Laser positioning marking method, device and equipment for helmet video acquisition

Abstract

The present invention relates to the field of image positioning and recognition, and in particular to a laser positioning and marking method, device and equipment for safety helmet video acquisition. The method comprises the following steps: real-time acquisition of all-round monitoring video of the construction site based on multiple cameras of the construction site; calculation of delay between adjacent frames of the all-round monitoring video of the construction site, thereby obtaining a global brightness enhanced monitoring video; recognition of bright spots on the construction site and visual detection of laser points of safety helmets on the global brightness enhanced monitoring video, thereby obtaining each laser point of a safety helmet in the construction site; marking of feature points of the construction site at multiple angles according to the all-round monitoring video of the construction site to construct a panoramic three-dimensional model of the construction site; and calculation of visual positioning differences of each laser point of a safety helmet in the construction site according to the panoramic three-dimensional model of the construction site, thereby obtaining the three-dimensional precise positioning coordinates of each laser point. The present invention realizes synchronous tracking of multiple workers and improves the safety of the construction site through efficient and accurate safety helmet positioning marking.

Description

Technical Field

[0001] The present invention relates to the field of image positioning and recognition, and in particular to a laser positioning and marking method, device and equipment for helmet video acquisition. Background Art

[0002] With the continuous development of intelligent technology, hard hats, as an important equipment for workers' safety protection, are widely used in industrial sites, construction sites and other fields. In order to improve the level of safety management on construction sites, real-time monitoring of workers' positions, activity trajectories and the wearing of hard hats has become an important technical requirement. Traditional hard hat position monitoring methods usually rely on manual observation and regular inspections. This method not only consumes a lot of human resources, but also has poor monitoring range and real-time performance, and cannot detect potential safety hazards in a timely manner.

[0003] With the continuous advancement of computer vision, laser scanning and image processing technology, the helmet positioning method based on video acquisition and laser marking has gradually become an intelligent solution. By marking laser points on the helmet and combining video monitoring, the position and movement information of workers can be collected in real time, thereby achieving accurate positioning and trajectory tracking of the helmet. This method not only eliminates the limitations of manual monitoring, but also improves the accuracy and real-time performance of monitoring.

[0004] However, in practical applications, the laser positioning and marking method of helmet video acquisition faces some technical difficulties. First of all, how to ensure that the laser marking point can still be clearly seen under different lighting conditions is a challenge. Secondly, the movement trajectory of the helmet is affected by many factors, such as the complexity of the construction site environment, the change of the camera's viewing angle, etc., which affect the accuracy of laser positioning. In addition, the existing video acquisition technology and laser positioning technology are prone to positioning errors in high-density worker scenes, affecting subsequent data analysis and safety warnings. Therefore, how to improve the stability of laser marking points in various environments, how to achieve synchronous tracking of multiple workers in complex construction site environments, and how to optimize the combination of video acquisition and laser marking have become key issues that need to be urgently solved in this field. Summary of the invention

[0005] In order to solve the above technical problems, the present invention proposes a laser positioning marking method, device and equipment for helmet video acquisition to solve at least one of the above technical problems.

[0006] To achieve the above object, the present invention provides a laser positioning and marking method for helmet video acquisition, comprising the following steps: Step S1: Based on multiple cameras on the construction site, a full-scale monitoring video of the construction site is collected in real time; adjacent frame delays are calculated for the full-scale monitoring video of the construction site, and global brightness enhancement is performed to obtain a global brightness enhanced monitoring video; Step S2: Performing construction site bright spot recognition and helmet laser point visual detection on the global brightness enhancement monitoring video, thereby obtaining each helmet laser point on the construction site; Step S3: marking the construction site feature points from multiple angles based on the construction site's omnidirectional monitoring video, and performing panoramic fusion reconstruction to construct a panoramic construction site three-dimensional model; Step S4: Calculate the visual positioning difference of each helmet laser point in the construction site according to the panoramic construction site three-dimensional model, and perform stereoscopic visual positioning compensation to obtain the three-dimensional precise positioning coordinates of each laser point; Step S5: Based on the three-dimensional precise positioning coordinates of each laser point, dynamic optical flow tracking is performed on the all-round monitoring video of the construction site, and real-time three-dimensional position change estimation is performed to construct a dynamic three-dimensional movement trajectory of each laser point; Step S6: predicting the worker's moving path based on the dynamic three-dimensional moving trajectory of each laser point, and adjusting the posture parameters of the multiple cameras on the construction site in real time to build an intelligent real-time positioning engine.

[0007] In this specification, a laser positioning marking device for helmet video acquisition is provided, which is used to perform the laser positioning marking method for helmet video acquisition as described above, including: The video enhancement module is used to collect all-round monitoring videos of the construction site in real time based on multiple cameras on the construction site; perform delay calculation between adjacent frames of the all-round monitoring videos of the construction site, and perform global brightness enhancement to obtain a global brightness enhanced monitoring video; Laser point visual recognition module, used to identify bright spots on the construction site and visually detect laser points on helmets in the global brightness enhancement monitoring video, so as to obtain each laser point of the helmet on the construction site; The panoramic fusion module is used to mark the construction site feature points from multiple angles based on the construction site's omnidirectional monitoring video, and to perform panoramic fusion reconstruction to build a panoramic construction site 3D model; The visual positioning compensation module is used to calculate the visual positioning difference of each helmet laser point in the construction site based on the panoramic three-dimensional model of the construction site, and perform stereoscopic visual positioning compensation to obtain the three-dimensional precise positioning coordinates of each laser point; Dynamic optical flow tracking module, which is used to perform dynamic optical flow tracking on the all-round monitoring video of the construction site based on the three-dimensional precise positioning coordinates of each laser point, and to perform real-time three-dimensional position change estimation to construct the dynamic three-dimensional movement trajectory of each laser point; The synchronous tracking module is used to predict the worker's movement path according to the dynamic three-dimensional movement trajectory of each laser point, and to adjust the posture parameters of the multiple cameras on the construction site in real time to build an intelligent real-time positioning engine.

[0008] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the laser positioning and marking method for helmet video acquisition described in any one of the above are implemented.

[0009] The beneficial effects of the present invention are as follows: by monitoring the construction site simultaneously with multiple cameras, blind spots under a single viewing angle are eliminated, and more comprehensive coverage is provided, especially in complex and dynamically changing construction site environments, ensuring that all areas are within the monitoring range. Multi-angle video data can more accurately obtain the movement and changes of objects, especially in large-scale construction sites. Different angles of workers or objects provide different information. By calculating the delay between adjacent frames, motion information can be captured more accurately and the accuracy of subsequent frame analysis can be improved. Enhancing the global brightness of the video helps to improve the visual effect of the image, especially in poor lighting environments, improve image quality, and reduce recognition errors caused by uneven lighting. Laser points, as reflective markers, provide stable and obvious recognition features for precise positioning. By detecting the laser points on the helmets, workers can be effectively distinguished and high-precision positioning can be performed. Laser points can be effectively identified under different lighting conditions and are more robust than other features in ordinary images. By identifying bright spots, the real-time position of workers' helmets is continuously monitored, the mobile status is quickly captured, and support is provided for subsequent dynamic tracking. Through the image data captured by multiple cameras, the key feature points of the construction site are marked and matched, and the three-dimensional spatial model of the construction site can be accurately constructed. This is a very important foundation for subsequent positioning and path prediction. Through panoramic fusion reconstruction, a panoramic view of the construction site can be effectively created, providing spatial information of the construction site, helping to analyze the structure and layout of the construction site. Through stereoscopic vision positioning compensation, combined with the panoramic construction site three-dimensional model, the laser points in the two-dimensional image are accurately converted into three-dimensional coordinates to achieve accurate positioning of each helmet laser point. Through difference calculation compensation, the positioning error caused by different angles or viewing distances can be effectively corrected to improve the positioning accuracy. The optical flow tracking technology can capture the motion trajectory of each helmet laser point in real time and dynamically update the worker's position. By analyzing the optical flow changes between adjacent frames, the system accurately estimates the position changes of the laser point in three-dimensional space. The dynamic three-dimensional trajectory of each laser point provides real-time data of the worker's movement, helps monitor and analyze the worker's behavior pattern, and identifies its movement path, stay area, etc. By analyzing the dynamic trajectory of each laser point, the system predicts the worker's future movement path, which is of great significance for construction site safety monitoring, emergency response, etc. Path prediction is also used to identify workers' behavior patterns, such as workers' regular movement routes, stay areas, etc., thereby improving construction site management efficiency. Through dynamic adjustment of the camera posture, the system adjusts the camera's viewing angle according to the prediction of the worker's position, and always keeps real-time tracking of the worker's position. Multiple cameras work together to ensure that no matter how the worker moves, the system can intelligently adjust the viewing angle to achieve accurate worker positioning. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A schematic diagram of the steps of a laser positioning and marking method for helmet video acquisition according to the present invention; Figure 2 Detailed implementation flow chart of step S1; Figure 3 Detailed implementation flow chart of step S2; Figure 4 Detailed implementation flow chart of step S3. DETAILED DESCRIPTION

[0011] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0012] The present application example provides a laser positioning marking method, device and equipment for helmet video acquisition. The execution subjects of the laser positioning marking method, device and equipment for helmet video acquisition include but are not limited to: mechanical equipment, data processing platform, cloud server node, network upload equipment, etc. equipped with the system can be regarded as the general computing node of the present application, and the data processing platform includes but is not limited to: at least one of an audio image management system, an information management system, and a cloud data management system.

