construction area, the construction element labels are mapped to the spatial map of the construction area to obtain the region-element distribution map.

[0031] It should be noted that a construction area spatial map is a method for visualizing and structuring the spatial relationships within a construction area. It includes: Nodes: Nodes in the map represent individual areas within the construction area. Each node's attributes include: Area coordinates: the geometric center of each construction area (e.g., centroid coordinates), representing the area's location; Boundary information: representing the shape and size of the construction area's boundaries, usually presented as polygons, containing all external boundary points to ensure accurate depiction of the area's geometry; Edges: representing the spatial connections or relationships between construction areas. For example, if two construction areas are adjacent or overlap, an edge can represent their relationship. Edges can have different attributes, such as distance and connection strength, depending on the type of relationship between the construction areas. In short, the construction area spatial map treats each construction area as a node, representing the spatial relationships between different areas through connections between nodes. This allows us to use graph theory tools to analyze and process the spatial relationships between construction areas. Construction element labels refer to the various elements involved in the construction process... Various factors or elements, such as different building structures, materials, equipment, construction stages, and personnel arrangements, are crucial to the management of the construction process and need to be closely linked to the spatial layout of the construction area. By mapping the coordinates and boundary information of the construction area with construction element labels, we can associate each construction element with a corresponding area node. For example, specific building materials, construction stages, or equipment may need to be labeled within a certain construction area, and the mapped map can accurately reflect the distribution of these elements. The area-element distribution map is a visual chart that shows how various construction elements are distributed within the construction area. Each node in the map not only contains area coordinates and boundary information but also labels the elements within that area. This distribution map provides construction managers with a clear view of the relationship between areas and elements, helping them optimize the construction process, avoid resource conflicts, rationally arrange the construction schedule, and ensure the coordination of various construction elements. For example, it can prevent different construction stages within the same construction area from interfering with each other or situations where certain equipment is incompatible with building materials.

[0032] In an optional embodiment, determining monitoring and early warning information for construction area clusters based on a region-feature distribution map includes: Analysis of the construction element status of each construction area at each historical time point in the regional-element distribution map; Based on the preset area-element standard table, determine whether the status of construction elements in each construction area meets the early warning conditions; The early warning information corresponding to the status of construction elements that meet the early warning conditions will be used as the monitoring and early warning information for the construction area cluster.

[0033] It should be noted that the Region-Element Standard Table is a predefined standard table used to compare the status of construction elements in a construction area, determining which statuses are normal and which may require vigilance. The standard table lists preset standards or threshold conditions corresponding to different construction areas and construction elements. For example, a construction element in a certain area might require "50% completion of rebar construction within one week" or "material supply should not lag by more than two days." Based on the Region-Element Standard Table, a set of early warning conditions can be defined to determine whether the current status of construction elements meets the standards. The status of construction elements at each time point in the construction area is compared with the preset standards in the standard table. If the status of a construction element in a certain area does not meet the standards (e.g., progress delays or resource shortages), it indicates a potential risk in that area, requiring an early warning. When the status of a construction element does not meet the preset standards, the system generates corresponding early warning information, which serves as a reminder of potential risks and prompts the construction team to take action. Construction managers or decision-makers need to take appropriate measures to make adjustments. On a construction site, there are usually multiple construction areas (such as multiple building blocks or multiple work surfaces). These areas may be divided into different clusters based on geographical location or construction tasks. In some cases, construction area clusters may share common risk factors, causing multiple areas to face the same early warning risks. For example, if multiple areas of a construction project rely on the same material supplier, insufficient material supply may affect multiple areas simultaneously. Within a construction area cluster, if multiple areas meet the early warning conditions, the system will collect these early warning information as monitoring and early warning information for the construction area cluster. In this way, managers can monitor the risks in the entire cluster simultaneously, not just a single area. For example, if all areas in a construction cluster experience equipment failure, the system will issue a cluster early warning, prompting construction managers to take comprehensive countermeasures, such as scheduling backup equipment or adjusting the construction schedule.

