Method for measuring spatial distance between hazard source and target using monocular vision

Abstract

The present invention discloses a method for measuring the spatial distance between a hazard source and a target using monocular vision, and relates to the field of distance measurement. The method comprises: deploying front-end acquisition equipment; controlling a laser projection device to perform a full-area scan of a target scene, transmitting laser point cloud data back, controlling a visual camera to perform full-field scene acquisition, and transmitting image data back; constructing a scene space model using the laser point cloud data and image data; and determining the target spatial distance by using the scene space model, by matching the image space with the point cloud two-dimensional projection space, and by mapping the point cloud two-dimensional projection space with the point cloud three-dimensional space, to determine the target point cloud position and perform spatial distance calculation. This method solves the technical problem of insufficient accuracy in existing methods for measuring the spatial distance between hazard sources and targets using monocular vision, and achieves the technical effect of improving the accuracy of spatial distance measurement.

Description

Technical Field

[0001] The present application relates to the field of distance measurement, and in particular to a method for measuring the spatial distance between a hazard source and a target using monocular vision. Background Art

[0002] In the operation, maintenance, and monitoring of power transmission lines, accurately measuring the spatial distance between hazardous sources and transmission lines helps assess potential safety hazards. Currently, a common approach is to extract depth information from a single image using complex image processing algorithms and geometric transformations, and then match it with three-dimensional point cloud data. However, in practical applications, due to factors such as lighting conditions, surface texture, and occlusion, a single image cannot directly provide sufficient depth information. This results in parallax errors when converting two-dimensional target positions to three-dimensional coordinates, which in turn affects the accuracy of distance measurements.

[0003] In the current related technologies, there is a technical problem of insufficient accuracy in measuring the spatial distance between the hazard source and the target using monocular vision. Summary of the Invention

[0004] The present application provides a method for measuring the spatial distance between a hazard source and a target using monocular vision, obtains laser point cloud data by scanning the target scene, captures a single image to obtain image data, combines the laser point cloud and image data, establishes a scene space model including three-dimensional space, two-dimensional projection space and image space, matches the image space with the two-dimensional projection space of the point cloud in the model, and then maps the two-dimensional projection space of the point cloud to the three-dimensional space of the point cloud to determine the target point cloud position. Based on the determined target point cloud position, spatial distance measurement is performed to obtain the target spatial distance and other technical means, thereby achieving the technical effect of improving the accuracy of spatial distance measurement.

[0005] The present application provides a method for measuring the spatial distance between a hazardous source and a target using monocular vision, including: deploying a front-end acquisition device, wherein the front-end acquisition device includes a laser projection device and a visual camera; controlling the laser projection device to perform a full-area scan of the target scene, returning laser point cloud data, controlling the visual camera to perform full-field scene acquisition, and returning image data, wherein the image data is a single image; constructing a scene space model using the laser point cloud data and the image data, wherein the scene space model includes point cloud three-dimensional space-point cloud two-dimensional projection space-image space; through the scene space model, based on the matching of the image space-point cloud two-dimensional projection space, using point cloud mapping of the point cloud two-dimensional projection space-point cloud three-dimensional space, determining the target point cloud position and performing spatial distance measurement to determine the target spatial distance, wherein the target point cloud position is the point cloud position of the hazardous source and the hazardous target.

[0006] In a possible implementation, the scene space model is constructed by performing the following processing: traversing the laser point cloud data to identify the point cloud coordinate data and scene feature data of each laser point cloud; constructing the point cloud three-dimensional space based on the point cloud coordinate data, wherein the point cloud three-dimensional space represents the point cloud spatial position; combining the scene feature data and taking the point cloud three-dimensional space as a reference, constructing a point cloud two-dimensional projection space, wherein the point cloud two-dimensional projection space uses the field of view coordinate system of the image data as the projection surface; constructing an image space based on the image data, hierarchically integrating the point cloud three-dimensional space-point cloud two-dimensional projection space-image space to determine the scene space model.

[0007] In a possible implementation, in combination with the scene feature data, the point cloud three-dimensional space is used as a reference to construct a point cloud two-dimensional projection space, and the following processing is performed: based on the acquisition parameters of the visual camera, the field of view coordinate system of the image data is determined; according to the field of view coordinate system, the projection surface is constructed, and a point cloud projection from the point cloud three-dimensional space to the projection surface is established to determine the point cloud two-dimensional projection; the scene feature data is traversed to determine the feature descriptor of each laser point cloud, the point cloud two-dimensional projection is projected point cloud identification, and the point cloud two-dimensional projection space is generated.