[0013] See also Figures 1 to 4 The present invention provides a laser positioning and marking method for helmet video acquisition, and the laser positioning and marking method for helmet video acquisition comprises the following steps: Step S1: Based on multiple cameras on the construction site, a full-scale monitoring video of the construction site is collected in real time; adjacent frame delays are calculated for the full-scale monitoring video of the construction site, and global brightness enhancement is performed to obtain a global brightness enhanced monitoring video; Step S2: Performing construction site bright spot recognition and helmet laser point visual detection on the global brightness enhancement monitoring video, thereby obtaining each helmet laser point on the construction site; Step S3: marking the construction site feature points from multiple angles based on the construction site's omnidirectional monitoring video, and performing panoramic fusion reconstruction to construct a panoramic construction site three-dimensional model; Step S4: Calculate the visual positioning difference of each helmet laser point in the construction site according to the panoramic construction site three-dimensional model, and perform stereoscopic visual positioning compensation to obtain the three-dimensional precise positioning coordinates of each laser point; Step S5: Based on the three-dimensional precise positioning coordinates of each laser point, dynamic optical flow tracking is performed on the all-round monitoring video of the construction site, and real-time three-dimensional position change estimation is performed to construct a dynamic three-dimensional movement trajectory of each laser point; Step S6: predicting the worker's moving path based on the dynamic three-dimensional moving trajectory of each laser point, and adjusting the posture parameters of the multiple cameras on the construction site in real time to build an intelligent real-time positioning engine.

[0014] In the embodiment of the present invention, refer to Figure 1 , is a schematic flow chart of a method of the present invention. In this example, the steps of the method include: Step S1: Based on multiple cameras on the construction site, a full-scale monitoring video of the construction site is collected in real time; adjacent frame delays are calculated for the full-scale monitoring video of the construction site, and global brightness enhancement is performed to obtain a global brightness enhanced monitoring video; In this embodiment, multiple cameras are arranged at key locations on the construction site to ensure coverage of all important areas. The viewing angles of each camera should overlap to form an all-round monitoring network. Select suitable cameras (such as 1080p or 4K resolution) and debug them to ensure their effective capture capabilities under different lighting conditions. Set the frame rate to 30 FPS to ensure video fluency. Use network video transmission protocols (such as RTSP or RTMP) to transmit the real-time video streams of each camera to the central server. Ensure that the network bandwidth is sufficient to support the simultaneous transmission of multiple video streams (the bandwidth is at least 5 Mbps per video stream). Use video encoding compression technology (such as H.264 or H.265) to reduce transmission delays and storage requirements while maintaining video quality. Use computer vision libraries (such as OpenCV) to extract images from real-time monitoring videos frame by frame. Set the frame extraction interval (such as 30 frames per second) to ensure coverage of the entire video segment. Preprocess each extracted frame, including image scaling and format conversion, to accommodate subsequent analysis. Use the differential method to calculate the delay between adjacent frames. Determine the delay of the video stream by calculating the time difference between the current frame and the previous frame. Set a time window (such as 500 milliseconds), calculate the timestamp of each frame within the window and record the time difference. Count the latency data and generate a latency analysis report, including the average, maximum and minimum latency information. These data will provide a reference for subsequent brightness enhancement. Select appropriate image enhancement algorithms, such as histogram equalization and gamma correction. Histogram equalization can improve the contrast of the image, while gamma correction adjusts the brightness of the image. Set parameters, such as gamma value (usually set between 0.5 and 2), to meet the enhancement needs under different lighting conditions. Apply the selected brightness enhancement algorithm to each frame of the image. First, perform histogram equalization to improve the global contrast of the image; then perform gamma correction as needed to adjust the overall brightness of the image. Record the brightness value and contrast of the enhanced image for effect evaluation.

[0015] Step S2: Performing construction site bright spot recognition and helmet laser point visual detection on the global brightness enhancement monitoring video, thereby obtaining each helmet laser point on the construction site; In this embodiment, a computer vision library (such as OpenCV) is used to preprocess each frame of the global brightness enhancement surveillance video, including color space conversion (such as from BGR to HSV) to facilitate subsequent bright spot recognition. The noise in the image is reduced by a denoising algorithm (such as median filtering or Gaussian blur) to ensure that the accuracy is improved in subsequent spot recognition. Set the filter parameters and select a suitable convolution kernel size (such as 5x5). A spot detection method based on color threshold is adopted. Set the threshold range in the HSV color space to extract bright spots. For the color features of the laser point (such as the red laser point), set the threshold range to H: 0-10, S: 100-255, V: 100-255. Use the cv2.inRange() function to generate a binary image, in which the bright spot area is white and the rest is black. Spots in the image are identified by contour detection (such as using the cv2.findContours() function). Feature extraction is performed on the identified bright spots. Calculate the area, center coordinates, and perimeter of each spot to help filter out the laser points on the helmet later. Set an area threshold (such as 10-500 pixels) to filter out spots that are too small or too large. Use shape feature analysis to determine whether the spot is close to a circle. Evaluate its shape by calculating the circularity of the spot (such as the relationship between perimeter and area). Filter the detected spots to ensure that only spots that meet the characteristics of the helmet laser point are retained. Filter out suitable laser points by setting the minimum and maximum radius (such as 5-15 pixels). Record the coordinates and feature information of the helmet laser point to form a data table containing information such as laser point ID, coordinates (X, Y), area, and circularity. Organize all the detected helmet laser point information into a data table to ensure that the data format is clear for subsequent analysis and use. The data table should contain fields such as laser point ID, coordinates, feature parameters, etc. Generate a test report based on the detection of the laser point. The report should include the number of detected laser points, their distribution, and their feature descriptions to help understand the distribution of helmets on the construction site.

[0016] Step S3: marking the construction site feature points from multiple angles based on the construction site's omnidirectional monitoring video, and performing panoramic fusion reconstruction to construct a panoramic construction site three-dimensional model; In this embodiment, a video processing library (such as OpenCV) is used to extract frames from the omnidirectional monitoring video. In order to ensure that sufficient dynamic changes are captured, 30 frames are extracted per second, and video frames are collected from different camera angles. Each frame is converted to grayscale and denoised (using methods such as Gaussian blur) to reduce unnecessary visual noise and improve the accuracy of subsequent feature point extraction. This step is crucial for reducing background interference and enhancing the robustness of target detection. Existing computer vision algorithms (such as SIFT, SURF, or ORB feature extraction algorithms) are used to extract feature points in the construction site from each video frame. These feature points are usually areas with high contrast, complex texture, or prominent corners in the image, and have high The stability and recognizability of the image are ensured by setting a threshold (such as the minimum response value of the feature point is 0.04) to ensure that the extracted feature points are significant and representative in the image. By matching the video frames at different angles (such as using the FLANN matching algorithm), the relative relationship between different camera perspectives is established. The matched feature points lay the foundation for the next step of 3D reconstruction. The checkerboard calibration method is used to shoot images of multiple perspectives using a checkerboard pattern of known size. By calculating the relationship between the feature points in the image and the actual physical world coordinates, the intrinsic parameters (focal length, principal point, distortion coefficient) and extrinsic parameters (camera position and posture) of the camera are estimated. Usually, the calibration tool library in OpenCV (such as cv2.calibrateCamera) to obtain the internal and external parameter matrix of each camera. Using the camera calibration parameters and the previously extracted feature points, triangulation is used to reconstruct each feature point in 3D. Triangulation uses multiple known viewpoints and camera positions to perform geometric calculations on matching feature points to determine their coordinates in 3D space. In the actual calibration process, there are often errors and noises, so it is necessary to use Bundle The Adjustment (bundle adjustment) algorithm globally optimizes the parameters of all cameras and the reconstructed 3D points. This method improves the accuracy of the model by minimizing the reprojection error (i.e., the difference between the actual observation point and the calculated model). The images obtained from different cameras are panoramically stitched, and image registration technology (such as the RANSAC algorithm based on feature points) is used to align the images. Through these registration algorithms, the perspective differences between images are eliminated to synthesize a seamless panorama. The image distortion and chromatic aberration need to be considered in the stitching process, and the radiation distortion model is used for correction. The 3D coordinates are calculated from the matching feature points of multiple perspectives through triangulation. The image features of each matching pair provide information about a 3D point. These 3D points form a dense point cloud on the construction site. The point cloud data is processed using an open source point cloud processing library (such as PCL, Point Cloud Library) to generate a high-precision 3D point cloud model. The 3D mesh model is generated by surface reconstruction of the point cloud using Delaunay triangulation or Poisson Surface. Reconstruction and other methods are used to convert discrete point cloud data into a continuous three-dimensional surface model. In this step, different reconstruction algorithms are selected according to the needs to balance accuracy and efficiency. In order to make the three-dimensional model more realistic, the texture of the panoramic image is mapped to the surface of the three-dimensional model. The registration information of the image stitching result is used to correctly paste the texture onto the three-dimensional grid, thus obtaining a complete three-dimensional site model with texture.

