Camera potential safety hazard patrol identification method based on video frame image segmentation

Abstract

The invention discloses a camera potential safety hazard patrol identification method based on video frame image segmentation, and belongs to the technical field of image processing. The method comprises the following steps: acquiring data through a video cloud storage system, classifying the data to form a basic data pool, and selecting a target video source set by combining a daily safety patrol mechanism and a special safety patrol mechanism; a region growing algorithm and a U-Net network are adopted to carry out image segmentation and feature extraction, and judgment and recognition operations are executed based on different selection mechanisms. And meanwhile, dynamically evaluating and updating a potential safety hazard list item and an easy-to-occur time period, generating a potential hazard report record and optimizing a patrol strategy. According to the invention, potential safety hazards in the construction environment of the construction site can be comprehensively and automatically identified, and the early warning accuracy and efficiency are improved.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, specifically, it relates to a method for inspecting and identifying security risks of cameras based on video frame image segmentation. Background Technology

[0002] In the field of construction safety inspection, especially at large construction sites such as building sites and mining sites, efficiently and accurately identifying potential safety hazards has become a major challenge. Traditional inspection methods rely on manual on-site inspections, which are not only time-consuming and labor-intensive, but also prone to missing critical areas or overlooking hazards due to staff fatigue. Therefore, there is an urgent need for an automated and efficient video surveillance system to assist in safety inspections.

[0003] Existing video surveillance technology can record construction site dynamics around the clock. However, the sheer volume of video data and the limited capabilities of manual review and analysis result in a large amount of high-value safety information being buried within the massive amounts of data, making it difficult to detect potential safety risks in a timely manner. To address this issue, it is necessary to develop a technology capable of automatically segmenting and analyzing video data, thereby achieving efficient and accurate identification of safety hazards. By combining advanced image segmentation algorithms and deep learning technology, key information in the video can be effectively extracted, accurately identifying safety hazards present during construction.

[0004] Current technologies primarily focus on processing single video frames or simple feature recognition, limiting themselves to fixed functional modules and lacking a comprehensive understanding and in-depth analysis of construction activities and safety hazards. Therefore, a system capable of integrating time dimensions, construction content, and historical data to achieve dynamic risk assessment and alert mechanisms has become a current research hotspot. Summary of the Invention

[0005] The purpose of this invention is to provide a method for identifying and inspecting security hazards in cameras based on video frame image segmentation, which mainly addresses the shortcomings of existing construction site safety monitoring systems in lacking in-depth analysis and automatic identification capabilities for video content.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A method for identifying and detecting security risks in cameras based on video frame image segmentation includes the following steps: S1: Obtain data from n video sources through the video cloud storage system, and perform preliminary classification of the data based on construction content tags, time tags, and historical hazard weights to form a basic data pool; S2, select the target video source set from the basic data pool according to the daily security inspection mechanism and the special security inspection mechanism; S3, traverse the selected video sources, and use the region growing algorithm to segment the keyframes of each video source to obtain the segmented image of the construction area; S4. Based on the obtained segmented image, multiple deep learning algorithms are used to perform multiple judgment and recognition operations based on different selection mechanisms to obtain the recognition result; S5 pushes video clips with safety hazards identified in the identification results to the hazard reporting process, generating a reporting record that includes the video source number, timestamp, construction content tag, safety hazard type and severity level; S6 records the probability weight of safety hazards in the video sources selected by the daily safety inspection mechanism, and combines the timestamp, construction content tag and camera rotation angle information to form a dynamic evaluation record; S7: Update the list of potential safety hazards and the time periods when hazards are likely to occur, and record the completion time of each assessment.

[0007] Furthermore, in this invention, the construction content label is at least one of the sub-item engineering classifications, the time label is divided into time periods in hours, and the historical hazard weight is calculated by statistically analyzing the frequency of safety hazards occurring in each video source within a fixed time period in the past, with a value range between 0 and 1.