[0008] In a possible implementation, the target point cloud position is determined by performing the following processing: based on the image space, the hazard source and the hazard target are framed; for the hazard source and the hazard target, matching is performed based on the point cloud two-dimensional projection space to determine the projection matching result, which includes position matching and feature matching; using the point cloud mapping of the point cloud two-dimensional projection space to the point cloud three-dimensional space, the three-dimensional mapping point cloud of the projection matching result is determined as the target point cloud position.

[0009] In a possible implementation, the hazard sources and hazard targets are framed and the following processing is performed: for the target scenario, based on historical risk control data, the risk database is mined; the image data is transmitted back and filtered and denoised to determine a pre-processed image; the risk database is traversed, the pre-processed image is matched, and the hazard sources and hazard targets are located.

[0010] In a possible implementation, matching based on the two-dimensional projection space of the point cloud is performed, and the following processing is performed: based on the image space, the hazard source is framed and the first coordinate is determined, and the hazard target is framed and the second coordinate is determined; based on the first coordinate and the second coordinate, the coordinate positioning of the two-dimensional projection space of the point cloud is performed to determine the first positioning coordinate; the pixel features of the hazard source and the hazard target are identified, and feature matching based on feature descriptors is performed to determine the second matching coordinate; the first positioning coordinate and the second matching coordinate are collated and fused to determine the projection matching result.

[0011] In a possible implementation, the spatial distance measurement is performed by performing the following processing: identifying the target point cloud position, determining the first point cloud range and the second point cloud range, wherein the first point cloud range is the point cloud distribution of the hazard source, and the second point cloud range is the point cloud distribution of the hazardous target; based on the first point cloud range and the second point cloud range, determining a spatial ranging point cloud pair, wherein the spatial ranging point cloud is a boundary point cloud pair; measuring the relative distance of the spatial ranging point cloud pair to determine the target spatial distance, wherein the relative distance includes straight-line distance, horizontal distance and vertical distance.

[0012] In a possible implementation, the spatial ranging point cloud pair is determined by performing the following processing: determining the center point cloud of the first point cloud range and the second point cloud range, and using the boundary point cloud on the line connecting the center point clouds as the first boundary point cloud and the second boundary point cloud; using the first boundary point cloud as a reference point, performing boundary extension calculation based on the reference point, and determining the first valid point cloud of the hazard source, wherein the boundary geometric features are determined based on the point cloud normal vector and curvature; using the second boundary point cloud as a reference point, performing boundary extension calculation based on the reference point, and determining the second valid point cloud of the hazard target; and using the first valid point cloud and the second valid point cloud as the boundary point cloud pair.

[0013] The method for measuring the spatial distance between a hazardous source and a target using monocular vision proposed in this application is to first deploy a front-end acquisition device, wherein the front-end acquisition device includes a laser projection device and a visual camera, and then control the laser projection device to perform a full-area scan of the target scene, and transmit back the laser point cloud data, and control the visual camera to perform full-field scene acquisition, and transmit back the image data, wherein the image data is a single image, and then use the laser point cloud data and the image data to construct a scene space model, wherein the scene space model includes point cloud three-dimensional space-point cloud two-dimensional projection space-image space, and finally use the scene space model to match the image space-point cloud two-dimensional projection space, and use the point cloud mapping of the point cloud two-dimensional projection space-point cloud three-dimensional space to determine the target point cloud position and perform spatial distance measurement to determine the target spatial distance, wherein the target point cloud position is the point cloud position of the hazardous source and the hazardous target, thereby achieving the technical effect of improving the accuracy of spatial distance measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention are briefly introduced below. Flowcharts are used in this application to illustrate the operations performed by the methods according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in precise order. Instead, various steps may be processed in reverse order or simultaneously as needed. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0015] Figure 1 A flow chart of a method for measuring the spatial distance between a hazard source and a target using monocular vision provided in an embodiment of the present application.

[0016] Figure 2 A schematic diagram of the process of constructing a scene space model in the method for measuring the spatial distance between a hazard source and a target using monocular vision provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below.