[0017] Step S4: Calculate the visual positioning difference of each helmet laser point in the construction site according to the panoramic construction site three-dimensional model, and perform stereoscopic visual positioning compensation to obtain the three-dimensional precise positioning coordinates of each laser point; In this embodiment, the image data of the construction site at each moment is extracted from the omnidirectional monitoring video frame. First, the image is gray-scale converted and denoised (such as using Gaussian blur) to reduce background interference and improve the contrast of the laser point. Image processing algorithms, such as Hough transform or threshold-based circle detection, are used to identify laser spots in the image. These spots represent the reflection points of the laser, which are usually located on the worker's helmet. The preliminary two-dimensional image coordinates are obtained by calculating the center coordinates of the laser point. The two-dimensional image coordinates of each laser point are converted into an uncertain position in three-dimensional space using known camera calibration parameters (such as focal length, principal point position, etc.). This is because in the absence of depth information, the two-dimensional image coordinates can only reflect the position of the laser point on the image plane. Images taken by different cameras are used for registration to ensure that the position of the same laser point in different images is consistent. Feature point matching is performed using a matching algorithm based on feature points (such as SIFT, ORB), and then the homography matrix is ​​used to register the image to ensure that images from different perspectives can be accurately aligned. By comparing the image coordinates of the same laser point at multiple perspectives, the positioning differences at different perspectives are calculated. Specifically, the deviation of the laser point coordinates under each viewing angle is calculated to obtain the difference vector (dx, dy), which helps to understand the error caused by different viewing angles between images. Combined with the calculated difference vector, a statistical analysis method (such as the least squares method) is used to evaluate the size of the error and correct the error. Specifically, a threshold (such as 1 pixel) is set to determine whether the error is significant, so as to determine whether the positioning result needs to be further corrected. Based on the internal and external parameters of the camera and the two-dimensional coordinates of the laser point under different viewing angles, the depth information of the laser point is calculated using triangulation. Triangulation uses the baseline of the camera (the distance between the two cameras) and the parallax of the same laser point seen from two different viewing angles to estimate the position of the point in three-dimensional space. First, based on the position of the laser point under the two camera views, the parallax between them is calculated, that is, the position difference in the two viewing angles. Then, using the intrinsic parameters (focal length, principal point, etc.) and external parameters (camera position and posture) of the camera, the depth information of the laser point is calculated using the parallax based on the known baseline distance. According to the triangulation formula: Z=(f⋅b) / d, Z is the depth of the laser point, f is the focal length of the camera, B is the baseline between the two cameras, and d is the parallax. According to the calculated depth Z, combined with the two-dimensional coordinates of the laser point, the accurate position of the laser point in three-dimensional space is obtained. The laser points of multiple perspectives are fused, and the weighted average method is used to integrate the depth information under different perspectives. By increasing the number of perspectives, the error under a single perspective is effectively reduced, and more accurate three-dimensional coordinates are obtained. Through the joint optimization of multiple perspectives, the parallax error is further reduced. At this time, optimization algorithms such as weighted least squares are used to globally optimize the coordinates under all perspectives to ensure that the final calculated three-dimensional coordinates are more accurate.By evaluating the accuracy of laser points at known locations (for example, comparing the error with known reference points) and adjusting parameters (such as camera calibration parameters, matching algorithm thresholds, etc.), positioning accuracy can be further improved.

[0018] Step S5: Based on the three-dimensional precise positioning coordinates of each laser point, dynamic optical flow tracking is performed on the all-round monitoring video of the construction site, and real-time three-dimensional position change estimation is performed to construct a dynamic three-dimensional movement trajectory of each laser point; In this embodiment, in the laser scanning system at the construction site, the three-dimensional coordinates of each laser point are obtained. The laser scanner (such as LiDAR system) calculates the three-dimensional coordinates of each scanning point through the time difference between the emission and reception of the laser pulse. In order to ensure the accuracy and reliability of the data, a high-precision laser scanner is required, such as Riegl VZ-400i, Leica BLK360, etc. These devices can provide high spatial resolution and measurement accuracy. On this basis, each frame of the construction site monitoring video is processed. Assuming that the video resolution is 1920×1080 pixels, the frame rate is 30FPS, and the field of view of the video camera is 90°, the internal parameters (focal length, distortion coefficient, etc.) and external parameters (position, posture) of the camera are determined by calibration technology. The three-dimensional coordinates of the laser point are combined with the internal and external parameters of the camera and projected into the two-dimensional image plane of the video to obtain the position of the laser point in each frame of the video. The classic Lucas-Kanade optical flow method or the more advanced Farneback optical flow method is selected to achieve tracking. The Lucas-Kanade optical flow method is a local method suitable for calculating dense optical flow, while the Farneback method performs global estimation through a multi-level image pyramid and is suitable for complex scenes. Assuming that the size of the video frame is 1920×1080, when the Lucas-Kanade method is selected, each time by calculation The image gradient of the local window is used to estimate the optical flow of the pixel. According to the estimated result of the optical flow, the position change of each laser point in the video frame is obtained. The change of its two-dimensional coordinates is estimated through the optical flow information of the laser point in each frame of video. However, the optical flow only provides the motion information in the image plane and does not consider the depth change. Therefore, the depth estimation must be carried out in combination with the initial three-dimensional coordinates of the laser point. The three-dimensional position of each laser point is updated by triangulation based on the relative position relationship between the camera and the laser scanner and the change of each laser point in the image. On the basis of real-time estimation of the three-dimensional position change of each laser point, the dynamic three-dimensional moving trajectory of the laser point is constructed. The motion trajectory of each laser point is regarded as a time series, which records the three-dimensional coordinate change of the laser point at each moment. Through multiple sampling, the complete motion trajectory of the laser point changing with time is obtained. In order to further improve the accuracy and reliability of the trajectory, the Kalman filter algorithm is used to smooth and correct the trajectory data. The Kalman filter effectively handles noise data and uncertainty and provides the best estimate of the position change of the laser point.

[0019] Step S6: predicting the worker's moving path based on the dynamic three-dimensional moving trajectory of each laser point, and adjusting the posture parameters of the multiple cameras on the construction site in real time to build an intelligent real-time positioning engine.

[0020] In this embodiment, by analyzing the laser points in the construction site, the laser points representing the positions of the workers are extracted. The workers usually move in a smaller range on the construction site. Compared with large equipment or buildings, their movement speed is faster and the paths are more tortuous. Therefore, by analyzing the movement patterns of the three-dimensional trajectories, it is inferred that these trajectories are related to the behavior of the workers. By clustering the laser point data and using algorithms such as DBSCAN (density-based spatial clustering algorithm), the positions of the workers are distinguished from the trajectories of other objects. Workers generally cross specific areas more frequently on the construction site, so their trajectories are relatively concentrated, and the moving paths show obvious curves or broken line features. Assuming that there are 20 cameras on the construction site, these cameras are synchronized through time and Spatial calibration and laser point joint positioning are performed. On this basis, the dynamic trajectory of the worker is further analyzed to build a motion model of the worker. According to the dynamic three-dimensional trajectory of the worker, the future movement path is predicted using machine learning methods. Here, deep learning methods such as recursive neural network (RNN) or long short-term memory network (LSTM) are used for modeling. LSTM network is a model commonly used for time series prediction. It can effectively handle the time correlation in the worker trajectory data. Assuming that the three-dimensional coordinate data of each worker is stored in time steps, the training set includes the trajectory data of the past few seconds (such as 5 seconds). The time dependency of the trajectory is learned through the LSTM network to predict the worker's movement path in the future (such as 1 second, 5 seconds). )’s moving path. Specifically, for each worker’s trajectory, we first convert its 3D coordinate data into a continuous time series, and then use the LSTM network for training to learn the law of workers’ movement. By reversely verifying the prediction results, we can get a more accurate prediction of the worker’s position. In the experiment, assuming that the video frame rate is 30FPS, the LSTM network estimates the worker’s moving path at the next moment based on the worker’s current position between each frame, and then predicts its position in the future. At this point, the prediction model provides the necessary data support for the subsequent positioning engine. In order to update the worker’s position and movement trajectory in real time, it is necessary to adjust the posture of the multiple cameras deployed on the construction site in real time. The posture parameters of each camera ( The three-dimensional coordinates of the laser point are matched with the three-dimensional coordinates of the laser point to deduce the actual posture parameters of the camera relative to the construction site. Through these matched feature points, the Pnp (Perspective-n-Point) algorithm is used to realize real-time camera posture estimation, ensuring that the camera's viewing angle matches the actual situation on the construction site.The intelligent real-time positioning engine provides real-time tracking and position prediction services for workers through the integration of workers' predicted paths and multi-camera posture adjustment. The core function of the engine is to dynamically adjust the viewing angles of multiple cameras on the construction site based on the workers' three-dimensional position data, predicted paths and real-time postures of the cameras, so as to accurately track the real-time positions of workers. In the implementation process, the worker's movement trajectory prediction is first combined with the real-time feedback of the camera, and a dynamic feedback control mechanism is adopted. By analyzing the workers' predicted paths and the camera's viewing angle data, the most suitable camera posture is calculated, and the direction of the camera is adjusted within a few milliseconds to ensure that the worker's position in each frame of the image is always within the camera's field of view. In addition, in order to improve the robustness of the system, Kalman filtering or particle filtering algorithms are introduced to deal with the noise problem in the worker's path prediction. In this way, the intelligent positioning engine can achieve high-precision real-time positioning of the worker's position in the construction site environment, and with the cooperation of multiple cameras and laser point data, it effectively avoids the influence of viewing blind spots, occlusions and other problems, and improves the reliability of positioning. ,

[0021] In this embodiment, refer to Figure 2 , is a flowchart of detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: Step S11: Real-time collection of all-round monitoring video of the construction site based on multiple cameras on the construction site; Step S12: Calculate the real-time video frame rate of the all-round monitoring video of the construction site and extract the real-time frame rate of the monitoring video; Step S13: Calculate the delay between adjacent frames according to the real-time frame rate of the monitoring video and extract the inter-frame delay parameter; Step S14: Optimizing the smooth transition of the construction site omnidirectional monitoring video frame by frame based on the inter-frame delay parameter to construct a time-series frame smooth optimization video; Step S15: performing global brightness enhancement on the time-series frame smoothing optimized video, thereby obtaining a global brightness enhanced monitoring video.

