A method and system for spatial localization of bridge fire sources based on multi-view stereo vision
By forming a set of candidate locations in bridge fires and performing spatial intersection screening, the problem of three-dimensional coordinate jumps caused by flame swaying and obstruction was solved, achieving stable spatial positioning of the fire source and accurate flame tracking and aiming of fire extinguishing devices.
Patent Information
- Application Number
- CN202611628029.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-21
- Publication Date
- 2026-11-17
AI Technical Summary
In vehicle fires in open spaces on bridges, when calculating the spatial coordinates of the fire source based on multi-view stereo vision, the problem of three-dimensional spatial coordinate jumps caused by the flames being swayed by crosswinds or blocked by vehicles makes it difficult to stably locate the fire source.
By forming a first set of candidate locations and a second set of candidate locations, the candidate locations are paired and the spatial intersection distance is filtered using the device attitude and spatial geometric relationship of the observation perspective to determine the corresponding location of the target. When the intersection relationship does not meet the conditions, the three-dimensional location of the fire source is recalculated, and the spatial positioning of the fire source is performed by combining the parallax of the fire source image and the device attitude.
It effectively avoids parallax errors and 3D coordinate offsets caused by flame swaying and obstruction, improves the reliability of 3D fire source location calculation, provides stable fire source spatial positioning results, and supports fire tracking and accurate aiming of jet extinguishing devices.
Smart Images

Figure CN123118751A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of computer vision image processing and bridge fire monitoring and location technology, and in particular to a method and system for spatial location of bridge fire sources based on multi-view stereo vision. Background Technology
[0002] In monitoring vehicle fires in open spaces on bridges, multi-view stereo vision technology can be used to acquire images of the fire scene from multiple observation angles. The spatial coordinates of the fire source can then be calculated based on image parallax for fire source tracking and intelligent jet targeting. This technology typically utilizes the positional differences between images from different viewpoints to obtain target depth information. It features non-contact operation and the ability to cover large open spaces, making it valuable for bridge fire monitoring and firefighting.
[0003] In vehicle fires in open spaces on bridges, when flames are swayed or stretched by crosswinds, or partially obscured by vehicles, the fire source areas identified from different observation angles may correspond to different actual parts of the flame. If image parallax is calculated directly based on the positional relationship of the fire source areas from different angles, the calculated 3D spatial coordinates of the same fire source may suddenly shift or jump at consecutive moments. For example, after a truck catches fire on the bridge, crosswinds cause the flames to tilt to the upper left. From one observation angle, the base and upper part of the flame are visible; from another, the base of the flame is obscured by the vehicle body, and only the upper high-temperature area is identified. Calculating the parallax at the center of the fire source area at these times may result in a spatial intersection position that falls behind the vehicle or in an adjacent lane. As the obstruction weakens in the next moment, the positioning result returns to the vicinity of the burning vehicle, resulting in continuous jumps in the fire source coordinates. Users will see the fire source location changing between the wrong lane and near the burning vehicle, causing the intelligent jet aiming direction to swing accordingly, making it difficult to stably aim at the base of the flame, potentially missing the opportunity to extinguish the fire and affecting subsequent fire source tracking. Therefore, how to reduce the jump in the three-dimensional spatial coordinates of the same fire source during the process of calculating the spatial coordinates of the fire source based on multi-view stereo vision in the case of vehicle fires in open spaces of bridges is an urgent problem to be solved in this field. Summary of the Invention
[0004] To address the shortcomings of existing technologies, the present invention aims to provide a method for spatially locating bridge fire sources based on multi-view stereo vision. This method involves forming a first set of candidate locations and a second set of candidate locations based on the fire source regions from each observation perspective. The candidate locations are then paired and the spatial intersection distance is used to determine the corresponding location of the target based on the device orientation and spatial geometric relationships of the observation perspectives. The three-dimensional location of the fire source is calculated from the corresponding location of the target. When the spatial intersection relationship does not meet the spatial intersection conditions, the corresponding location of the target is re-determined, thereby achieving the technical effect of spatially locating bridge fire sources.