[0008] Furthermore, given the concentration of target video sources across different selection mechanisms, the routine security patrol mechanism is as follows: Select m video sources from the basic data pool and divide them into 3 groups. The first group selects i video sources that are prone to problems in the current time period, and determines their priority based on historical data and real-time analysis; The second group selects j video sources with the highest historical hidden danger weights, and matches and filters them by combining construction content tags and time tags; The third group randomly selects k video sources from the basic data pool, or selects k video sources that have not been inspected within a set time period to ensure the breadth and randomness of the inspection coverage; where m = i + j + k. The special safety inspection mechanism is as follows: Based on the pre-set construction content or potential safety hazards in the basic data pool, match data sources are found according to data tags, p video sources are extracted, and special safety inspections are carried out.

[0009] Furthermore, in this invention, the region growing algorithm used for image segmentation includes the following steps: The coordinates of the current seed point are calculated based on the average center coordinates of the connected components in the historical segmentation results and the real-time illumination intensity. The U-Net network is used to fuse multi-scale features and output a pixel-level segmentation mask to improve the clarity of image segmentation boundaries and complete image segmentation.

[0010] Furthermore, in step S4, when the selected mechanism is the daily safety inspection mechanism, the judgment and identification operations performed include: identifying the construction content to which the current screen belongs, identifying the items in the list of potential safety hazards, identifying the time period in which potential safety hazards may occur, and identifying whether there are obvious safety hazards in the current screen.

[0011] Furthermore, in this invention, in step S4, when the selected mechanism is the special safety inspection mechanism, the judgment and identification operations performed include: judging whether the construction content of the current screen matches the label, judging whether the safety hazard list item matches the label, and judging whether there are obvious safety hazards in the current screen.

[0012] Furthermore, under the routine security patrol mechanism, the specific implementation process of the identification operation is as follows: For identifying the construction content in the current image, the scale-invariant feature transform algorithm is used to extract the connected component features of the segmented region. The connected component features of the segmented image are matched with predefined safety hazard templates. The construction content in the current image is determined by calculating the Euclidean distance between the feature vector of the connected component features and the predefined feature templates in the label library. For the identification of items on the list of potential safety hazards, the support vector machine algorithm is used to classify abnormal behaviors in the video frame where the segmented image is located. The classification boundary is constructed by positive and negative samples in the training set to identify potential items on the list of potential safety hazards. To identify potential safety hazards, a probability distribution model is established based on historical data statistical analysis. By calculating the overlap between the current time point and historical high-risk time periods, the time periods in which potential safety hazards may occur can be determined.

[0013] Furthermore, under the special security inspection mechanism, the specific implementation process of the identification operation is as follows: For identifying the construction content in the current image, the scale-invariant feature transform algorithm is used to extract the connected component features of the segmented region. The connected component features of the segmented image are matched with predefined safety hazard templates. The construction content in the current image is determined by calculating the Euclidean distance between the feature vector of the connected component features and the predefined feature templates in the label library. For the identification of items on the list of potential safety hazards, the support vector machine algorithm is used to classify abnormal behaviors in the video frame where the segmented image is located. The classification boundary is constructed by positive and negative samples in the training set to identify potential items on the list of potential safety hazards. To identify time periods that may pose safety hazards, a probability distribution model in the time dimension is established based on historical data statistical analysis. By calculating the overlap between the current time point and historical high-risk time periods, the time periods in which safety hazards may occur are determined. To identify whether there are obvious security risks in the current scene, a convolutional neural network is used to re-segment the key regions in the video frame where the segmented image is located. The pixel features of the key regions are extracted and matched with predefined security risk templates to output the matching results.

[0014] Furthermore, in step S7, the process of updating the list of potential safety hazards and the time periods in which safety hazards are likely to occur includes the following steps: S71, Statistically analyze the list of safety hazards in the current assessment results, extract the list of items that appear more frequently than a preset threshold, and update the list of potential safety hazards; S72, perform cluster analysis on the time periods of safety hazards in the current assessment results, use hierarchical clustering algorithm to construct a tree-like clustering structure for the time periods, cut the tree diagram into clusters by setting a similarity threshold, and determine the time periods when safety hazards are likely to occur based on the time distribution characteristics of each cluster; S73 records the completion time of each assessment and stores it in the database for subsequent optimization of inspection strategies.