[0022] In this embodiment, multiple cameras are arranged in key areas of the construction site to form all-round coverage. The camera should be installed at a high place to maximize the field of view and avoid blind spots. The viewing angle of each camera should be debugged to ensure that overlapping areas are covered. Configure a real-time data transmission system to ensure that the video signal collected by the camera can be transmitted to the central monitoring server in real time. Use a wired network (such as Ethernet) or a wireless network (such as Wi-Fi or 4G / 5G) for video transmission. Set the parameters of video acquisition, such as resolution, frame rate, and compression format (such as H.264), to ensure the balance between video quality and bandwidth during transmission. Set up a monitoring center at the construction site and configure high-performance servers and display devices to ensure that video streams from multiple cameras can be received in real time. The monitoring center should be equipped with video management software (such as Milestone or Genetec) to facilitate real-time monitoring, playback, and data storage. Set a system alarm mechanism so that when an abnormal situation (such as a safety hazard or illegal intrusion) is detected, the system can issue an alarm in time and record the corresponding video clips. Use a video stream processing library (such as OpenCV or FFmpeg) to receive real-time video streams from the monitoring center. Ensure that the video stream can be seamlessly connected to the analysis module and extract frame data. Set the number of frames extracted per second (FPS), which can usually be set to 30 frames per second to ensure the smoothness and real-time performance of the video stream. Use a timer to record the capture timestamp of each frame in the video stream and calculate the number of frames successfully captured per second. Use the time difference method, that is, compare the timestamp difference between two consecutive frames to calculate the frame rate per second. Implement real-time monitoring of the frame rate in the program to ensure that the frame rate can be accurately calculated under different conditions (such as lighting changes or camera movement). Record the real-time frame rate data in the database for subsequent analysis and query. Set the frame rate data to be recorded once a minute and generate a visual chart to show the trend of the frame rate over time. Inter-frame delay refers to the time difference between two adjacent frames of the image, usually in milliseconds (ms). The calculation of delay directly affects the smoothness and real-time performance of the video. Too high delay causes the screen to freeze. Set the delay calculation formula: delay = timestamp (current frame) - timestamp (previous frame) to ensure that the timestamp information can be accurately extracted during calculation. Implement the delay calculation logic in the video stream processing program, loop through each frame, and calculate the delay between adjacent frames. Record the delay value of each calculation and store it in a data structure for subsequent analysis. Set a threshold to monitor delay changes. When the delay exceeds the set reasonable range (such as 100ms), trigger the alarm mechanism to indicate that there is a network or device problem. Perform statistical analysis on the calculated delay data to calculate indicators such as average delay, maximum delay, and minimum delay. Use chart tools (such as Matplotlib or Excel) to generate a delay change curve for intuitive observation of delay trends. Frame smoothing optimization is mainly to solve the problem of screen freeze or jump caused by uneven delay between frames.Generate transition frames between adjacent frames through interpolation algorithms (such as linear interpolation or bilinear interpolation) to make video playback smoother. Set optimization principles to ensure that the generated transition frames are visually smooth and undistorted. Use motion estimation methods to analyze the motion trajectory of objects to ensure that the generation of transition frames conforms to physical laws. Implement frame-by-frame smooth optimization logic in the video processing program. For each pair of adjacent frames, calculate the inter-frame delay and decide whether to generate a transition frame based on the delay. While generating transition frames, ensure that the display time of each transition frame matches the time interval of the original frame to avoid introducing additional delay effects. Record the video data after smooth optimization to ensure that there is no obvious lag in the video stream during playback. Verify the optimization effect through subjective evaluation (watching the video) and objective evaluation (calculating the frame rate). Generate an optimization report to record the optimization status of each video segment, including the comparison before and after the delay change, to ensure that the results are traceable. Select a suitable image brightness enhancement algorithm, such as histogram equalization or gamma correction method. Histogram equalization can effectively improve the contrast, while gamma correction can fine-tune the overall brightness. Consider the overall lighting conditions of the video and set the parameters for brightness enhancement to avoid overexposure or overdarkness and ensure the naturalness of the video. Implement the brightness enhancement logic in the video processing program and process the optimized video frame by frame. For each frame, extract its pixel value and apply the selected brightness enhancement algorithm to generate the enhanced image. Ensure the real-time performance of the enhancement process, usually set to a processing time of no more than 33ms per frame to ensure smooth playback at 30 frames per second. Record the video data of each frame after brightness enhancement and evaluate the enhancement effect. Analyze the changes in image quality before and after enhancement through visual inspection and brightness histogram comparison. Generate a comparison report to show the image effects before and after brightness enhancement to ensure the objectivity and visualization of the results.

[0023] In this embodiment, step S15 includes the following steps: Identify contrast difference areas in the time-series frame smoothing optimized video and extract contrast deviation areas; Performing local gamma correction based on the contrast deviation area to obtain a local gamma corrected video; Extracting a first video frame based on the local gamma corrected video; Calculating the grayscale histogram distribution of the first video frame to obtain a video frame grayscale histogram; Calculate the cumulative distribution function based on the grayscale histogram of the video frame; Dynamically stretching the pixel value range according to the cumulative distribution function to obtain a range-stretched pixel value; Performing pixel mapping transformation on the first video frame based on the range-stretched pixel value, thereby obtaining a brightness-enhanced video frame; Repeat the above steps to process each frame in the video sequence to obtain a global brightness enhanced surveillance video.

[0024] In this embodiment, the contrast of the video frame is calculated using an image processing library (such as OpenCV). The contrast is determined by calculating the standard deviation of the pixel value, and the larger the standard deviation, the higher the contrast. A contrast threshold (such as 50) is set to distinguish between high-contrast and low-contrast areas. The threshold is determined by global statistical analysis to adapt to videos under different ambient lighting conditions. Local area analysis is performed on each frame, and the video frame is divided into multiple small blocks (such as 8x8 or 16x16 pixel areas), and the contrast value of each area is calculated. The contrast value is compared with the set threshold to identify the contrast deviation area. The position coordinates of these areas are recorded for subsequent processing. The core formula of gamma correction is: I_out = I_in ^γ, where I_in is the input pixel value, I_out is the output pixel value, and γ is the gamma value. Select a suitable gamma value according to the degree of contrast deviation (such as 0.5 for brightness enhancement and 2.0 for dark enhancement). For each contrast deviation area, select a suitable gamma value for local gamma correction, and dynamically adjust the gamma value according to the local contrast. For the identified contrast deviation area, apply local gamma correction. First, extract the pixel values ​​of the area, and then perform gamma correction. Update the pixel values ​​of the area and replace the corresponding area in the original video frame. Make sure that the corrected video frame still maintains a natural transition with the entire video sequence to avoid obvious boundaries or gamma correction artifacts. Record the changes in pixel values ​​of each frame after local gamma correction, and generate a comparison chart before and after correction for subsequent analysis. Save the processed video frames to form a local gamma corrected video. Use a video processing library (such as OpenCV) to open the local gamma corrected video and read the first frame of the image. Make sure that the read frame is the corrected video frame to ensure the accuracy of subsequent processing. Use the cv2.VideoCapture() method to obtain the video object and use the read() method to read the first frame. Save the extracted first frame image as a file (such as PNG or JPEG format) to prepare for the subsequent histogram calculation. Make sure that the image quality is not lost when saving, and usually choose a lossless format for saving. Convert the extracted first frame image to a grayscale image to facilitate histogram calculation. Use OpenCV's cv2.cvtColor() method to convert the BGR image to grayscale. Calculate the grayscale histogram and use the numpy library or OpenCV's cv2.calcHist() function to count the pixel value distribution of the grayscale image. Each bar of the histogram represents the number of pixels with a specific grayscale value. Set the grayscale value range (0-255) and calculate the pixel count in each grayscale interval to form the histogram data. Generate a visual chart of the grayscale histogram to intuitively display the pixel distribution. Use the Matplotlib library to draw the histogram and save it as an image file. Calculate the cumulative distribution function (CDF) based on the grayscale histogram.CDF is obtained by accumulating the pixel counts of each gray level in the histogram, indicating the proportion of pixels less than or equal to a certain gray value. The calculation formula is: CDF(i) = ∑ P(x ≤ i), where P is the probability of each gray level and i is the current gray level. Use the numpy library to calculate CDF. In the histogram, the pixel counts of each gray level are first normalized to probability and then accumulated to form a CDF array. Make sure that the final value of CDF is 1 to conform to the characteristics of probability distribution. Generate a visual chart of the cumulative distribution function for subsequent analysis. Use Matplotlib to draw the CDF curve and save it as an image file to intuitively display the cumulative distribution of gray values. According to the CDF, the pixel values ​​of the video frame are dynamically stretched. The stretching formula is: I_out = CDF(I_in) * (L-1), where I_in is the input pixel, I_out is the output pixel, and L is the number of gray levels (usually 256). This method can enhance the contrast of the image, make the pixel value distribution more uniform, and improve the visual effect. Traverse each pixel of the first frame of the extracted image and map the pixel value according to the CDF. The for loop traverses each pixel and replaces its value with the stretched value to ensure that the stretching range is between 0 and 255. For extreme values ​​(such as 0 and 255), ensure that it remains within the legal grayscale range after stretching. Record the pixel value changes after range stretching and generate a comparison chart before and after processing to verify the stretching effect. Save the processed image for subsequent pixel mapping transformation. Use the obtained range stretching pixel value to perform pixel mapping transformation on the first frame of the extracted image. Traverse each pixel of the image and replace the original pixel value with the range stretching result so that each pixel is enhanced. Ensure that the image details are not lost during the processing process and avoid obvious artifacts or blurring. Record the image data after brightness enhancement to ensure the traceability of the results. Save the enhanced image as a file (such as PNG or JPEG format) to prepare for subsequent processing steps. Implement the above processing steps for each frame in the local gamma correction video. Use a loop structure (such as a for loop) to read each frame of the image in turn and repeat the steps of contrast difference area identification, gamma correction, grayscale histogram calculation, etc. Set up a processing progress monitoring mechanism, such as displaying the processing progress every 10 frames to track the overall processing progress. Save each processed frame of the image as a new video sequence to form a global brightness enhancement monitoring video. Use OpenCV's cv2.VideoWriter() function to write each frame to the output video file. Ensure that the video format and frame rate settings are consistent with the original video to maintain smooth playback. Evaluate the final generated global brightness enhancement monitoring video, check the visibility and contrast of the video, and ensure that the processing effect meets expectations. Perform a comprehensive evaluation through visual inspection and objective indicators (such as contrast, brightness histogram).