[0005] To achieve the above objectives, the present invention adopts the following technical solution:
[0006] In a first aspect, the present invention provides a method for spatial localization of bridge fire sources based on multi-view stereo vision, comprising:
[0007] A first set of candidate locations and a second set of candidate locations are formed based on the fire source area from each observation perspective;
[0008] Based on the device attitude and spatial geometric relationship of the observation perspective, the candidate positions in the first candidate position set and the second candidate position set are paired, and the corresponding positions of the target are determined by filtering according to the spatial intersection distance of the corresponding spatial observation direction of each paired position.
[0009] The fire source image parallax is determined from the location corresponding to the target, and the three-dimensional position of the fire source is calculated by combining the fire source image parallax, the spatial geometric relationship, and the device attitude.
[0010] When the spatial intersection relationship of the target location does not meet the spatial intersection conditions, the target location is re-determined based on the first candidate location set and the second candidate location set of the current positioning cycle, the three-dimensional location of the fire source is recalculated, and the spatial positioning result of the bridge fire source is formed.
[0011] Furthermore, the fire source area includes:
[0012] The infrared high-temperature region is mapped to the color image coordinate system according to the parameters corresponding to the image position. The minimum pixel distance from the pixel in the color flame region to the mapped infrared high-temperature region is calculated. Color flame pixels whose minimum pixel distance is not greater than the cross-band registration allowable distance are formed into the fire source region.
[0013] The cross-band registration allowable distance is determined based on the registration residual distribution of the corresponding pixel positions of the same calibration target at different working distances.
[0014] Further, the first candidate location set and the second candidate location set include:
[0015] Contour tracing and internal temperature mapping are performed on the corresponding fire source regions to obtain the fire source boundary pixel sequence and temperature pixel set;
[0016] The process of determining the interval of candidate locations is performed based on the size of the fire source area itself.
[0017] Based on the candidate position interval, boundary candidate positions are extracted from the fire source boundary pixel sequence, and internal candidate positions are extracted from the temperature pixel set. The boundary candidate positions and internal candidate positions are then written into the corresponding candidate position set.
[0018] Furthermore, the candidate position interval determination process includes:
[0019] Calculate the equivalent diameter of the current fire source area based on the current fire source area area, and determine the initial candidate position interval based on the equivalent diameter of the current fire source area;
[0020] Based on the initial candidate position interval, internal candidate positions are selected. When the number of internal candidate positions exceeds the upper limit, the candidate position interval is increased and internal candidate positions are reselected until the number of internal candidate positions meets the limit or the candidate position interval reaches the corresponding upper limit.
[0021] Furthermore, the screening includes:
[0022] Establish cross-viewpoint candidate position pairings according to candidate position categories, obtain the corresponding epipolar line of the first candidate position in another observation view based on the dual-target calibration parameters, and retain the pairings that are no more than the allowable distance of the epipolar line and have the same candidate position category as the corresponding epipolar line.
[0023] The paired positions are converted into spatial observation directions, the nearest point and spatial intersection distance of the two spatial observation directions are calculated, and a valid candidate correspondence set is formed based on the spatial intersection distance and the candidate intersection positions.
[0024] Furthermore, the spatial intersection distance includes:
[0025] The Euclidean distance between the first and second nearest points along two spatial observation directions is taken as the spatial intersection distance, and the midpoint of the line connecting the first and second nearest points is taken as the candidate intersection position.
[0026] A spatial intersection residual sample set is formed based on the repeated positioning data of known spatial coordinate test points within the target working distance range, and the spatial intersection distance threshold is determined based on the quantile value of the spatial intersection residual sample set.
[0027] Furthermore, the target location includes:
[0028] When the set of valid candidate correspondences contains more than two valid candidate correspondences, they are sorted according to spatial intersection distance.