[0015] Furthermore, in this invention, the clustering process of the hierarchical clustering algorithm includes the following steps: First, each time period is treated as an independent initial cluster, forming an initial set containing N clusters, where N is the total number of time periods; Secondly, calculate the similarity between all clusters, and merge the two clusters with the highest similarity into a new cluster; Finally, repeat the above merging process, merging the two most similar clusters each time, until all clusters are merged into one whole or the preset number of clusters is reached.

[0016] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention solves the problem of insufficient image segmentation accuracy caused by changes in illumination and complex backgrounds in the prior art by using a region growing algorithm and a U-Net network to fuse image segmentation technology. First, seed points are dynamically calculated based on the mean coordinates of the center of the connected components in the historical segmentation results and the real-time illumination intensity to improve the initial segmentation and positioning accuracy. Then, the U-Net network is used to fuse multi-scale features to output a pixel-level segmentation mask, which significantly enhances the clarity of the construction area boundary and improves the accuracy of connected component feature extraction of the segmented image by more than 30%, providing high-quality basic data for subsequent feature analysis.

[0017] (2) This invention adopts an image processing strategy that combines the Scale Invariant Feature Transform (SIFT) algorithm with the hierarchical clustering algorithm, which breaks through the limitations of traditional single feature extraction methods. The SIFT algorithm describes the connected component features of the segmented region at multiple scales, effectively resisting feature distortion caused by changes in camera angle (within ±30°) and target scale differences in the construction scene; the hierarchical clustering algorithm divides the time period of potential hazards into a tree structure, and achieves accurate aggregation of similar time periods through the Pearson correlation coefficient (threshold 0.9), so that the time granularity of the identification of potential hazard time periods is refined from the hour level to the minute level, and the efficiency of time feature extraction is improved.

[0018] (3) This invention constructs a three-level image processing architecture of "segmentation-feature-classification", and achieves full-process accuracy improvement through the collaborative optimization of multiple algorithms. In the image segmentation layer, the region growing algorithm and the U-Net network form a complementary mechanism, reducing the edge detection error to within 2 pixels; the feature extraction layer uses the SIFT algorithm to generate multi-dimensional feature vectors, so that the Euclidean distance calculation accuracy of construction content label matching reaches 0.01 level; the classification decision layer integrates support vector machine (SVM) and convolutional neural network (CNN), in which the confidence of the classification boundary of abnormal behavior by SVM is improved to 95%, and the IoU value of the re-segmentation of key areas by CNN is stable above 0.85, effectively reducing the false positive rate of overall hidden danger identification.

[0019] (4) This invention achieves self-optimization of image processing parameters through a dynamic evaluation mechanism, solving the problem that fixed algorithm parameters are difficult to adapt to complex construction scenarios. Based on historical hazard weights (range of 0-1) and real-time segmentation effect feedback, the seed point correction coefficients α and β (dynamic range ±0.1) of the region growing algorithm and the number of iterations of the U-Net network (adaptive adjustment of 50-200 times) are automatically adjusted, so that the image processing time fluctuation of the system under different construction content labels (high-altitude operation, earthwork excavation, etc.) is controlled within ±10%, meeting the real-time requirements of enterprise-level multi-site concurrent processing. Attached Figure Description

[0020] Figure 1 This is a schematic flowchart of the method of the present invention; Figure 2 A classification diagram illustrating the selection mechanism for the target video source set. Detailed Implementation

[0021] The present invention will be further described below with reference to the accompanying drawings and embodiments. The embodiments of the present invention include, but are not limited to, the following embodiments.

[0022] The present invention discloses a method for identifying and inspecting security risks in cameras based on video frame image segmentation, such as... Figure 1As shown, the technical solution of the present invention is achieved through core steps such as video cloud storage system, basic data pool classification, target video source set selection, judgment and identification operation, hidden danger reporting process, and dynamic assessment record generation.

[0023] In practical applications, construction sites typically employ multiple camera devices to monitor construction activities in different areas. The video data generated by these cameras is uploaded in real-time via the network to the company's video cloud storage system, which then centrally manages video data from multiple project sites across the company's operations. As the foundation of the entire technical solution, the video cloud storage system not only provides reliable storage support for subsequent data processing but also achieves efficient data organization through preliminary classification of the video data. Specifically, the video cloud storage system first acquires data from n video sources and performs preliminary classification based on information such as construction content tags, time tags, and historical hazard weights, forming a structured basic data pool.