[0018] In order to make the purpose, technical solutions and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be regarded as limiting this application. All other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0019] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict, and the terms “first\second” involved are merely used to distinguish similar objects and do not represent a specific ordering of the objects. The terms “including” and “having” and any variations are intended to cover non-exclusive inclusions. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or modules that are not clearly listed or that are inherent to these processes, methods, products or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. The terms used herein are for the purpose of describing the embodiments of this application only.

[0020] The embodiment of the present application provides a method for measuring the spatial distance between a danger source and a target using monocular vision, such as Figure 1 As shown, the method includes:

[0021] Step S100 , deploying front-end acquisition equipment, wherein the front-end acquisition equipment includes a laser projection device and a visual camera.

[0022] Specifically, front-end acquisition equipment is deployed, including laser projection devices and visual cameras. The laser projection device is a device that emits a laser beam and receives the reflected laser signal, generating laser point cloud data. The visual camera is used to capture image data of the target scene. It uses a single-frame camera and can capture a two-dimensional image of the scene.

[0023] Step S200, controlling the laser projection device to perform full-area scanning of the target scene, returning laser point cloud data, controlling the visual camera to perform full-field scene acquisition, and returning image data, wherein the image data is a single image.

[0024] Specifically, a laser projection device is controlled to emit a laser beam, scanning the entire target scene. During the scanning process, the laser beam is reflected back, forming a laser point cloud. This data contains the 3D coordinate information of each point in the scene. Simultaneously, while the laser projection device is scanning, a visual camera is controlled to capture the entire field of view of the scene, capturing 2D image data.

[0025] Step S300: constructing a scene space model using the laser point cloud data and the image data, wherein the scene space model includes point cloud three-dimensional space-point cloud two-dimensional projection space-image space.

[0026] Specifically, laser point cloud data and image data are used to construct a scene space model through image matching, feature point extraction and matching. The scene space model is deployed on the front-end intelligent monitoring equipment, and the intelligent monitoring equipment can perform hazard source identification and ranging processing. The scene space model is a model that includes the correspondence between the point cloud three-dimensional space, the point cloud two-dimensional projection space and the image space. Among them, the point cloud three-dimensional space is a three-dimensional space composed of laser point cloud data, which contains the three-dimensional coordinate information of each point in the scene. The point cloud two-dimensional projection space is a space formed by projecting the points in the point cloud three-dimensional space onto a two-dimensional plane. This projection space is a bridge connecting the point cloud three-dimensional space and the image space. The image space is a space composed of image data captured by a visual camera, which contains the two-dimensional image information of the scene.

[0027] like Figure 2As shown, in a possible implementation, the scene space model is constructed, and step S300 further includes step S310, traversing the laser point cloud data and identifying the point cloud coordinate data and scene feature data of each laser point cloud. Specifically, the laser point cloud data returned by the laser projection device is processed point by point by programming (such as using Python, C++ and other languages). The point cloud data contains the three-dimensional coordinates (X, Y, Z) and intensity information of each laser point. During the traversal process, the three-dimensional coordinates of each laser point are identified, and these coordinates are the basis for constructing the three-dimensional space of the point cloud. At the same time, based on the density, shape, color and other characteristics of the point cloud, specific objects or feature areas in the scene are identified.

[0028] Step S320 constructs the point cloud 3D space based on the point cloud coordinate data, where the point cloud 3D space represents the spatial location of the point cloud. Specifically, a 3D point cloud model of the target scene is constructed based on the point cloud coordinate data identified in step S310 using 3D modeling software or a programming library (such as PCL or Open3D). This model represents the 3D structure and shape of the scene and is composed of countless 3D points, each representing a specific location in the scene.

[0029] In step S330, a two-dimensional point cloud projection space is constructed based on the scene feature data and the three-dimensional point cloud space. The two-dimensional point cloud projection space uses the field of view coordinate system of the image data as the projection surface. Specifically, a two-dimensional projection surface is determined based on the field of view coordinate system of the image data captured by the visual camera. This projection surface is the mapping plane from the three-dimensional point cloud space to the image space. Then, based on the point cloud coordinates and scene feature data in the three-dimensional point cloud space, these points are mapped onto the projection surface according to projection rules (such as orthographic projection, perspective projection, etc.), forming a two-dimensional point cloud projection space. This space retains the three-dimensional position information of the point cloud but represents it in a two-dimensional form.