[0025] In this embodiment, refer to Figure 3 , is a flowchart of detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: Step S21: performing construction site bright spot recognition on the global brightness enhanced monitoring video to extract all bright spots; Step S22: performing spot region image segmentation on all bright spots to generate a plurality of spot region image frames; Step S23: performing spot geometric morphology analysis on multiple spot area image frames to obtain geometric morphology features of each spot; Step S24: performing deep semantic analysis on each spot geometric morphological feature, thereby obtaining semantic features of bright spots; Step S25: mining environmental interference light spots based on the semantic features of the bright spots to mark the environmental interference light spots; Step S26: Perform visual detection of the helmet laser point on the global brightness enhancement monitoring video based on the environmental interference light points, so as to obtain each helmet laser point in the construction site.

[0026] In this embodiment, the top of the helmet is provided with a laser emitter, and a suitable bright spot detection algorithm is selected, such as Hough transform or color-based threshold segmentation. Hough transform is suitable for detecting circular light spots, while color threshold segmentation extracts bright spots according to brightness and color features. Algorithm parameters, such as brightness threshold (such as pixel values ​​set to 200 or more) are set to identify bright parts. A suitable color space (such as HSV color space) is selected to improve the accuracy of spot recognition. The global brightness enhancement monitoring video is read frame by frame using a video processing library (such as OpenCV), and the bright spot recognition algorithm is applied to each frame of the image. Ensure that the processing time for each frame is as short as possible to maintain the real-time nature of the video. During the processing, the pixel value of each spot is detected, and a binary image is generated using image binarization technology, in which the bright area is white and the other areas are black. After the recognition is completed, the coordinates and size information of the bright spots extracted in each video frame are recorded. The location information of the spots is stored in a data structure, such as a list or a dictionary, for subsequent processing. The bright spots in all frames are summarized to form a list containing all bright spot information for subsequent analysis. The connected component labeling algorithm is used to segment the extracted bright spots. This algorithm can identify all connected bright areas in the image and assign a unique label to each area. During the segmentation process, the minimum spot area threshold (such as 10 pixels) is set to filter out noise and small spots to ensure that only valid bright spot areas are retained. The connected component labeling algorithm is applied to the bright spots extracted in each frame to generate a spot area image frame. The bounding box of each spot is calculated, and the coordinate information of each area (upper left corner and lower right corner coordinates) is recorded. The findContours function of OpenCV is used to further extract the contour information of each spot and generate the corresponding image frame for subsequent analysis. The spot area information obtained by segmentation is recorded, including the coordinates, area and bounding box of each area. During the processing, the segmented image frame is superimposed on the original image to generate a visual result to check the accuracy of the segmentation effect. The information of each area is stored in structured data (such as a CSV file or database) for subsequent analysis and processing. The morphological analysis method is used to extract the geometric features of the spots, including area, perimeter, roundness, aspect ratio, etc. The roundness can be calculated by the formula roundness = (4 × π × area) / perimeter² to evaluate the shape characteristics of the spots. Use the functions provided by OpenCV (such as contourArea and arcLength) to quickly calculate the geometric features of each spot to ensure the efficiency and accuracy of the calculation. Analyze each stored spot area and calculate its geometric features. Record the area, perimeter, roundness, aspect ratio and other information of each spot to form a feature vector.For each spot, when calculating the aspect ratio, make sure to obtain the size of the spot's minimum bounding rectangle to facilitate accurate measurement of the aspect ratio. Organize the geometric features of each spot into a table, including information such as spot ID, area, perimeter, roundness, and aspect ratio. Ensure that the data is formatted clearly for subsequent analysis. When recording the analysis results, use data visualization tools to generate distribution maps of each feature to help further analyze the geometric characteristics of the spots. Select a suitable deep learning model (such as a convolutional neural network CNN) for semantic feature parsing. Use pre-trained models (such as ResNet or VGG) for fine-tuning to suit the feature extraction task of bright spots. Build a deep learning framework (such as TensorFlow or PyTorch) and set the model's input size, batch size, and rate to ensure efficient model training. Prepare a training dataset, including annotated bright spot feature images and their corresponding semantic labels. Increase the diversity of training data through data augmentation (such as rotation, scaling, and flipping). Train the model, record the loss function and accuracy of each training cycle to evaluate the performance of the model, and tune it on the validation set. After the model training is completed, use the model to perform deep semantic analysis on the geometric features of each spot, and extract the semantic features of the bright spots, such as "safety sign", "warning signal", etc. Record the semantic features of each spot and combine them with the geometric features to form a comprehensive feature vector for subsequent use. Based on the semantic features of the bright spots extracted previously, set the feature standards for interfering light spots. Interfering light spots usually have a small area, high brightness, and irregular shape. Determine the classification standards for interfering light spots, such as brightness thresholds (such as pixels with brightness greater than 255) and area thresholds (such as less than 30 pixels). Screen each bright spot and identify the interfering light spots based on the set feature standards. Use conditional statements to check the geometric and semantic features of each spot one by one to determine whether it meets the standards for interfering light spots. Mark the interfering light spots that meet the conditions, record their coordinates, area, and other relevant features, and store them in a data structure for subsequent use. Organize the information of interference light spots and form a detailed report of interference light spots, including the feature description of each light spot and its location in the surveillance video. Use visualization tools to overlay interference light spots on the original surveillance video to facilitate observation and verification of the identification effect of interference light spots. Define the characteristics of the helmet laser spot, such as brightness, color, and shape. Laser spots usually have a specific brightness value (such as above 200) and appear as small dots. Set the detection algorithm and combine it with the previous interference light spot information to ensure that the helmet laser spot can be accurately distinguished from other bright spots. Use a video processing library (such as OpenCV) to analyze the global brightness enhanced surveillance video frame by frame, and apply the laser point detection algorithm to each frame of the image.Each detected bright spot is further analyzed to check whether it meets the characteristic standards of the helmet laser point to ensure the accuracy of the detection.

[0027] In this embodiment, refer to Figure 4 , is a flowchart of detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include: Step S31: Marking construction site feature points at multiple angles based on the construction site's omnidirectional surveillance video to generate key construction site feature points at different angles; Step S32: performing multi-directional matching on key feature points of the construction site at different angles, thereby generating key feature point position matching data; Step S33: performing video stream global posture correction on the key feature point position matching data to obtain global posture correction parameters at multiple angles; Step S34: Calculate the spatial position of the multiple cameras on the construction site and extract the spatial position coordinates of each camera; Step S35: fitting the three-dimensional environment of the construction site according to the spatial position coordinates of each camera to construct a three-dimensional environment map of the construction site; Step S36: Perform panoramic fusion reconstruction on the construction site three-dimensional environment map based on the global posture correction parameters of multiple angles to construct a panoramic construction site three-dimensional model.