[0034] Example 2, please refer to Figure 2 This invention provides a technical solution: a construction monitoring system based on image recognition technology, applicable to the aforementioned construction monitoring method based on image recognition technology, comprising: Construction monitoring unit 1 is used to acquire construction videos of each construction area in the construction area cluster through multiple video monitoring devices, wherein the construction video includes at least one frame of construction image. The monitoring and filtering unit 2 is used to perform color statistics on construction images to obtain a color distribution matrix; determine the monitoring coefficient of the construction images based on the color distribution matrix; and perform monitoring and filtering on the construction videos based on the monitoring coefficient to obtain at least one target construction image. The target detection unit 3 is used to perform target detection on the target construction image according to the pre-trained target detection model to obtain at least one frame of construction element labels; wherein, the construction element labels include personnel location, equipment status and material stacking status; Distribution determination unit 4 is used to determine the spatial feature information of the construction area cluster corresponding to the construction area cluster based on the location information of the video surveillance equipment; and to determine the area-element distribution map of the construction area cluster corresponding to each historical time node based on the spatial feature information of the construction area and the construction element tags. Monitoring and early warning unit 5 is used to determine the monitoring and early warning information of construction area clusters based on the regional-element distribution map.

[0035] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the spirit of the present invention.

Claims

1. A construction monitoring method based on image recognition technology, characterized in that, include: Construction videos of each construction area in the construction area cluster are acquired through multiple video surveillance devices, wherein each construction video includes at least one frame of construction image. Color statistics are performed on the construction images to obtain a color distribution matrix; monitoring coefficients of the construction images are determined based on the color distribution matrix; and the construction video is monitored and filtered based on the monitoring coefficients to obtain at least one target construction image. The target construction image is subjected to target detection based on a pre-trained target detection model to obtain at least one frame of construction element labels; wherein, the construction element labels include personnel location, equipment status and material stacking status; Based on the location information of the video surveillance equipment, determine the spatial feature information of the construction area corresponding to the construction area cluster; based on the spatial feature information of the construction area and the construction element tags, determine the area-element distribution map of the construction area cluster corresponding to each historical time node; Based on the region-element distribution map, the monitoring and early warning information of the construction area cluster is determined; The process of determining the monitoring coefficients of the construction image based on the color distribution matrix includes: Based on the color distribution matrix, the percentage of pixels for each color within a preset color gamut is calculated to obtain a percentage sequence. The number of pixels within a preset color gamut interval corresponding to each color in the color distribution matrix is ​​taken as the number of colors; The maximum number of colors in the color distribution matrix corresponding to the construction images is counted to obtain the first monitoring indicator; the maximum number of colors in the color distribution matrix corresponding to all construction images in the construction video is counted to obtain the second monitoring indicator. The monitoring coefficient of the construction image is determined based on the ratio of the first monitoring indicator to the second monitoring indicator. Specifically, the target construction image is subjected to target detection based on a pre-trained target detection model to obtain at least one frame of construction element labels, including: Each frame of the target construction image is input into the pre-trained target detection model for target detection. After the target detection of each frame of the target construction image is completed, the corresponding construction element label is output according to the detection result. If the target detection model detects a target, it stops detecting the target construction image in the remaining frames; If no target is detected in any frame of target construction images, then each frame of target construction images will be divided into multiple image regions with priority. Feature extraction is performed on each image region in each frame of the target construction image in order of priority from high to low. After feature extraction of each image region is completed, target detection is performed based on the feature extraction results. If the feature extraction detects a target, then feature extraction for the remaining image region is stopped; If no target is detected in any image region, target detection is performed based on the image region of the target construction image in the next frame; Specifically, based on the spatial feature information of the construction area and the construction element labels, the region-element distribution map of the construction area cluster corresponding to each historical time node is determined, including: A spatial map of the construction area is determined based on the spatial feature information of the construction area; wherein, the spatial map of the construction area includes construction area nodes and edges, and the node attributes of the construction area nodes include the area coordinates and boundary information of the corresponding construction area; Based on the regional coordinates and boundary information of the construction area, the construction element labels are mapped to the spatial map of the construction area to obtain a region-element distribution map.

2. The construction monitoring method based on image recognition technology according to claim 1, characterized in that, Each frame of the target construction image is divided into multiple image regions with priority, including: The image region where the target was most recently detected is obtained as the central image region. In the target construction image, the central image region and the image region of its surrounding first preset range are used to obtain the first priority image region. Determine the remaining image regions in the target construction image other than the first priority image region, and take the image regions within a second preset range surrounding the first priority image region in the remaining image regions as the second priority image regions; The image area to be monitored is determined from each frame of target construction images based on the preset monitoring range.