[0030] Step S340 is to construct an image space using the image data, hierarchically integrate the point cloud three-dimensional space, point cloud two-dimensional projection space, and image space to determine the scene space model. Specifically, image processing software or a programming library (such as OpenCV) is used to construct an image space based on the image data captured by the visual camera. This space is a two-dimensional space that represents the pixel information and color information in the image. The point cloud three-dimensional space, the point cloud two-dimensional projection space, and the image space are hierarchically integrated to form a complete scene space model. This model can reflect the three-dimensional structure and shape of the target scene, as well as the feature information in the image. After the integration is completed, the scene space model is saved through programming or modeling software for applications such as spatial distance measurement and hazard source identification. This implementation method, by introducing the point cloud two-dimensional projection space, can map the target position in the image onto the projection surface, thereby avoiding three-dimensional ranging errors caused by differences in viewing angles.

[0031] In one possible implementation, a two-dimensional point cloud projection space is constructed based on the three-dimensional point cloud space in combination with the scene feature data. Step S330 further includes step S331, where a field of view coordinate system for the image data is determined based on the acquisition parameters of the visual camera. Specifically, the acquisition parameters of the visual camera include the focal length, optical center position, pixel size, etc., which determine the correspondence between the image captured by the camera and the actual scene. Based on the camera acquisition parameters, the field of view coordinate system of the image is calculated. This coordinate system is a two-dimensional coordinate system used to represent the correspondence between pixel positions in the image and spatial positions in the actual scene.

[0032] Step S332, construct the projection surface according to the field of view coordinate system, establish the point cloud projection from the point cloud three-dimensional space to the projection surface, and determine the two-dimensional projection of the point cloud. Specifically, construct the projection surface, which is a two-dimensional plane for receiving points projected from the point cloud three-dimensional space. This plane corresponds to the field of view coordinate system of the image, that is, each point on the projection surface corresponds to a pixel in the image. The points in the point cloud three-dimensional space are projected onto the projection surface according to the projection rules to form a two-dimensional projection of the point cloud. After the projection is completed, each point on the projection surface corresponds to a point in the point cloud three-dimensional space, but at this time these points have been converted into two-dimensional coordinates.

[0033] Step S333 traverses the scene feature data, determines the feature descriptors for each laser point cloud, identifies the two-dimensional projection of the point cloud, and generates the two-dimensional projection space of the point cloud. Specifically, the scene feature data is traversed. These feature data include information such as the shape, color, and texture of the point cloud. A feature descriptor is determined for each laser point cloud, which is used to uniquely identify the point cloud. The feature descriptor can be a combination of one or more shape features, color features, and texture features of the point cloud. Using feature matching algorithms such as SIFT and SURF, the feature descriptors of each laser point cloud are associated with their corresponding points in the two-dimensional projection of the point cloud to form a projection point cloud identification. After the projection point cloud identification is completed, a space containing the positions of all laser point clouds on the two-dimensional projection surface and their feature descriptors is obtained, namely the two-dimensional projection space of the point cloud. This space contains both the position information and feature information of the point clouds, which can be used to match the pixel features of the image. This implementation method constructs a two-dimensional projection space for point clouds, which can project point clouds in three-dimensional space onto a two-dimensional plane while retaining their feature information. In this way, image processing algorithms can be used to perform feature matching on the two-dimensional plane, thereby improving the accuracy of hazard source identification and spatial distance measurement.

[0034] In step S400, the target point cloud position is determined and the spatial distance is measured through the scene space model based on the matching of the image space-point cloud two-dimensional projection space and the point cloud mapping of the point cloud two-dimensional projection space-point cloud three-dimensional space to determine the target spatial distance, wherein the target point cloud position is the point cloud position of the danger source and the danger target.