[0028] In this embodiment, the selection criteria of the feature points are determined, such as edge significance, corner detection, texture change, etc. Harris corner detection algorithm or FAST algorithm is used to identify important feature points in the image, and these feature points should have strong stability and repeatability under different viewing angles. Set detection parameters, such as corner response threshold and detection window size, to ensure that the extracted feature points are both representative and can filter out noise. Use a computer vision library (such as OpenCV) to process the surveillance video frame by frame, and apply the selected feature point detection algorithm to extract key feature points in the construction site. During the processing, the feature points of each frame image are marked and stored. For each feature point, its coordinates, response value and other relevant information (such as feature descriptors) are recorded to ensure data integrity. Use multiple cameras to shoot the construction site from different angles to ensure that all key areas are covered. Perform the same feature point detection and marking on the images captured by each camera to form a multi-angle feature point database. Organize the recorded feature point information to form structured data containing multi-angle feature point positions and descriptors for subsequent matching and analysis. Select appropriate feature point matching algorithms, such as descriptor-based matching (such as SIFT, SURF, or ORB), which can effectively cope with the effects of rotation, scaling, and illumination changes. Set matching parameters, such as the maximum distance ratio (usually set to 0.75) and the minimum number of matches to ensure the accuracy and reliability of the matching results. Match each pair of feature point sets from different cameras, use the selected matching algorithm to calculate the distance between the descriptors, and filter out valid matches based on the set threshold. Record the location, descriptor, and matching quality of each pair of matching feature points for subsequent analysis and correction. Organize the matching results into a matching data table, which contains the coordinates of each pair of matching feature points and the information of the camera from which they come. Perform statistical analysis on the accuracy of the matching, calculate the matching rate and the false matching rate, and ensure the reliability of the matching results. Use visualization tools to display the matching results to help intuitively understand the matching effect and the distribution of feature points. Select an appropriate global pose correction algorithm, such as the PnP (Perspective-n-Point) algorithm, which calculates the pose of the camera based on known 3D points and corresponding 2D image points. Set algorithm parameters such as initial guess, maximum number of iterations, and reprojection error threshold to ensure the accuracy and convergence of the correction. Using the matched feature point position data, apply the PnP algorithm to calculate the pose parameters (rotation matrix and translation vector) of each camera. Combine the 3D coordinates of the feature points with the corresponding 2D image coordinates for pose calculation. Perform multiple corrections on each camera to ensure the stability of the results. Record the output of each correction and calculate the reprojection error to evaluate the accuracy of the correction. Using the pose parameters obtained in the previous step and the known camera installation positions, calculate the position coordinates of each camera in 3D space. Use basic geometric transformation formulas for calculations.Set a coordinate system (such as a right-handed coordinate system) to ensure that the spatial positions of all cameras are in the same coordinate reference system. Parse the posture parameters of each camera and calculate its actual position in 3D space. Record each step in the calculation process to ensure the accuracy of the results. Combine the installation height and direction of the camera to obtain the complete spatial coordinates (x, y, z) of each camera. Organize the spatial position coordinates of each camera into a data table to ensure that the data is clearly formatted and includes the camera ID and other relevant information. Generate a spatial position report that describes the location of each camera and its distribution in the construction site in detail for subsequent analysis and visualization. Select a suitable 3D environment fitting algorithm, such as Delaunay triangulation or grid-based fitting methods. These algorithms can effectively process irregular 3D point cloud data and generate smooth 3D models. Determine fitting parameters, including point cloud density, triangle generation parameters, etc., to ensure the accuracy and richness of details of the generated 3D environment map. Generate 3D point cloud data of the construction site based on the spatial position coordinates of each camera. Convert the feature points in the field of view of each camera into 3D coordinates to form a complete point cloud data set. Filter and de-noise the point cloud data to remove outliers and noise to ensure the quality of subsequent fitting. Apply the selected 3D fitting algorithm to fit the processed point cloud data to generate a 3D environmental map of the construction site. Ensure that the details and structure of the map can truly reflect the actual situation of the construction site. Combine the generated 3D map with the original surveillance video to form a visualization effect for subsequent analysis and display. Select a suitable panoramic fusion reconstruction algorithm, such as the Multi-View Stereo (MVS) algorithm. This algorithm can fuse the perspective information from different cameras to generate a high-quality 3D model. Set reconstruction parameters, such as perspective overlap and depth map accuracy, to ensure the quality of the reconstruction results. Use the previously obtained global posture correction parameters to align and fuse the 3D environmental maps from different angles. Ensure that the perspective of each camera is accurately reflected in the final model. Record each step in the reconstruction process to ensure the traceability of the results. Ensure that the overlap and occlusion problems that occur during the fusion process are handled to generate a complete panoramic 3D model. After completing the panoramic fusion, generate the final 3D model of the construction site and visualize it. Use 3D visualization tools (such as Meshlab or Blender) to display the details of the model for further analysis and application. Generate a detailed model report to record the parameters and results of each step, including the accuracy and details of the model, for subsequent use and sharing.

[0029] In this embodiment, step S4 includes the following steps: Step S41: Perform multi-angle detection and identification on each safety helmet laser point in the construction site to obtain laser point detection parameters at all angles; Step S42: Calculate the laser point coordinates according to the laser point detection parameters at all angles to obtain the laser point coordinates at each angle; Step S43: performing visual positioning difference calculation on the laser point coordinates at each angle according to the panoramic construction site three-dimensional model to generate the laser point parallax parameters at each angle; Step S44: Perform stereoscopic vision positioning compensation based on the parallax parameters of the laser point at each angle, so as to obtain the three-dimensional precise positioning coordinates of each laser point.

[0030] In this embodiment, a suitable laser point detection algorithm is selected, such as a combination method based on color threshold and edge detection. Laser points usually have high brightness and specific colors (such as red or green), so they are extracted by setting color thresholds. Hough transform is used to detect the circular features of laser points. Parameters such as minimum and maximum radius are set to ensure that only actual laser points are detected. Use multiple cameras to shoot the construction site from different angles to capture the laser points on the safety helmet in real time. The input video stream of each camera is processed and the laser point detection algorithm is applied frame by frame. The position information of the laser point is extracted in each frame, and relevant parameters such as brightness value, center coordinates and radius are recorded for subsequent analysis. The detected laser point parameters are organized into a data table containing camera ID, laser point coordinates, detection timestamp and other relevant information. These data will provide a basis for subsequent steps. Generate a laser point detection report to show the number and distribution of laser points detected by different cameras to evaluate the effectiveness and accuracy of the detection. The perspective projection model is used to convert the detected laser points from image coordinates to actual world coordinates in combination with the camera's internal parameters (focal length, principal point coordinates) and external parameters (rotation matrix and translation vector). According to the detection parameters of the laser point (such as image coordinates and depth information), set the coordinate calculation formula: X = (u - cx) * Z / fx, Y = (v - cy) * Z / fy, where (u, v) is the image coordinate, (cx, cy) is the main point, and (fx, fy) is the focal length. For each detected laser point, use the above model to calculate its position in three-dimensional space. It is necessary to ensure that the depth information of the laser point (such as obtained by the sensor or estimated by other methods) is accurate. Store the calculated three-dimensional coordinates of each laser point and record the corresponding camera information and calculation time for subsequent processing. Organize the three-dimensional coordinates of all laser points into a data table, including the laser point ID, coordinates (X, Y, Z), corresponding camera ID and other relevant information. Ensure that the data structure is clear for subsequent analysis and use. Generate a coordinate calculation report that describes the calculation process and results of each laser point in detail to ensure the traceability and accuracy of the results. Parallax refers to the change in the position of an object on the image plane when the same object is observed from different perspectives. For laser points, parallax is calculated by comparing the coordinates of the laser points under different camera perspectives. Set the parallax calculation formula: d = X1 - X2, where X1 and X2 are the X coordinates of the same laser point detected from the two cameras. For each laser point, calculate the corresponding parallax parameters according to its coordinates under different cameras. Ensure that the matching of laser points is accurate when calculating parallax to avoid confusion between different laser points. Record the parallax parameters of each laser point and store the corresponding camera ID and coordinate information for subsequent analysis. Organize the parallax parameters into a data table containing the laser point ID, parallax value, corresponding camera ID and other relevant information. Ensure that the data is in a clear format for subsequent analysis.Use visualization tools (such as Matplotlib) to generate disparity distribution maps to help intuitively understand the disparity situation and its impact on laser point positioning. Select a suitable stereo vision compensation algorithm, such as binocular stereo vision calculation, which uses the disparity information of laser points under two perspectives to reconstruct the accurate three-dimensional position of the laser point. Set compensation parameters, including disparity threshold and geometric model parameters, to ensure the accuracy of compensation calculation. For each laser point, perform stereo vision compensation according to its disparity parameters. By combining the laser point coordinates and disparity under different perspectives, calculate the final three-dimensional precise positioning coordinates. Use the formula for compensation calculation: Z = f *B / d, where f is the focal length of the camera, B is the baseline distance between the two cameras, and d is the disparity. Organize the final three-dimensional coordinates of each laser point into a data table, including information such as laser point ID, three-dimensional coordinates (X, Y, Z), corresponding camera ID and disparity parameters. Generate a compensation result report, describing the compensation process and results in detail to ensure the accuracy and traceability of the results.

[0031] In this embodiment, step S5 includes the following steps: Step S51: performing time-series frame analysis on the all-round monitoring video of the construction site to obtain time-series frame image change characteristics; Step S52: performing dynamic optical flow tracking on the three-dimensional precise positioning coordinates of each laser point according to the change characteristics of the time-series frame image to obtain the optical flow change vector of each laser point; Step S53: identifying the optical flow change direction of the optical flow change vector of each laser point, and generating the optical flow change direction of the laser point; Step S54: calculating the optical flow fluctuation amplitude of the optical flow change vector of each laser point to generate the optical flow fluctuation amplitude of the laser point; Step S55: perform real-time three-dimensional position change estimation based on the change direction of the laser point optical flow and the fluctuation amplitude of the laser point optical flow, and construct a dynamic three-dimensional movement trajectory of each laser point.