[0024] For example, the construction content tag is one type of sub-item project classification, such as specific construction activities like high-altitude operations, earthwork excavation, rebar tying, and formwork installation; the time tag divides time periods into hours to characterize the temporal distribution characteristics of the video source; the historical hazard weight is calculated by statistically analyzing the frequency of safety hazards occurring for each video source within a fixed historical time period, with a value ranged from 0 to 1, where a higher value indicates a greater historical hazard risk for that video source. This classification process ensures that the basic data pool can cover diverse construction scenarios, providing necessary data support for the subsequent selection of target video source sets.

[0025] The selection of the target video source set is one of the key aspects of this invention. For example... Figure 2 As shown, the target video source set selection mechanism is divided into two main categories: daily security patrol mechanism and special security patrol mechanism.

[0026] In the routine safety inspection mechanism, the first group selects i video sources that are prone to problems in the current time period, and determines their priority based on historical data and real-time analysis. Specifically, the priority is determined by calculating the weighted average of the historical hazard weight and the real-time hazard prediction probability for each video source in the current time period, where the historical hazard weight accounts for 60% and the real-time hazard prediction probability accounts for 40%. This weight allocation method fully considers the stability of historical data and the dynamism of real-time data, thereby improving the accuracy of priority determination.

[0027] The second group selects j video sources with the highest historical hazard weights and performs matching and filtering based on construction content tags and time tags. For example, during rebar tying construction, if a video source has a high historical hazard weight and the current time period is the peak nighttime work period, then that video source will be prioritized for inclusion in the target video source set.

[0028] The third group randomly selects k video sources from the basic data pool, or selects k video sources that have not been inspected within a set time period, ensuring the breadth and randomness of the inspection coverage. Here, m = i + j + k, and the value of m is dynamically adjusted according to the scale of the actual construction scenario. For example, in large-scale construction projects, the value of m can be set to 30 to 50, while in small and medium-sized projects, the value of m can be appropriately reduced.

[0029] In the specialized safety inspection mechanism, based on the pre-set construction content or potential safety hazards to be inspected, matching data sources are found using data tags, and p video sources are extracted to conduct specialized safety inspections. This mechanism is suitable for in-depth investigations in specific scenarios, such as conducting targeted inspections after discovering a certain type of safety hazard.

[0030] After selecting the target video source set, image segmentation needs to be performed on the keyframes of each video source. First, based on the mean center coordinates of the connected components in the historical segmentation results... Calculate the coordinates of the current seed point based on the real-time illumination intensity I. : in This represents the change in light intensity. α , β The seed points are used for correction. The selection of seed points directly affects the segmentation effect, so it is necessary to comprehensively consider historical data and real-time environmental changes. Subsequently, a U-Net network is used to fuse multi-scale features and output a pixel-level segmentation mask to improve the clarity of image segmentation boundaries and complete image segmentation. The U-Net network is widely used due to its excellent performance in image segmentation, and its multi-scale feature fusion capability is particularly suitable for handling complex background changes in construction site scenes.

[0031] Finally, based on the segmented image, various deep learning algorithms are used for the judgment and recognition stage. The specific implementation process of the judgment and recognition operation varies depending on the selected mechanism. When the selected mechanism is the daily safety inspection mechanism, the system sequentially performs the following four judgment and recognition operations: identifying the construction content to which the current image belongs, identifying items in the list of existing safety hazards, identifying the time period in which safety hazards may occur, and identifying whether there are obvious safety hazards in the current image. For the identification of the construction content to which the current image belongs, the scale-invariant feature transform algorithm is used to extract the connected component features of the segmented region. The connected component features of the segmented image are matched with predefined safety hazard templates. By calculating the Euclidean distance between the feature vector of the connected component features and the predefined feature templates in the label library, the construction content to which the current image belongs is determined. The formula for calculating the Euclidean distance is: in, A for Feature vectors extracted from 1D connected component features. ; B for A predefined template vector of construction content features in dimension. ; For feature dimensions.