3. The construction monitoring method based on image recognition technology according to claim 2, characterized in that, Feature extraction is performed on each image region in each frame of the target construction image in descending order of priority. After feature extraction of each image region is completed, target detection is performed based on the feature extraction results, including: The image regions in each frame of the target construction image are sorted from high to low priority to obtain the image region sequence; According to the image region sequence, feature extraction is performed on each current image region to obtain a region feature map of the current image region; wherein, the feature extraction of the current image region to obtain a region feature map of the current image region includes: The current image region is convolved by a pre-trained multi-scale convolutional network to obtain multi-scale initial features. The multi-scale initial features are fused to obtain a fused initial feature map; The fused initial feature map is subjected to channel attention enhancement processing to highlight key channel features, thereby obtaining the region feature map of the current image region; After feature extraction of each current image region is completed, target detection is immediately performed based on the region feature map of the current image region to obtain the target detection result of the current image region; wherein, the target detection based on the region feature map of the current image region to obtain the target detection result of the current image region includes: Target classification and location regression are performed on the region feature map of the current image region to obtain the target detection result of the current image region; The target detection result of the current image region is used as the input data for feature extraction of the next image region. Feature extraction and target detection are then performed on the next image region until all image regions have been processed.

4. The construction monitoring method based on image recognition technology according to claim 3, characterized in that: Based on the location information of the video surveillance equipment, determine the spatial characteristic information of the construction area corresponding to the construction area cluster, including: Spatial clustering analysis is performed on the location information of the video surveillance equipment. A clustering threshold is set according to the Euclidean distance between the video surveillance equipment, and adjacent video surveillance equipment are classified into the same construction area to obtain the initial construction area division result. Based on the initial construction area division results, the geometric center coordinates of each construction area are determined, and the representative points of the construction area are obtained. Based on the geometric center coordinates, a boundary expansion analysis is performed on each construction area. Taking a representative point as the center, the area is expanded outward according to a preset radius. The distribution density of video surveillance equipment in the expanded area is statistically analyzed. The expansion radius is adjusted until the equipment distribution density is stable, and the optimized construction area boundary is obtained. Extract the location information of all video surveillance devices within the optimized construction area boundary to obtain a spatial point set of the construction area; perform spatial interpolation processing on the spatial point set of the construction area to obtain a spatial distribution density map of the construction area.

5. The construction monitoring method based on image recognition technology according to claim 4, characterized in that, Based on the location information of the video surveillance equipment, determining the spatial characteristic information of the construction area corresponding to the construction area cluster further includes: The spatial distribution density map of the construction area is analyzed based on the connected component analysis algorithm to identify connected components as hotspots for construction activities, thus obtaining a set of construction hotspots. Calculate the geometric features of the set of construction hotspot areas to obtain basic spatial feature information; wherein, the basic spatial feature information includes area, perimeter, and aspect ratio; Based on the aforementioned basic spatial characteristic information, the spatial clustering degree of construction activity hotspot areas is calculated using spatial autocorrelation analysis to obtain spatial clustering characteristics; Based on the spatial clustering characteristics and basic spatial characteristics of the construction activity hotspots, multi-scale analysis is performed to identify the spatial distribution patterns of construction areas at different scales, thereby obtaining the spatial characteristics of the construction area clusters.

6. The construction monitoring method based on image recognition technology according to claim 5, characterized in that, Based on the region-element distribution map, the monitoring and early warning information for the construction area cluster is determined, including: Analyze the status of construction elements in each construction area at each historical time point in the region-element distribution map; Based on the preset area-element standard table, determine whether the status of construction elements in each construction area meets the early warning conditions; The early warning information corresponding to the status of construction elements that meet the early warning conditions will be used as the monitoring and early warning information for the construction area cluster.

7. A construction monitoring system based on image recognition technology, used to implement the construction monitoring method based on image recognition technology as described in any one of claims 1-6, characterized in that, include: A construction monitoring unit is used to acquire construction videos of each construction area in a construction area cluster through multiple video monitoring devices, wherein the construction video includes at least one frame of construction image. The monitoring and filtering unit is used to perform color statistics on the construction images to obtain a color distribution matrix; determine the monitoring coefficient of the construction images based on the color distribution matrix; and perform monitoring and filtering on the construction video based on the monitoring coefficient to obtain at least one frame of target construction image. The target detection unit is used to perform target detection on the target construction image according to the pre-trained target detection model to obtain at least one frame of construction element labels; wherein, the construction element labels include personnel location, equipment status and material stacking status; The distribution determination unit is used to determine the spatial feature information of the construction area corresponding to the construction area cluster based on the location information of the video surveillance equipment; and to determine the area-element distribution map of the construction area cluster corresponding to each historical time node based on the spatial feature information of the construction area and the construction element tags. The monitoring and early warning unit is used to determine the monitoring and early warning information of the construction area cluster based on the area-element distribution map.

Images