[0032] In this embodiment, a video processing library (such as OpenCV) is used to extract images from the all-round monitoring video of the construction site frame by frame. The frame extraction interval (such as extracting 30 frames per second) is set to ensure that the entire video segment is covered to capture dynamic changes. Each frame of the extracted image is preprocessed, including grayscale conversion and noise removal (such as using Gaussian blur) to improve the accuracy of subsequent analysis. The image difference algorithm is used to calculate the changes between adjacent frames. The frame difference method (Frame Differencing) is used, that is, the change area is identified by calculating the absolute difference between two frames. A change threshold is set (such as 50), and the area with a change amplitude exceeding the threshold is marked as a significant change for subsequent analysis. The change features of each frame of the image are recorded in a data table, including the coordinates, change amplitude and timestamp of the change area. Ensure that the data structure is clear for subsequent use. The Lucas-Kanade optical flow algorithm is used, which is suitable for optical flow calculation under small displacement conditions. The algorithm analyzes the image brightness changes through a local window to estimate the motion vector of each feature point. The window size of the optical flow calculation (such as 5x5 pixels) is set to balance the calculation accuracy and efficiency. For each laser point's 3D coordinates, apply the optical flow algorithm to perform dynamic tracking in adjacent time frames. Make sure to use the laser point's initial position as input in each tracking. Calculate the optical flow change vector for each laser point and record its coordinate changes in each frame. Organize the optical flow change vector for each laser point into a data table containing the laser point ID, timestamp, optical flow change vector (dx, dy, dz), and corresponding frame number. Generate an optical flow tracking result report to show the movement of the laser point in different time periods for subsequent analysis. By calculating the direction of the optical flow change vector, use the inverse tangent function (atan2(dy, dx)) to determine the direction of movement of each laser point. This method can handle 360-degree direction changes and avoids quadrant confusion. Set the output range of the direction to 0 to 360 degrees for subsequent analysis and visualization. Calculate the direction of the optical flow change vector for each laser point and record the direction of movement of each laser point in each frame. Organize the calculation results together with the optical flow change vector for subsequent analysis. Organize the optical flow change direction of each laser point into a data table, including information such as laser point ID, timestamp, and optical flow change direction. Ensure that the data structure is clear for subsequent use. The fluctuation amplitude can be obtained by calculating the modulus of the optical flow change vector, using the formula: amplitude = sqrt(dx^2 +dy^2 + dz^2), where (dx, dy, dz) are the components of the optical flow change vector. Set an amplitude threshold (such as 0.1) to distinguish between significant fluctuations and minor fluctuations. Calculate the amplitude of the optical flow change vector of each laser point, record the fluctuation amplitude of each laser point in each frame, and mark whether it exceeds the set threshold. Store the calculation results for subsequent analysis and visualization.The fluctuation amplitude of each laser point is organized into a data table, including information such as laser point ID, timestamp, fluctuation amplitude, etc. Ensure that the data format is clear for subsequent analysis. Generate a fluctuation amplitude analysis report to show the fluctuation of the laser point in different time periods to help understand its dynamic characteristics. Based on the direction of optical flow change and fluctuation amplitude, a simple motion model (such as a uniform linear motion model) is used to estimate the dynamic movement trajectory of each laser point. Set the time interval (such as 1 frame time) as the basic unit of trajectory calculation. By calculating the predicted position of the laser point at each time step, combined with its optical flow change vector and fluctuation amplitude, the position is updated. For each laser point, its three-dimensional position coordinates are updated frame by frame according to its optical flow change vector and fluctuation amplitude. Record the new position of each laser point in each frame to form complete dynamic trajectory data. The dynamic three-dimensional movement trajectory of each laser point is organized into a data table, including information such as laser point ID, timestamp, trajectory coordinates, etc. Ensure that the data structure is clear for subsequent use.

[0033] In this embodiment, step S6 includes the following steps: Step S61: performing dynamic movement trajectory rendering on the panoramic construction site three-dimensional model according to the dynamic three-dimensional movement trajectory of each laser point, and constructing a dynamic trajectory construction site rendering model; Step S62: performing segment-by-segment movement evolution analysis on the dynamic trajectory construction site rendering model to obtain movement change characteristics of each segment of the trajectory; Step S63: predicting the worker's moving path based on the moving change characteristics of each trajectory, thereby obtaining the predicted moving path of each worker; Step S64: adjusting the real-time posture parameters of the multiple cameras on the construction site according to the predicted movement path of each worker to obtain real-time posture adjustment data; Step S65: Perform intelligent synchronous capture based on the real-time posture adjustment data to build an intelligent real-time positioning engine to perform real-time positioning and marking operations for safety helmets on the construction site.

[0034] In this embodiment, a suitable 3D rendering engine is selected, such as Unity or Unreal Engine, which can efficiently handle dynamic rendering and real-time interaction, and rendering parameters, including lighting, material, and camera perspective, are determined to ensure that the rendering effect is realistic and visually attractive. The dynamic 3D movement trajectory data of the laser point is imported into the rendering engine and formatted into a data structure suitable for engine processing. The trajectory of each laser point is ensured to be continuous and smooth. The movement trajectory of the laser point is represented in the 3D model in the form of lines or dots, and the color and transparency of the trajectory are set so that it can be clearly seen in a complex environment. During the rendering process, the position of the laser point is dynamically updated, and its movement trajectory is displayed in real time. An interpolation algorithm (such as linear interpolation or spline interpolation) is used to smooth the trajectory to ensure smooth animation. Performance optimization is performed, and the rendering frame rate (such as 60 FPS) and LOD (Level of Depth) are adjusted. Detail) settings to ensure smooth operation on different devices, divide the dynamic trajectory into several segments according to time or distance, set segmentation parameters (such as one segment every 5 seconds or every 10 meters), and analyze the movement characteristics of each segment. For each segment of the trajectory, calculate basic motion characteristics such as displacement, speed, and acceleration to quantify the changes in each segment of the trajectory. Use statistical analysis tools (such as NumPy and Pandas) to summarize the movement characteristics of each segment of the trajectory, calculate indicators such as average speed and maximum displacement, and generate visual charts to show the movement change characteristics of each segment of the trajectory, including speed change curves and displacement graphs, for intuitive understanding. Organize the movement change characteristics of each segment of the trajectory into a data table, including information such as trajectory segment ID, displacement, speed, and acceleration, to ensure a clear data structure for subsequent use, prepare a trajectory evolution analysis report, describe the movement characteristics and change trends of each segment of the trajectory in detail, and provide a basis for analysis. Use machine learning algorithms (such as linear regression, random forest, or LSTM) to predict the movement path of workers. These The algorithm can learn the workers' movement patterns based on historical data, determine training parameters such as learning rate, number of iterations and cross-validation method to ensure the accuracy and generalization ability of the model, use the analyzed trajectory segment movement change characteristics as training data, including displacement, speed, acceleration and timestamp information, divide the data set into training set and test set (such as 80% training, 20% test), standardize the data to ensure the uniformity of input features to improve the learning effect of the model, and after training, use the test set to evaluate the model, calculate the prediction accuracy (such as mean square error) and tune it, apply the model to real-time data to predict the future movement path of the workers, record the predicted movement path of each worker, and output it as a visualization result to show the comparison between the predicted path and the actual movement trajectory. According to the predicted movement path of the workers, the posture parameters of the camera are adjusted based on the optimization algorithm (such as particle swarm optimization or genetic algorithm), and the objective function is set to minimize the error between the predicted path and the actual path. The adjustment parameters are determined.Including rotation angle, translation distance and camera field of view to ensure the best capture of workers' movements. In real-time monitoring, use artificial intelligence algorithms (such as reinforcement learning) to automatically adjust the posture parameters of each camera to adapt to the movement trajectory of workers, ensure the real-time and efficiency of the adjustment process, record the posture adjustment parameters of each camera, and update them in real time to ensure the optimization of monitoring effects. Organize the real-time posture adjustment data into a data table, including camera ID, adjustment parameters (rotation, translation) and timestamp information, ensure the data structure is clear for subsequent use, generate a posture adjustment report, describe in detail the adjustment process of each camera and its impact on the monitoring effect, provide a basis for improvement, and design the system architecture of the intelligent real-time positioning engine, including data acquisition module, data processing Modules and real-time monitoring modules to ensure data flow between modules, determine the system workflow, including data input, processing, output and feedback, to achieve real-time positioning and marking functions, develop intelligent capture algorithms based on real-time posture adjustment data, combine laser point detection with real-time monitoring data, realize automatic positioning and marking of safety helmets on construction sites, set capture parameters, such as capture frequency (such as 10 times per second) and marking accuracy, ensure the accuracy and efficiency of real-time positioning, conduct system tests on the intelligent real-time positioning engine, evaluate its performance in real-time construction site environments, including positioning accuracy and real-time performance, optimize the system based on the test results, improve the algorithm and parameter settings, record system operation data, and generate performance analysis reports to demonstrate the effectiveness and reliability of the intelligent capture system and provide reference for subsequent improvements. ,

[0035] The present invention also provides a laser positioning and marking device for helmet video acquisition, which is used to execute the laser positioning and marking method for helmet video acquisition as described above, comprising: The video enhancement module is used to collect all-round monitoring videos of the construction site in real time based on multiple cameras on the construction site; perform delay calculation between adjacent frames of the all-round monitoring videos of the construction site, and perform global brightness enhancement to obtain a global brightness enhanced monitoring video; Laser point visual recognition module, used to identify bright spots on the construction site and visually detect laser points on helmets in the global brightness enhancement monitoring video, so as to obtain each laser point of the helmet on the construction site; The panoramic fusion module is used to mark the construction site feature points from multiple angles based on the construction site's omnidirectional monitoring video, and to perform panoramic fusion reconstruction to build a panoramic construction site 3D model; The visual positioning compensation module is used to calculate the visual positioning difference of each helmet laser point in the construction site based on the panoramic three-dimensional model of the construction site, and perform stereoscopic visual positioning compensation to obtain the three-dimensional precise positioning coordinates of each laser point; Dynamic optical flow tracking module, which is used to perform dynamic optical flow tracking on the all-round monitoring video of the construction site based on the three-dimensional precise positioning coordinates of each laser point, and to perform real-time three-dimensional position change estimation to construct the dynamic three-dimensional movement trajectory of each laser point; The synchronous tracking module is used to predict the worker's movement path according to the dynamic three-dimensional movement trajectory of each laser point, and to adjust the posture parameters of the multiple cameras on the construction site in real time to build an intelligent real-time positioning engine.