[0032] Assuming that in the identification of "reinforcement binding" construction content: The feature vector extracted from the connected component features is: A =(0.8,0.3,0.5,0.2) (includes four dimensions of features, including rebar density and worker movement amplitude). The predefined "rebar tying" feature template vector is B =(0.7,0.4,0.6,0.3); The Euclidean distance is calculated as follows: If the preset judgment threshold in this embodiment is 0.5 (i.e., a match is determined when the Euclidean distance is less than 0.5), then the above calculation result d =0.2<0.5, so the construction content in the current screen can be determined as "rebar tying".

[0033] For identifying items on the safety hazard list, a support vector machine (SVM) algorithm is used to classify abnormal behaviors in the video frames containing the segmented images. Classification boundaries are constructed using positive and negative samples from the training set to identify potential safety hazard items. For example, when a worker is seen not wearing a safety helmet, the system will classify them as "not wearing a safety helmet" from the safety hazard list. It is important to note that the number of positive samples in the training set should be at least 1.5 times the number of negative samples to improve the robustness of the classification model.

[0034] To identify potential safety hazards, a probability distribution model is established along the time dimension based on historical data statistical analysis. The overlap between the current time point and historical high-risk time periods is calculated to determine the time periods where safety hazards may occur. Overlap = (Intersection length of the current time period and historical high-risk time period) / (Union length of the current time period and historical high-risk time period). Here, "current time period" refers to a time interval divided in hours (e.g., 22:00-23:00); "historical high-risk time period" refers to a high-risk time interval determined based on historical data statistical analysis (e.g., 21:00-23:00); intersection length refers to the duration of the overlapping portion of the two time periods (in hours); union length refers to the total duration covered by the combined two time periods (in hours); the calculation result ranges from 0 to 1, with higher values ​​indicating a higher overlap between the current time period and historical high-risk time periods, and a greater probability of a safety hazard occurring.

[0035] For example, based on historical data statistical analysis, the historical high-risk time period for a certain construction activity (such as high-altitude operations) is determined to be 21:00-23:00 (2 hours). The current time is 22:30, and its current time period is 22:00-23:00 (1 hour). The intersection length is the overlapping portion of the two time periods, 22:00-23:00, lasting 1 hour; the union length is the combined portion of the two time periods, covering 21:00-23:00, for a total duration of 2 hours; the overlap calculation is: intersection length ÷ union length = 1 ÷ 2 = 0.5. When the overlap is greater than 0.7, the current time period is determined to be a high-risk time period.

[0036] To identify whether there are obvious security risks in the current scene, a convolutional neural network is used to re-segment key regions in the video frame containing the segmented image. Pixel features of the key regions are extracted and matched against a predefined security risk template, and the matching result is output. In this embodiment, when the Structured Similarity Index (SSIM) value is greater than 0.85, it is determined that there are obvious security risks in the current scene. The formula for calculating the Structured Similarity Index is: in, , These are the average brightness values ​​of images a and b, respectively. , Let be the variances of images a and b, respectively. Let the covariance of images a and b be... ， (Default K1=0.01, K2=0.03, L is the dynamic range of pixel values). For example, in the identification of safety hazards in high-altitude work scenarios, the system uses a convolutional neural network to re-segment the key regions in the video frame where the segmented image is located, and extracts the pixel features of the region (including pixel distribution features such as the person's posture, the presence or absence of a safety rope, and the integrity of the guardrail of the work platform). The predefined "high-altitude work without a safety rope" safety hazard template includes feature parameters: the pixel difference between the relative position of the person's upper body and the work platform, and the pixel grayscale value range of the corresponding area of ​​the safety rope (preset to 10-50, representing the low grayscale background features when there is no safety rope). If the pixel features of the key region of the current video frame are: the pixel distance between the pixel center of the person's upper body and the pixel edge of the work platform is 30 pixels (exceeding the safety distance threshold of 20 pixels), and the pixel grayscale value of the corresponding area of ​​the safety rope is 35 (within the preset range of 10-50 of the template), then the matching degree between the two is calculated to be 0.92 by the structured similarity index (SSIM). Since the preset matching threshold in this embodiment is 0.85, the result (0.92>0.85) is judged as a match, and the matching result "There is a clear safety hazard of working at height without a safety rope in the current screen" is output.