[0029] When the difference in spatial intersection distance between adjacent valid candidate corresponding records is greater than the spatial intersection distance resolvable threshold, the valid candidate corresponding record with the smaller spatial intersection distance is selected.
[0030] When the difference in spatial intersection distance is not greater than the resolvable threshold of spatial intersection distance, the corresponding position of the target is determined based on the spatial distance between the candidate intersection position and the three-dimensional position of the effective fire source in the previous positioning cycle.
[0031] Furthermore, the resolvable threshold for spatial intersection distance includes:
[0032] Spatial intersection calculations are repeatedly performed on control points with the same known spatial location to form a repeated spatial intersection distance sequence;
[0033] Calculate the median of the repeated spatial intersection distance sequence and the absolute deviation between each spatial intersection distance and the median, and determine the resolvable threshold of the spatial intersection distance based on the distribution of the absolute deviation.
[0034] Furthermore, the three-dimensional location of the fire source includes:
[0035] The positions corresponding to the first and second observation angles in the target positions are determined as the first target position and the second target position, respectively. The first target spatial observation direction and the second target spatial observation direction are formed according to the first target position and the second target spatial observation direction, respectively. The closest point of the first target and the closest point of the second target in the two target spatial observation directions are obtained. The midpoint of the line connecting the closest point of the first target and the closest point of the second target is determined as the three-dimensional position of the fire source.
[0036] In a fixed binocular setup, when the corrected first target position and second target position are located in a common effective image area and the longitudinal position difference is less than or equal to the epipolar allowable distance, the initial depth is obtained based on the parallax of the fire source image, and a consistency check is performed between the initial depth and the depth corresponding to the three-dimensional position of the fire source.
[0037] Furthermore, the re-determining of the target's corresponding location includes:
[0038] Using the target location from the previous positioning cycle as the search center for the current positioning cycle, the nearest candidate location of the same category is determined in the current first candidate location set and second candidate location set respectively;
[0039] The average sampling interval of the boundary is determined based on the total arc length of the fire source boundary in the current fire source area and the number of candidate positions of the boundary. The nearest candidate position is then determined to be a continuous candidate position based on the average sampling interval of the boundary or the candidate position interval.
[0040] For pairs of candidate positions, the spatial intersection distance is recalculated. If the spatial intersection condition is met, the target position is updated. If the spatial intersection condition is not met, the candidate position filtering is re-executed.
[0041] Furthermore, the spatial location results of the bridge fire source include:
[0042] The three-dimensional location of the fire source is transformed into the bridge spatial coordinate system to obtain the spatial coordinates of the fire source. The spatial coordinates of the fire source are then projected onto the bridge surface along the bridge surface normal. The lane area where the fire source is located is determined based on the inclusion relationship between the fire source projection point on the bridge surface and the lane boundary.
[0043] The nearest key components are determined based on the shortest spatial distance between the spatial coordinates of the fire source and the actual spatial outline of the key components of the bridge, thus forming the spatial positioning result of the bridge fire source.
[0044] Secondly, the present invention provides a bridge fire source spatial positioning system based on multi-view stereo vision, used to implement the above-mentioned bridge fire source spatial positioning method based on multi-view stereo vision, including:
[0045] The fire source area processing module is used to form a first candidate location set and a second candidate location set based on the fire source area from each observation perspective;
[0046] The candidate correspondence filtering module is used to perform spatial intersection filtering on the candidate positions in the first candidate position set and the second candidate position set according to the device attitude and spatial geometric relationship of the observation perspective, so as to determine the target correspondence position;
[0047] The three-dimensional position calculation module is used to determine the parallax of the fire source image based on the target's corresponding position, calculate the three-dimensional position of the fire source by combining the fire source image parallax, the spatial geometric relationship, and the device attitude, and re-determine the target's corresponding position when the spatial intersection relationship of the target's corresponding position does not meet the spatial intersection condition.
[0048] The bridge spatial mapping module is used to generate spatial positioning results of the bridge fire source based on the three-dimensional location of the fire source.