[0036] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the laser positioning and marking method for helmet video acquisition described in any one of the above are implemented.

[0037] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0038] The above is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but should conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A laser positioning and marking method for helmet video acquisition, characterized in that: The following steps are involved: Step S1: Real-time collection of all-round monitoring video of the construction site based on multiple cameras on the construction site; The delay between adjacent frames of the all-round monitoring video of the construction site is calculated, and the global brightness is enhanced to obtain the monitoring video with global brightness enhancement; Step S2: Performing construction site bright spot recognition and helmet laser point visual detection on the global brightness enhancement monitoring video, thereby obtaining each helmet laser point on the construction site; Step S3: marking the construction site feature points from multiple angles based on the construction site's omnidirectional monitoring video, and performing panoramic fusion reconstruction to construct a panoramic construction site three-dimensional model; Step S4: Calculate the visual positioning difference of each helmet laser point in the construction site according to the panoramic construction site three-dimensional model, and perform stereoscopic visual positioning compensation to obtain the three-dimensional precise positioning coordinates of each laser point; Step S5: Based on the three-dimensional precise positioning coordinates of each laser point, dynamic optical flow tracking is performed on the all-round monitoring video of the construction site, and real-time three-dimensional position change estimation is performed to construct a dynamic three-dimensional movement trajectory of each laser point; Step S6: predicting the worker's moving path based on the dynamic three-dimensional moving trajectory of each laser point, and adjusting the posture parameters of the multiple cameras on the construction site in real time to build an intelligent real-time positioning engine.

2. The laser positioning and marking method for helmet video acquisition according to claim 1, characterized in that: The specific steps of step S1 are: Step S11: Real-time collection of all-round monitoring video of the construction site based on multiple cameras on the construction site; Step S12: Calculate the real-time video frame rate of the all-round monitoring video of the construction site and extract the real-time frame rate of the monitoring video; Step S13: Calculate the delay between adjacent frames according to the real-time frame rate of the monitoring video and extract the inter-frame delay parameter; Step S14: Optimizing the smooth transition of the construction site omnidirectional monitoring video frame by frame based on the inter-frame delay parameter to construct a time-series frame smooth optimization video; Step S15: performing global brightness enhancement on the time-series frame smoothing optimized video, thereby obtaining a global brightness enhanced monitoring video.

3. The laser positioning and marking method for helmet video acquisition according to claim 2 is characterized in that: The specific steps of step S15 are: Identify contrast difference areas in the time-series frame smoothing optimized video and extract contrast deviation areas; Performing local gamma correction based on the contrast deviation area to obtain a local gamma corrected video; Extracting a first video frame based on the local gamma corrected video; Calculating the grayscale histogram distribution of the first video frame to obtain a video frame grayscale histogram; Calculate the cumulative distribution function based on the grayscale histogram of the video frame; Dynamically stretching the pixel value range according to the cumulative distribution function to obtain a range-stretched pixel value; Performing pixel mapping transformation on the first video frame based on the range-stretched pixel value, thereby obtaining a brightness-enhanced video frame; Repeat the above steps to process each frame in the video sequence to obtain a global brightness enhanced surveillance video.

4. The laser positioning and marking method for helmet video acquisition according to claim 1, characterized in that: The specific steps of step S2 are: Step S21: performing construction site bright spot recognition on the global brightness enhanced monitoring video to extract all bright spots; Step S22: performing spot region image segmentation on all bright spots to generate a plurality of spot region image frames; Step S23: performing spot geometric morphology analysis on multiple spot area image frames to obtain geometric morphology features of each spot; Step S24: performing deep semantic analysis on each spot geometric morphological feature, thereby obtaining semantic features of bright spots; Step S25: mining environmental interference light spots based on the semantic features of the bright spots to mark the environmental interference light spots; Step S26: Perform visual detection of the helmet laser point on the global brightness enhancement monitoring video based on the environmental interference light points, so as to obtain each helmet laser point in the construction site.

5. The laser positioning and marking method for helmet video acquisition according to claim 1, characterized in that: The specific steps of step S3 are: Step S31: Marking construction site feature points at multiple angles based on the construction site's omnidirectional surveillance video to generate key construction site feature points at different angles; Step S32: performing multi-directional matching on key feature points of the construction site at different angles, thereby generating key feature point position matching data; Step S33: performing video stream global posture correction on the key feature point position matching data to obtain global posture correction parameters at multiple angles; Step S34: Calculate the spatial position of the multiple cameras on the construction site and extract the spatial position coordinates of each camera; Step S35: fitting the three-dimensional environment of the construction site according to the spatial position coordinates of each camera to construct a three-dimensional environment map of the construction site; Step S36: Perform panoramic fusion reconstruction on the construction site three-dimensional environment map based on the global posture correction parameters of multiple angles to construct a panoramic construction site three-dimensional model.

6. The laser positioning and marking method for helmet video acquisition according to claim 1, characterized in that: The specific steps of step S4 are: Step S41: Perform multi-angle detection and identification on each safety helmet laser point in the construction site to obtain laser point detection parameters at all angles; Step S42: Calculate the laser point coordinates according to the laser point detection parameters at all angles to obtain the laser point coordinates at each angle; Step S43: performing visual positioning difference calculation on the laser point coordinates at each angle according to the panoramic construction site three-dimensional model to generate the laser point parallax parameters at each angle; Step S44: Perform stereoscopic vision positioning compensation based on the parallax parameters of the laser point at each angle, so as to obtain the three-dimensional precise positioning coordinates of each laser point.

7. The laser positioning and marking method for helmet video acquisition according to claim 1, characterized in that: The specific steps of step S5 are: Step S51: performing time-series frame analysis on the all-round monitoring video of the construction site to obtain time-series frame image change characteristics; Step S52: performing dynamic optical flow tracking on the three-dimensional precise positioning coordinates of each laser point according to the change characteristics of the time-series frame image to obtain the optical flow change vector of each laser point; Step S53: identifying the optical flow change direction of the optical flow change vector of each laser point, and generating the optical flow change direction of the laser point; Step S54: calculating the optical flow fluctuation amplitude of the optical flow change vector of each laser point to generate the optical flow fluctuation amplitude of the laser point; Step S55: perform real-time three-dimensional position change estimation based on the change direction of the laser point optical flow and the fluctuation amplitude of the laser point optical flow, and construct a dynamic three-dimensional movement trajectory of each laser point.

8. The laser positioning and marking method for helmet video acquisition according to claim 1, characterized in that: The specific steps of step S6 are: Step S61: performing dynamic movement trajectory rendering on the panoramic construction site three-dimensional model according to the dynamic three-dimensional movement trajectory of each laser point, and constructing a dynamic trajectory construction site rendering model; Step S62: performing segment-by-segment movement evolution analysis on the dynamic trajectory construction site rendering model to obtain movement change characteristics of each segment of the trajectory; Step S63: predicting the worker's moving path based on the moving change characteristics of each trajectory, thereby obtaining the predicted moving path of each worker; Step S64: adjusting the real-time posture parameters of the multiple cameras on the construction site according to the predicted movement path of each worker to obtain real-time posture adjustment data; Step S65: Perform intelligent synchronous capture based on the real-time posture adjustment data to build an intelligent real-time positioning engine to perform real-time positioning and marking operations for safety helmets on the construction site.

9. A laser positioning and marking device for helmet video acquisition, characterized in that: The laser positioning marking method for performing the helmet video acquisition as claimed in claim 1 comprises: The video enhancement module is used to collect all-round monitoring videos of the construction site in real time based on multiple cameras on the construction site; perform delay calculation between adjacent frames of the all-round monitoring videos of the construction site, and perform global brightness enhancement to obtain a global brightness enhanced monitoring video; Laser point visual recognition module, used to identify bright spots on the construction site and visually detect laser points on helmets based on the global brightness enhancement monitoring video, so as to obtain each laser point of the helmet on the construction site; The panoramic fusion module is used to mark the construction site feature points from multiple angles based on the construction site's omnidirectional monitoring video, and to perform panoramic fusion reconstruction to build a panoramic construction site 3D model; The visual positioning compensation module is used to calculate the visual positioning difference of each helmet laser point in the construction site based on the panoramic three-dimensional model of the construction site, and perform stereoscopic visual positioning compensation to obtain the three-dimensional precise positioning coordinates of each laser point; Dynamic optical flow tracking module, which is used to perform dynamic optical flow tracking on the all-round monitoring video of the construction site based on the three-dimensional precise positioning coordinates of each laser point, and to perform real-time three-dimensional position change estimation to construct the dynamic three-dimensional movement trajectory of each laser point; The synchronous tracking module is used to predict the worker's movement path according to the dynamic three-dimensional movement trajectory of each laser point, and to adjust the posture parameters of the multiple cameras on the construction site in real time to build an intelligent real-time positioning engine.

10. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the laser positioning and marking method for helmet video acquisition according to any one of claims 1 to 8 are implemented.

Images