[0035] Specifically, first, the image feature points of the hazard source and the hazard target are identified in the image space. Then, the corresponding relationship in the scene space model is used to map these feature points to the point cloud two-dimensional projection space. Next, the corresponding relationship between the point cloud two-dimensional projection space and the point cloud three-dimensional space is used to further map the feature points mapped to the point cloud two-dimensional projection space to the point cloud three-dimensional space, thereby determining the point cloud position of the hazard source and the hazard target. Finally, based on the point cloud position of the hazard source and the hazard target in the point cloud three-dimensional space, the spatial distance between them is calculated. The embodiment of the present application adopts the method of scanning the target scene to obtain laser point cloud data, taking a single image to obtain image data, combining the laser point cloud and the image data, establishing a scene space model including three-dimensional space, two-dimensional projection space and image space, matching the image space with the point cloud two-dimensional projection space in the model, and then mapping the point cloud two-dimensional projection space to the point cloud three-dimensional space to determine the target point cloud position, and performing spatial distance measurement based on the determined target point cloud position to obtain the target space distance and other technical means, thereby achieving the technical effect of improving the accuracy of spatial distance measurement.

[0036] In a possible implementation, the target point cloud position is determined, and step S400 further includes step S410, which frames the hazard source and the dangerous target based on the image space. Specifically, the image space is a space composed of image data collected by a visual camera, which contains two-dimensional pixel information in the scene. In the image space, image processing technology (such as target detection, etc.) is used to identify and frame the positions of the hazard source and the dangerous target. Specifically, it includes pre-processing the image (such as denoising, contrast enhancement, etc.), and then applying a target detection algorithm (such as a convolutional neural network, a support vector machine, etc.) to identify the target object and mark it in the image with a rectangular frame or other shape. Among them, the hazard source is the source that may cause danger or accidents, such as tree growth, construction activities, etc. Dangerous targets are objects that need to be monitored or protected, such as power transmission lines and key areas around them.

[0037] Step S420, for the hazardous source and the hazardous target, performs matching based on the point cloud two-dimensional projection space, and determines the projection matching result, which includes position matching and feature matching. Specifically, the point cloud two-dimensional projection space is a space formed by the projection of the laser point cloud data in the image two-dimensional coordinate system, which contains the two-dimensional projection information in the scene. In the point cloud two-dimensional projection space, a feature matching algorithm is used to find the projection points corresponding to the hazardous source and hazardous target framed in the image space. Specifically, it includes extracting the feature descriptors of the projection points (such as shape, color, texture, etc.), and then performing feature matching in the projection space to find the most similar projection points. After the matching is completed, a projection matching result including position matching and feature matching is obtained. Position matching refers to the position of the projection point in the projection space corresponding to the target position framed in the image space; feature matching refers to the feature descriptor of the projection point matching the feature descriptor of the target object in the image space.

[0038] In step S430, the three-dimensional mapping point cloud of the projection matching result is determined as the target point cloud location using the point cloud mapping from the two-dimensional projection space of the point cloud to the three-dimensional space of the point cloud. Specifically, the point cloud mapping from the two-dimensional projection space of the point cloud to the three-dimensional space of the point cloud establishes a mapping relationship between the two-dimensional projection space of the point cloud and the three-dimensional space of the point cloud, so that each point in the projection space can find a corresponding point in the three-dimensional space. Using the point cloud mapping relationship, the projection points in the projection matching result are mapped into the three-dimensional space of the point cloud to obtain the corresponding three-dimensional point clouds. These three-dimensional point clouds are the target point cloud locations, representing the actual locations of the hazardous sources and hazardous targets in three-dimensional space. This implementation method constructs a scene space model, combines laser point cloud data with image data, and utilizes the matching relationship between the two-dimensional projection space of the point cloud and the image space to achieve rapid identification and location of hazardous sources and hazardous targets. Then, using the point cloud mapping relationship, the projection matching result is mapped into the three-dimensional space of the point cloud to obtain the target point cloud location, thereby improving the accuracy of determining the locations of hazardous sources and hazardous targets in three-dimensional space.

[0039] In one possible implementation, the hazard sources and hazard targets are framed, and step S410 further includes step S411, mining the risk database based on historical risk control data for the target scenario. Specifically, historical risk control data is collected from various sources, and the historical risk control data includes past line fault records, inspection reports, meteorological data, geographic information, etc. These data contain various factors that lead to line failures. The collected data is pre-processed by cleaning, deduplication, format unification, etc. to improve the efficiency and quality of data mining. Feature extraction algorithms are used to extract features related to line faults from historical data, such as fault type, fault location, environmental factors, etc. The extracted feature information is integrated into the risk database to form a structured knowledge base.