[0049] Compared to existing technologies, the advantages of this invention are as follows: In vehicle fire scenarios in open bridge spaces, this invention does not directly use the center of the fire source area for cross-view stereo matching. Instead, it retains boundary candidate positions and internal candidate positions within the fire source area from each observation perspective. It then utilizes the intersection relationship between the observation equipment's attitude, epipolar relationship, and spatial observation direction to determine the target corresponding position that can both point to the same spatial area from the candidate positions. This avoids erroneous parallax and 3D coordinate offsets caused by the fire source area centers actually corresponding to different spatial locations when the flame is swayed, stretched, or partially obscured by vehicles. Furthermore, by filtering through spatial intersection distance, filtering the bridge entity spatial boundaries, and checking depth consistency, erroneous candidate combinations can be eliminated, improving the reliability of the 3D position calculation of the fire source. During continuous positioning, the continuity of the correspondence is determined based on the nearest neighbor relationship between the candidate position of the current positioning cycle and the target position of the previous positioning cycle. If necessary, the target corresponding position is re-filtered, thereby suppressing coordinate jumps caused by changes in the obstruction range and flame contour. Ultimately, the effective three-dimensional location of the fire source is mapped to the bridge's spatial coordinate system, determining the lane area where the fire source is located and its distance from key bridge components, providing stable spatial positioning results for fire tracking and aiming of jet fire extinguishing devices. Attached Figure Description
[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0051] Figure 1 This is a schematic diagram of the multi-view fire source observation deployment in the open space of a bridge, as described in an embodiment of the present invention.
[0052] Figure 2 This is a schematic diagram showing the correspondence between the positions of the colored flame area and the infrared high-temperature area in an embodiment of the present invention;
[0053] Figure 3 This is a schematic diagram of the fire source area and candidate locations from different observation angles in an embodiment of the present invention;
[0054] Figure 4 This is a schematic diagram of the spatial intersection screening of candidate positions in an embodiment of the present invention;
[0055] Figure 5 This is a schematic diagram showing the correspondence between the spatial coordinates of the fire source and the bridge lanes and key components in an embodiment of the present invention;
[0056] Figure 6 This is a schematic diagram illustrating the target's position update under varying occlusion conditions in an embodiment of the present invention;
[0057] Figure 7 This is a schematic diagram illustrating the relationship between the spatial positioning result of the bridge fire source and the aiming of the jet fire extinguishing device in an embodiment of the present invention. Detailed Implementation
[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0059] Example 1:
[0060] This embodiment provides a method for spatial localization of bridge fire sources based on multi-view stereo vision. The fire source region refers to the image region formed by corresponding the flame identification area in a color image with the corresponding abnormal high-temperature area in an infrared image within the same observation viewpoint. The observation viewpoint refers to a single image observation state with a defined optical center position, optical axis direction, and imaging parameters. Different fixed installation positions form different observation viewpoints, and observation equipment with a rotatable support platform can also form different observation viewpoints when in different postures. Candidate corresponding positions refer to image positions extracted from fire source regions corresponding to different observation viewpoints and used to establish cross-view spatial correspondence. Spatial intersection conditions refer to the spatial observation directions formed by back-projection of different candidate corresponding positions, which, within the allowable range of equipment positioning error, can all point to the same spatial region. Target corresponding positions refer to a set of cross-view candidate corresponding positions used to determine the parallax of the fire source image and the three-dimensional position of the fire source after being filtered by spatial intersection conditions. The bridge spatial coordinate system refers to an engineering coordinate system established with the bridge longitudinal direction, bridge deck transverse direction, and bridge deck normal as three coordinate directions. The positioning cycle refers to the image acquisition and positioning processing cycle corresponding to one multi-view fire source spatial localization process.