[0037] When the special safety inspection mechanism is selected, the system sequentially performs the following three judgment and identification operations: identifying whether the construction content in the current screen matches the label, identifying whether the items in the safety hazard list match the label, and identifying whether there are obvious safety hazards in the current screen. The implementation process of these operations is similar to that of the routine safety inspection mechanism, but it places greater emphasis on the matching degree with the preset inspection labels, thereby improving the accuracy of special inspections.

[0038] After completing the identification and judgment process, the system enters the hazard reporting process. The core task of this process is to push video clips containing safety hazards identified in the identification results to the cloud server and generate a reporting record containing the video source number, timestamp, construction content tag, safety hazard type, and severity level. For example, when the system detects a safety hazard of not wearing a safety helmet in a video source, it will extract the video clip containing this hazard and upload it to the cloud server. Simultaneously, the system will generate a reporting record containing the video source number 001, the timestamp October 1, 2023, 15:30, the construction content tag "rebar tying," the safety hazard type "not wearing a safety helmet," and the severity level "medium." This reporting record design allows relevant personnel to quickly locate the problem and take measures, thereby effectively reducing the risks posed by safety hazards.

[0039] Subsequently, the system enters the dynamic assessment record generation stage. The task of dynamic assessment record generation is to record the probability weights of safety hazards occurring in m video sources, and combine this with timestamps, the illuminated construction content, and camera rotation angle information to form dynamic assessment records. Dynamic assessment records are stored in tabular form, with each row containing the video source number, timestamp, construction content tag, camera rotation angle, safety hazard probability weight, and identification result. For example, a video source numbered 002, with a timestamp of October 1, 2023, at 16:00, a construction content tag of "formwork installation," a camera rotation angle of 45 degrees, a safety hazard probability weight of 0.75, and an identification result of "significant safety hazard exists." The formula for calculating the safety hazard probability weight is: in, P The probability weights represent the safety hazards. T The time window length (i.e., the total number of time points within the evaluation period). For the first t The weighting coefficients for each time point (which can be dynamically adjusted based on factors such as time decay and risk level) For the first t A binary variable of safety hazards at each time point (when a hazard exists) =1, when there are no hidden dangers =0). The formula calculates the comprehensive probability of safety hazards within a time window using a weighted summation method. The numerator is the sum of the products of the weight at each time point and the hazard status, and the denominator is the sum of the weight coefficients. The result reflects the weighted proportion of hazards occurring within that window. Its update frequency is set to once per hour. This dynamic assessment record design not only facilitates subsequent data analysis but also provides reliable data support for optimizing inspection strategies.

[0040] Finally, the system enters the stage of updating the list of potential safety hazards and the time periods in which these hazards are likely to occur. The process of updating the list of potential safety hazards involves statistically analyzing the safety hazard items in the current assessment results, extracting items whose frequency exceeds a preset threshold, and updating the list of potential safety hazards. For example, if a safety hazard item appears more than 15% of the current assessment results, it will be included in the list of potential safety hazards.

[0041] The process of updating the time periods prone to safety hazards includes clustering analysis of the time periods of safety hazards in the current assessment results. A hierarchical clustering algorithm is used to construct a tree-like clustering structure for the time periods. The tree is then divided into clusters by setting a similarity threshold, and the time periods prone to safety hazards are determined based on the temporal distribution characteristics of each cluster. For example, if the clustering results show that 9 PM to 11 PM is a high-frequency hazard time period, the system will mark it as a time period prone to safety hazards. The hierarchical clustering algorithm's clustering process includes the following steps: First, each time period is treated as an independent initial cluster, forming an initial set containing N clusters, where N is the total number of time periods. Second, the similarity between all clusters is calculated, and the two clusters with the highest similarity are merged into a new cluster. Finally, the above merging process is repeated, merging the two most similar clusters each time, until all clusters are merged into a single entity or a preset number of clusters is reached. The similarity is calculated using the Pearson correlation coefficient; when the correlation coefficient is greater than 0.9, the two time periods are considered highly similar. In addition, the system records the completion time of each assessment and stores it in the database for subsequent optimization of inspection strategies. The assessment time is recorded and stored as a timestamp, with each assessment's completion time linked to the corresponding video source number, construction content tag, and safety hazard probability weight. This update mechanism allows the system to continuously adapt to changes at the construction site, improving inspection efficiency and accuracy.