[0040] In step S412, the image data is returned and filtered and denoised to determine a preprocessed image. Specifically, filtering and denoising are performed on the returned image data to improve image quality. This includes using filtering algorithms such as median filtering and Gaussian filtering to remove noise and interference from the image. Contrast enhancement and sharpening are also performed on the image to improve image clarity and legibility.

[0041] Step S413, traverse the risk database, match the pre-processed image, and locate the dangerous sources and dangerous targets. Specifically, extract template images of known dangerous sources and dangerous targets from the risk database and match them with the pre-processed image. Use a feature point matching algorithm to search for feature points similar to the template image in the image to locate the dangerous sources and dangerous targets. Combined with deep learning algorithms (such as convolutional neural networks (CNNs)), identify and classify targets in the image to further confirm the identity of the dangerous sources and dangerous targets. This implementation method improves the accuracy of identifying dangerous sources and dangerous targets in images through the mining of historical risk control data and image preprocessing.

[0042] In one possible implementation, matching is performed based on the two-dimensional projection space of the point cloud. Step S420 further includes step S421, where, based on the image space, the hazard source is framed and a first coordinate is determined, and the hazard target is framed and a second coordinate is determined. Specifically, the approximate areas of the hazard source and the hazard target are framed in the image space. Points in these areas are extracted as the first coordinates (hazard source coordinates) and the second coordinates (hazard target coordinates).

[0043] Step S422: Based on the first coordinate and the second coordinate, coordinate positioning is performed in the two-dimensional projection space of the point cloud to determine a first positioning coordinate. Specifically, the first coordinate and the second coordinate are mapped into the two-dimensional projection space of the point cloud using a projection plane to obtain the first positioning coordinate, i.e., the first positioning coordinate includes the projected coordinates of the first coordinate and the second coordinate.

[0044] Step S423, identify the pixel features of the dangerous source and the dangerous target, perform feature matching based on feature descriptors, and determine the second matching coordinates. Specifically, extract the feature descriptors of the dangerous source and the dangerous target from the image. These feature descriptors can be local feature descriptors such as SIFT, SURF, ORB, or global feature descriptors such as HOG, LBP. Then, search for point cloud features that match these feature descriptors in the two-dimensional projection space of the point cloud, and determine the matching results by calculating the similarity between the feature descriptors (such as Euclidean distance, Hamming distance, etc.). The matching results include the coordinates of the point cloud features corresponding to the dangerous source and the dangerous target on the projection surface, that is, the second matching coordinates, and the second matching coordinates also include the matching coordinates of the first coordinates and the second coordinates.

[0045] Step S424, proofread and fuse the first positioning coordinates and the second matching coordinates to determine the projection matching result. Specifically, compare the difference between the first positioning coordinates and the second matching coordinates to evaluate their accuracy and consistency. If the difference is large, it is necessary to re-perform feature matching or adjust the parameters of the coordinate mapping. The coordinates that are considered accurate and reliable after proofreading are fused to obtain a final coordinate that combines the advantages of the two methods, and this final coordinate is used as the basis for subsequent spatial distance measurement. This implementation method can accurately determine the position of dangerous sources and dangerous targets in the two-dimensional projection space of the point cloud by combining the two methods of coordinate positioning and feature matching. Coordinate positioning provides preliminary position information, and feature matching verifies and corrects this position information by comparing feature descriptors, thereby improving the accuracy of positioning.

[0046] In one possible implementation, the spatial distance measurement step S400 further includes step S440, identifying the target point cloud position, determining a first point cloud range and a second point cloud range, wherein the first point cloud range is the point cloud distribution of the hazard source, and the second point cloud range is the point cloud distribution of the dangerous target. Specifically, the target point cloud position is the specific position of the hazard source and the dangerous target in the three-dimensional space of the point cloud, and their range is determined by traversing these point clouds, specifically including calculating the boundary, center of mass, size and other parameters of the point cloud. The point cloud range refers to the distribution area of ​​the point clouds of the hazard source and the dangerous target in the three-dimensional space.

[0047] Step S450 determines a spatial ranging point cloud pair based on the first point cloud range and the second point cloud range, where the spatial ranging point cloud is a boundary point cloud pair. Specifically, boundary point clouds (the outer point clouds of the hazard source and hazard target point cloud ranges, defining the target boundaries) are extracted from the point cloud ranges of the hazard source and hazard target by calculating the convex hull, minimum bounding box, minimum bounding sphere, or other methods of the point cloud. By calculating the Euclidean distance between each pair of points, the closest point pair between the two boundary point clouds is found as the spatial ranging point cloud pair.