[0061] The flame outline in the open space of the bridge is constantly changing. Crosswinds can deflect and stretch the flames, and vehicle bodies, guardrails, and adjacent vehicles can obscure flame areas at different heights. The flame outline obtained from different observation directions at the same time will also differ. When two observation perspectives are used for stereo matching using the center position of the fire source area, one center position may be located below the flame, and the other above it. The two image positions actually correspond to different spatial locations, and the spatial position calculated from the two image positions will deviate from the actual location of the fire source. During continuous positioning, as the obstruction range changes, the center position of the fire source area will also shift with changes in the visible outline, causing jumps in the calculated spatial coordinates of the fire source.
[0062] To address the aforementioned situation, this embodiment retains boundary candidate positions and internal candidate positions within each fire source area. Utilizing the intersection relationship between the observation equipment's attitude, epipolar relationships, and spatial observation directions, it determines cross-viewpoint combinations from the candidate positions that all point to the same spatial region. Then, based on the selected target corresponding positions, it determines the three-dimensional position of the fire source. When the original target corresponding position changes due to flame morphology changes or vehicle obstruction during continuous positioning, the cross-viewpoint correspondence is re-evaluated based on the candidate positions of the current positioning cycle.
[0063] See Figure 1Bridge 10 includes a bridge deck 11, a first lane 12, a second lane 13, a third lane 14, guardrails 15, and key bridge components 16. A first observation unit 20 is located on one side of bridge 10, and a second observation unit 30 is located at a spatial baseline distance from the first observation unit 20. The first observation unit 20 includes a first color imaging unit 21 and a first infrared imaging unit 22, and the second observation unit 30 includes a second color imaging unit 31 and a second infrared imaging unit 32. The observation ranges of the first observation unit 20 and the second observation unit 30 share a common bridge deck area. Vehicle 40 is located in the second lane 13, and the fire source 50 is located in the area where vehicle 40 is located.
[0064] The first color imaging unit 21 and the first infrared imaging unit 22 are fixedly installed, as are the second color imaging unit 31 and the second infrared imaging unit 32. Image position correspondence parameters are pre-obtained between the first color imaging unit 21 and the first infrared imaging unit 22, and between the second color imaging unit 31 and the second infrared imaging unit 32.
[0065] The determination of image position corresponding parameters covers the target's working distance range. During the deployment phase, calibration targets that can be simultaneously identified by both the color imaging unit and the infrared imaging unit are deployed at the near, middle, and far ends of the target's working distance range. Corresponding color and infrared images are acquired, and the corresponding pixel positions of the same calibration target in the color and infrared images are obtained at different distances. The position transformation parameters from infrared image coordinates to color image coordinates are solved using all corresponding pixel positions. At the same time, the pixel residuals between the transformed color image position of each calibration position and the corresponding real color image position of the same calibration target are calculated.
[0066] The 95th percentile of the pixel residuals corresponding to all working distances is recorded as the allowable cross-band registration distance. This allowable cross-band registration distance covers the working distance-related positional differences caused by the installation baseline between the color imaging unit and the infrared imaging unit, pixel quantization errors, and corresponding pixel position extraction errors. During operation, within the allowable cross-band registration distance, it is determined whether the mapped infrared high-temperature region and the colored flame region form a positional correspondence, ensuring that the near-end, middle, and far-end fire sources all adopt positional correspondence conditions verified by actual working distances.
[0067] In other embodiments, the first observation unit 20 and the second observation unit 30 employ a dual-band imaging device that has undergone joint color and infrared calibration. The dual-band imaging device directly outputs a joint color and infrared fire mask. This joint fire mask refers to a fire source area mask formed after positional registration of the color flame identification results and the infrared high-temperature identification results. In this implementation, the joint fire mask is subsequently used directly to determine the fire source area within the corresponding observation viewpoint. Both implementations ultimately form fire source areas with the same data structure.
[0068] The first observation unit 20 and the second observation unit 30 share the same time reference for the bridge monitoring system. Image frames output by the first color imaging unit 21, the first infrared imaging unit 22, the second color imaging unit 31, and the second infrared imaging unit 32 all carry acquisition timestamps. A single multi-view positioning operation selects image frames whose acquisition time difference is less than or equal to one image sampling period. When the image sampling frequency is 25 frames per second, one image sampling period is 40 milliseconds. Using images within the same sampling period can limit additional cross-view differences caused by changes in flame morphology at different acquisition times.