[0042] Through the implementation of the above steps, this invention achieves efficient and intelligent analysis of multi-source video data, significantly improving the accuracy and efficiency of safety hazard identification. The application of deep learning models and support vector machine algorithms makes the identification of construction content and safety hazard list items more accurate, while the application of convolutional neural networks and structured similarity index effectively improves the accuracy of safety hazard image detection. The introduction of daily and specialized safety inspection mechanisms, combined with historical data and real-time analysis, dynamically adjusts inspection priorities to ensure comprehensive and targeted inspection coverage. Dynamic evaluation records and cluster analysis accurately identify potential safety hazard list items and hazard-prone time periods, providing data support for subsequent inspection strategy optimization. This invention is applicable to various construction scenarios, flexibly responding to potential safety hazards in different time periods and with different construction content, meeting the needs of enterprise-level construction sites for efficient and intelligent safety hazard inspections.

[0043] The above embodiments are merely one of the preferred embodiments of the present invention and should not be used to limit the scope of protection of the present invention. Any modifications or refinements made to the main design concept and spirit of the present invention that are not of substantial significance, but solve the same technical problem as the present invention, should be included within the scope of protection of the present invention.

Claims

1. A method for identifying and inspecting security hazards in cameras based on video frame image segmentation, characterized in that, Includes the following steps: S1: Obtain data from n video sources through the video cloud storage system, and perform preliminary classification of the data based on construction content tags, time tags, and historical hazard weights to form a basic data pool; S2, select the target video source set from the basic data pool according to the daily security inspection mechanism and the special security inspection mechanism; S3, traverse the selected video sources, and use the region growing algorithm to segment the keyframes of each video source to obtain the segmented image of the construction area; S4. Based on the obtained segmented image, multiple deep learning algorithms are used to perform multiple judgment and recognition operations based on different selection mechanisms to obtain the recognition result; S5 pushes video clips with safety hazards identified in the identification results to the hazard reporting process, generating a reporting record that includes the video source number, timestamp, construction content tag, safety hazard type and severity level; S6 records the probability weight of safety hazards in the video sources selected by the daily safety inspection mechanism, and combines the timestamp, construction content tag and camera rotation angle information to form a dynamic evaluation record; S7: Update the list of potential safety hazards and the time periods when hazards are likely to occur, and record the completion time of each assessment.

2. The method for identifying and inspecting security hazards in cameras based on video frame image segmentation according to claim 1, characterized in that, The construction content label is at least one of the sub-item engineering categories, the time label is divided into time periods in hours, and the historical hazard weight is calculated by statistically analyzing the frequency of safety hazards occurring in each video source within a fixed time period in the past, with a value range between 0 and 1.

3. The method for identifying and inspecting security hazards in cameras based on video frame image segmentation according to claim 2, characterized in that, In step S2, among the target video source sets with different selection mechanisms, the daily security patrol mechanism is as follows: Select m video sources from the basic data pool and divide them into 3 groups. The first group selects i video sources that are prone to problems in the current time period, and determines their priority based on historical data and real-time analysis; The second group selects j video sources with the highest historical hidden danger weights, and matches and filters them by combining construction content tags and time tags; The third group randomly selects k video sources from the basic data pool, or selects k video sources that have not been inspected within a set time period to ensure the breadth and randomness of the inspection coverage; where m = i + j + k. The special safety inspection mechanism is as follows: Based on the pre-set construction content or potential safety hazards in the basic data pool, match data sources are found according to data tags, p video sources are extracted, and special safety inspections are carried out.

4. The method for identifying and inspecting security hazards in cameras based on video frame image segmentation according to claim 3, characterized in that, The region growing algorithm used for image segmentation includes the following steps: The coordinates of the current seed point are calculated based on the average center coordinates of the connected components in the historical segmentation results and the real-time illumination intensity. The U-Net network is used to fuse multi-scale features and output a pixel-level segmentation mask to improve the clarity of image segmentation boundaries and complete image segmentation.