[0048] Step S460, measure the relative distance of the spatial ranging point cloud pair to determine the target spatial distance, wherein the relative distance includes straight-line distance, horizontal distance and vertical distance. Specifically, the straight-line distance between the spatial ranging point cloud pairs is calculated by calculating the Euclidean distance between the two points. The horizontal distance and the vertical distance are calculated by decomposing the straight-line distance into horizontal and vertical components. Finally, the calculated relative distance is output as the target spatial distance for subsequent risk assessment, safety monitoring and other applications. This implementation method uses boundary point cloud pairs for ranging. The boundary point cloud is relatively stable and less affected by noise, thereby improving the accuracy of the measurement.

[0049] In a possible implementation, the step of determining the spatial ranging point cloud pair S450 further includes a step S451, determining the center point cloud of the first point cloud range and the second point cloud range, and using the boundary point cloud on the line connecting the center point clouds as the first boundary point cloud and the second boundary point cloud. Specifically, by calculating the average coordinates of all points within the point cloud range, the center point cloud of the hazard source and hazard target point cloud range is calculated, and the center point cloud represents the geometric center of the hazard source and the hazard target. Then, based on the line connecting the center point clouds, the boundary point clouds on the line are found as the first boundary point cloud and the second boundary point cloud. These boundary point clouds are located at the edge of the hazard source and hazard target point cloud ranges, and are the key points connecting the two target point clouds.

[0050] Step S452 uses the first boundary point cloud as a reference point and performs boundary extension calculation based on the reference point to determine the first valid point cloud for the hazard source. The boundary geometric features are determined based on the point cloud normal and curvature. Specifically, using the first boundary point cloud as a reference point, boundary extension calculation is performed using geometric features such as the point cloud normal and curvature to determine the range of the valid point cloud. Within the boundary extension range, valid point clouds are screened based on specific conditions (such as distance and angle) to determine the first valid point cloud for the hazard source. The valid point cloud is obtained through boundary extension calculation and is used for spatial distance measurement. The normal vector is a vector that describes the direction of the point cloud surface, and the curvature is a quantity that describes the degree of curvature of the point cloud surface. By calculating the point cloud normal and curvature, the geometric features of the point cloud can be understood, thereby performing boundary extension calculation. Boundary extension calculation is an algorithm based on the geometric features of the point cloud and is used to determine the boundaries of the point cloud range.

[0051] Step S453: Using the second boundary point cloud as a reference point, boundary extension calculation is performed based on the reference point to determine a second valid point cloud of the dangerous target. Specifically, using the second boundary point cloud as a reference point, boundary extension calculation is performed to determine a second valid point cloud of the dangerous target.

[0052] In step S454, the first valid point cloud and the second valid point cloud are used as the boundary point cloud pair. Specifically, the first valid point cloud and the second valid point cloud are used as the boundary point cloud pair for spatial distance measurement. Hazard source or target point clouds may have irregular distributions, so the boundary point cloud may not directly represent the shortest distance between the two targets. This implementation method, by determining the valid point cloud, can reduce the influence of these factors, more accurately find the shortest path between the two targets, and improve the accuracy and reliability of the calculation results.