[0069] S1: Determine the fire source area from different observation angles based on the color and infrared images of the open space of the bridge, and determine the candidate corresponding position within each fire source area.
[0070] Step S1 converts the bridge fire identification results into a set of image locations that can be further processed using stereo vision. When the fire source area is partially obscured, the visible outline and internal temperature distribution change with the observation direction. By storing a set of candidate locations within the fire source area, the subsequent spatial localization process can select matching locations based on actual spatial correspondences, allowing the cross-view location determination process to adapt to changes in the flame outline.
[0071] Further, step S1 includes:
[0072] S11: Determine the fire source areas corresponding to the first and second observation perspectives respectively.
[0073] Specifically, the first observation viewpoint corresponds to the first color imaging unit 21 and the first infrared imaging unit 22, and the second observation viewpoint corresponds to the second color imaging unit 31 and the second infrared imaging unit 32. The bridge fire sensing module outputs a first color flame mask and a second color flame mask, and simultaneously outputs a first infrared high-temperature mask and a second infrared high-temperature mask. The color flame mask represents the pixel area in the color image identified as a flame, and the infrared high-temperature mask represents the pixel area in the infrared image determined by the bridge fire sensing module to be at an abnormally high temperature.
[0074] The background temperature used for the infrared high-temperature mask is updated in real time by the bridge fire sensing module according to the background temperature of the bridge surface in the infrared image. As one implementation, the bridge fire sensing module selects a bridge surface background area from the current infrared image, statistically analyzes the temperature of that background area, generates a background temperature corresponding to the current image based on the statistical temperature, and then determines abnormally high-temperature areas based on the difference between the current pixel temperature and the background temperature.
[0075] The first infrared high-temperature mask is mapped to the first color image coordinate system according to the image position transformation parameters corresponding to the first observation unit 20, resulting in the first mapped infrared high-temperature region 52; the second infrared high-temperature mask is mapped to the second color image coordinate system in the same manner, resulting in the second mapped infrared high-temperature region. (See also...) Figure 2 , specifically, Figure 2 (a) shows the first colored flame region 51. Figure 2 (b) shows the first mapped infrared high-temperature region 52. Figure 2 (c) illustrates the positional correspondence between the first colored flame region 51, the first mapped infrared high-temperature region 52, and the first fire source region 53. The first colored flame region 51 and the first mapped infrared high-temperature region 52 overlap within the allowable cross-band registration distance. For the current colored flame pixel in the first colored flame region 51, the minimum pixel distance from the current colored flame pixel to the first mapped infrared high-temperature region 52 is calculated. When the minimum pixel distance is less than or equal to the allowable cross-band registration distance, the current colored flame pixel is written into the first fire source region 53. The first fire source region 53 is obtained by traversing the first colored flame region 51.
[0076] Using the same processing method, a second fire source region 54 is formed based on the positional relationship between the second colored flame region and the second mapped infrared high-temperature region. When the first observation unit 20 or the second observation unit 30 directly outputs a combined colored and infrared fire mask, the corresponding combined fire mask is used as the first fire source region 53 or the second fire source region 54. Subsequent steps uniformly process the first fire source region 53 and the second fire source region 54.
[0077] S12: Determine the boundary of the corresponding fire source area and the internal temperature distribution based on the first fire source area 53 and the second fire source area 54.
[0078] Specifically, connected component extraction is performed on the first fire source region 53. When there are more than two unconnected fire source regions in the same image, each connected component is saved separately. The bridge fire perception module provides fire identifiers for the fire source regions based on the existing fire tracking results. The first fire source region 53 and the second fire source region 54, which belong to the same fire identifier, enter the same multi-view localization.