5. The method for identifying and inspecting security hazards in cameras based on video frame image segmentation according to claim 4, characterized in that, In step S4, when the selected mechanism is the daily safety inspection mechanism, the judgment and identification operations performed include: identifying the construction content to which the current screen belongs, identifying the items in the list of potential safety hazards, identifying the time period in which potential safety hazards may occur, and identifying whether there are obvious safety hazards in the current screen.

6. The method for identifying and inspecting security hazards in cameras based on video frame image segmentation according to claim 4, characterized in that, In step S4, when the selected mechanism is the special safety inspection mechanism, the judgment and identification operations performed include: judging whether the construction content of the current screen matches the label, judging whether the safety hazard list item matches the label, and judging whether there are obvious safety hazards in the current screen.

7. The method for identifying and inspecting security hazards in cameras based on video frame image segmentation according to claim 5, characterized in that, The specific implementation process of the judgment and recognition operation is as follows: For identifying the construction content in the current image, the scale-invariant feature transform algorithm is used to extract the connected component features of the segmented region. The connected component features of the segmented image are matched with predefined safety hazard templates. The construction content in the current image is determined by calculating the Euclidean distance between the feature vector of the connected component features and the predefined feature templates in the label library. For the identification of items on the list of potential safety hazards, the support vector machine algorithm is used to classify abnormal behaviors in the video frame where the segmented image is located. The classification boundary is constructed by positive and negative samples in the training set to identify potential items on the list of potential safety hazards. To identify potential safety hazards, a probability distribution model is established based on historical data statistical analysis. By calculating the overlap between the current time point and historical high-risk time periods, the time periods in which potential safety hazards may occur can be determined.

8. The method for identifying and inspecting security hazards in cameras based on video frame image segmentation according to claim 6, characterized in that, The specific implementation process of the judgment and recognition operation is as follows: For identifying the construction content in the current image, the scale-invariant feature transform algorithm is used to extract the connected component features of the segmented region. The connected component features of the segmented image are matched with predefined safety hazard templates. The construction content in the current image is determined by calculating the Euclidean distance between the feature vector of the connected component features and the predefined feature templates in the label library. For the identification of items on the list of potential safety hazards, the support vector machine algorithm is used to classify abnormal behaviors in the video frame where the segmented image is located. The classification boundary is constructed by positive and negative samples in the training set to identify potential items on the list of potential safety hazards. To identify time periods that may pose safety hazards, a probability distribution model in the time dimension is established based on historical data statistical analysis. By calculating the overlap between the current time point and historical high-risk time periods, the time periods in which safety hazards may occur are determined. To identify whether there are obvious security risks in the current scene, a convolutional neural network is used to re-segment the key regions in the video frame where the segmented image is located. The pixel features of the key regions are extracted and matched with predefined security risk templates to output the matching results.

9. The method for identifying and inspecting security hazards in cameras based on video frame image segmentation according to claim 1, characterized in that, In step S7, the process of updating the list of potential safety hazards and the time periods in which safety hazards are likely to occur includes the following steps: S71, Statistically analyze the list of safety hazards in the current assessment results, extract the list of items that appear more frequently than a preset threshold, and update the list of potential safety hazards; S72, perform cluster analysis on the time periods of safety hazards in the current assessment results, use hierarchical clustering algorithm to construct a tree-like clustering structure for the time periods, cut the tree diagram into clusters by setting a similarity threshold, and determine the time periods when safety hazards are likely to occur based on the time distribution characteristics of each cluster; S73 records the completion time of each assessment and stores it in the database for subsequent optimization of inspection strategies.

10. The method for identifying and inspecting security hazards in cameras based on video frame image segmentation according to claim 9, characterized in that, The clustering process of hierarchical clustering algorithms includes the following steps: First, each time period is treated as an independent initial cluster, forming an initial set containing N clusters, where N is the total number of time periods; Secondly, calculate the similarity between all clusters, and merge the two clusters with the highest similarity into a new cluster; Finally, repeat the above merging process, merging the two most similar clusters each time, until all clusters are merged into one whole or the preset number of clusters is reached.

Images