[0053] The above specific embodiments do not constitute a limitation to the scope of protection of this application. It should be understood by those skilled in the art that various modifications, combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of this application should be included in the scope of protection of this application. In some cases, the actions or steps recorded in this application can be performed in an order different from that in the embodiments and can still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A method for measuring the spatial distance between a hazard source and a target using monocular vision, characterized in that: The method comprises: Deploy front-end acquisition equipment, wherein the front-end acquisition equipment includes a laser projection device and a visual camera; Controlling the laser projection device to perform full-area scanning of the target scene and transmitting back laser point cloud data; controlling the visual camera to perform full-field scene acquisition and transmitting back image data, wherein the image data is a single image; Constructing a scene space model using the laser point cloud data and the image data, wherein the scene space model includes a point cloud three-dimensional space, a point cloud two-dimensional projection space, and an image space; By using the scene space model, based on the matching of the image space and the point cloud two-dimensional projection space, and by using the point cloud mapping of the point cloud two-dimensional projection space to the point cloud three-dimensional space, the target point cloud position is determined and the spatial distance is measured to determine the target spatial distance, wherein the target point cloud position is the point cloud position of the hazard source and the hazard target; The step of constructing the scene space model includes: Traversing the laser point cloud data, identifying point cloud coordinate data and scene feature data of each laser point cloud; Based on the point cloud coordinate data, construct the point cloud three-dimensional space, wherein the point cloud three-dimensional space represents the point cloud spatial position; In combination with the scene feature data, the point cloud three-dimensional space is used as a reference to construct a point cloud two-dimensional projection space, wherein the point cloud two-dimensional projection space uses the field of view coordinate system of the image data as a projection surface; Constructing an image space using the image data, hierarchically integrating the point cloud three-dimensional space-point cloud two-dimensional projection space-image space to determine the scene space model; The step of combining the scene feature data and taking the point cloud three-dimensional space as a reference to construct a point cloud two-dimensional projection space includes: Determining a field of view coordinate system of the image data based on acquisition parameters of the visual camera; Constructing the projection surface according to the field of view coordinate system, establishing a point cloud projection from the point cloud three-dimensional space to the projection surface, and determining a two-dimensional projection of the point cloud; Traversing the scene feature data, determining the feature descriptor of each laser point cloud, performing projection point cloud identification on the two-dimensional projection of the point cloud, and generating the two-dimensional projection space of the point cloud; Wherein, determining the target point cloud position includes: Based on the image space, framing the danger source and the danger target; For the hazard source and the hazard target, performing matching based on the two-dimensional projection space of the point cloud to determine a projection matching result, which includes position matching and feature matching; The three-dimensional mapping point cloud of the projection matching result is determined as the target point cloud position by using the point cloud mapping between the two-dimensional projection space of the point cloud and the three-dimensional space of the point cloud.

2. The method for measuring the spatial distance between a danger source and a target using monocular vision according to claim 1, wherein: The framing of hazard sources and hazard targets includes: For the target scenario, mining the risk database based on historical risk control data; Returning the image data and performing filtering and noise reduction to determine a pre-processed image; The risk database is traversed, the pre-processed images are matched, and the hazard sources and hazard targets are located.

3. The method for measuring the spatial distance between a danger source and a target using monocular vision according to claim 1, wherein: Performing matching based on the two-dimensional projection space of the point cloud, including: Based on the image space, framing the danger source and determining a first coordinate, framing the danger target and determining a second coordinate; Performing coordinate positioning on the two-dimensional projection space of the point cloud according to the first coordinate and the second coordinate to determine a first positioning coordinate; Identifying pixel features of the danger source and the dangerous target, performing feature matching based on feature descriptors, and determining second matching coordinates; The first positioning coordinates and the second matching coordinates are collated and merged to determine the projection matching result.

4. The method for measuring the spatial distance between a danger source and a target using monocular vision according to claim 1, wherein: The spatial distance measurement includes: Identify the target point cloud position, and determine a first point cloud range and a second point cloud range, wherein the first point cloud range is the point cloud distribution of the hazard source, and the second point cloud range is the point cloud distribution of the hazard target; Determine a spatial ranging point cloud pair based on the first point cloud range and the second point cloud range, wherein the spatial ranging point cloud is a boundary point cloud pair; The relative distance of the spatial ranging point cloud pair is measured to determine the target spatial distance, wherein the relative distance includes straight-line distance, horizontal distance and vertical distance.

5. The method for measuring the spatial distance between a danger source and a target using monocular vision according to claim 4, wherein: The determining of the spatial ranging point cloud pair includes: Determine the center point cloud of the first point cloud range and the second point cloud range, and use the boundary point clouds on the line connecting the center point clouds as the first boundary point cloud and the second boundary point cloud; Taking the first boundary point cloud as a reference point, performing boundary extension measurement based on the reference point to determine a first valid point cloud of the hazard source, wherein the boundary geometric features are determined based on the point cloud normal vector and curvature; Taking the second boundary point cloud as a reference point, performing boundary extension measurement based on the reference point to determine a second valid point cloud of the dangerous target; The first valid point cloud and the second valid point cloud are used as the boundary point cloud pair.

Images