[0079] Contour tracing is performed on the first fire source region 53 to obtain the first fire source boundary pixel sequence according to the pixel connection order. The infrared temperature value corresponding to the first infrared high-temperature mask is mapped to the first fire source region 53, forming a first temperature pixel set inside the first fire source region 53.
[0080] The same processing is performed on the second fire source region 54 to obtain the second fire source boundary pixel sequence and the second temperature pixel set. The first and second fire source boundary pixel sequences are used to store the actual visible flame outline positions under different viewing directions, while the first and second temperature pixel sets are used to store the image positions with relatively high temperatures inside the fire source region. Both types of positions are ultimately represented by the pixel horizontal and vertical coordinates in the corresponding viewing angle.
[0081] S13: Determine the sampling parameters of the candidate positions based on the dimensions of the first fire source region 53 and the second fire source region 54 respectively, to form the first candidate position set 55 and the second candidate position set 56.
[0082] Further, step S13 includes:
[0083] S131: Determine the candidate position intervals corresponding to the first fire source region 53 and the second fire source region 54 respectively.
[0084] Specifically, the number of pixels contained in the current fire source region is calculated, and this number is denoted as the area A of the current fire source region. The diameter of a circle with the same area as the current fire source region is used as the equivalent diameter of the current fire source region, which is calculated according to the following formula:
[0085]
[0086] in, This represents the equivalent diameter of the current fire source area, in pixels. This represents the area of the current fire source region, expressed in square pixels. The value is pi. One-eighth of the current fire source area's equivalent diameter is used as the initial candidate position interval. Using this initial candidate position interval, a candidate selection is performed once according to the internal candidate position selection rules of step S133. When the number of internal candidate positions that can be formed exceeds eight, the candidate position interval is increased incrementally using one-twenty-fourth of the current fire source area's equivalent diameter. Each increment triggers a re-execution of the internal candidate position selection until the number of internal candidate positions is less than or equal to eight, or the candidate position interval reaches one-quarter of the current fire source area's equivalent diameter. The aforementioned upper and lower limits for the number of internal candidate positions, the candidate position interval increment, and the upper limit for the candidate position interval are preset parameters for the deployment phase and can be determined based on imaging resolution, target working distance, typical fire source area size, positioning accuracy requirements, and the real-time computational load of a single positioning cycle.
[0087] If the number of internal candidate positions formed when using the initial candidate position interval is less than 4, the initial candidate position interval is retained. This yields the candidate position interval corresponding to the current fire source region. The first fire source region 53 and the second fire source region 54 independently calculate their corresponding candidate position intervals based on their respective areas.
[0088] S132: Extract candidate boundary positions from the boundaries of the first fire source region 53 and the second fire source region 54 respectively.
[0089] Specifically, the total arc length of the boundary corresponding to the first fire source boundary pixel sequence is calculated. This total arc length is then divided by the interval between candidate positions corresponding to the first fire source region 53. The nearest integer of the result is taken to obtain the number of initial segments. When the number of initial segments is less than 12, the number of first boundary segments is set to 12; when the number of initial segments is greater than 24, the number of first boundary segments is set to 24; when the number of initial segments is between 12 and 24, the number of initial segments is directly used as the number of first boundary segments. The lower and upper limits of the number of boundary segments are preset parameters for the deployment phase, determined based on the minimum sampling density of the fire source contour and the maximum number of candidate boundary positions allowed to participate in spatial intersection calculations in a single positioning cycle.
[0090] The first fire source boundary pixel sequence is divided into continuous segments along the boundary arc length according to the number of first boundary segments. In each continuous segment, the boundary pixel closest to the midpoint of the corresponding segment arc length is selected as the first boundary candidate position. The second fire source region 54 is determined independently using the same method based on the total arc length of the second fire source boundary and the interval between the corresponding candidate positions of the second fire source region 54, and then the second boundary candidate positions are formed.
[0091] When the first observation viewpoint can see both the lower and upper parts of the flame, while the second observation viewpoint can only see the upper part of the flame due to obstruction by the vehicle 40, the boundary length of the second fire source region 54 shortens with the visible range. The number of second boundary segments is determined by the actual boundary length of the second fire source region 54, thus maintaining the match between the candidate positions of the second boundary and the current visible contour size. The current average sampling interval of the first fire source region 53 is equal to the total arc length of the first fire source boundary divided by the number of first boundary segments. The current average sampling interval of the second fire source region 54 is calculated independently in the same way. The average sampling interval of the boundary is also used for the continuation judgment of the candidate positions of the boundary in subsequent continuous positioning cycles.
[0092] S133: Extract internal candidate positions from the first temperature pixel set and the second temperature pixel set respectively.
[0093] Specifically, the pixels in the first temperature pixel set are arranged from highest to lowest temperature value, and the first internal candidate position is selected starting from the pixel with the highest temperature value. After determining a first internal candidate position, the pixel distance between the remaining temperature pixels and the determined first internal candidate position is calculated. Pixels whose distance is less than the candidate position interval corresponding to the first fire source region 53 are excluded. Then, the pixel with the highest temperature value is selected from the remaining temperature pixels as the next first internal candidate position. This process is repeated until all temperature pixels that meet the candidate position interval requirement have been processed or 8 first internal candidate positions have been formed.
[0094] When the number of pixels meeting the candidate position spacing requirement is less than 4, the actual number of pixels obtained is used as the number of first internal candidate positions. When the first temperature pixel set does not form internal candidate positions, the first candidate position set 55 is composed of the first boundary candidate positions.
[0095] The second temperature pixel set is processed in the same way, and the corresponding candidate positions of the second fire source region 54 are used as intervals to form the second internal candidate positions. The boundary candidate positions and the internal candidate positions in the current observation view together constitute the candidate position set corresponding to the current observation view.
[0096] S134: Write the boundary candidate positions and the internal candidate positions into the corresponding candidate position set.
[0097] Each record in the first candidate location set 55 includes at least the current positioning cycle number, candidate location number, pixel x-coordinate, pixel y-coordinate, candidate location category, and associated fire situation identifier. The candidate location category can be either a boundary candidate location or an internal candidate location.
[0098] The second candidate location set 56 uses the same data structure. The candidate location number is regenerated at the beginning of each positioning cycle, so each positioning cycle forms a candidate location record for the current cycle based on the actual outline and temperature distribution of the current fire source area.
[0099] See Figure 3 The first fire source region 53 and the second fire source region 54 have different visible outlines due to different viewing directions and the occlusion relationship of the vehicle 40. The first candidate position set 55 is distributed within the outline and interior of the first fire source region 53, and the second candidate position set 56 is distributed within the outline and interior of the second fire source region 54. When the vehicle 40 occludes the lower part of the second fire source region 54, the second candidate position set 56 can still retain the upper boundary and interior position of the second fire source region 54, providing candidate positions for subsequent cross-view spatial correspondence.
[0100] Step S1 retains candidate positions that can continue to undergo spatial geometric verification. After the fire source area is partially obscured, the candidate positions corresponding to the obscured parts are updated with the current fire source area, while the unobscured parts still form candidate positions according to the current actual visible contour and temperature distribution, so that cross-view matching can reselect spatial correspondences from the current periodic candidate position set.
[0101] For example, for ease of explanation, when vehicle 40 catches fire, the first fire source area 53 can see the lower and upper parts of the flames from the first observation perspective, while the second fire source area 54 is obstructed by the side wall of vehicle 40, and only the upper middle part of the flames can be seen. The first candidate position set 55 forms 16 boundary candidate positions and 6 internal candidate positions, and the second candidate position set 56 forms 13 boundary candidate positions and 5 internal candidate positions based on the size of the second fire source area 54 itself. The candidate positions then proceed to step S2 for spatial correspondence filtering.
[0102] S2: Based on the device attitude and spatial geometric relationship corresponding to each observation angle, perform spatial intersection screening on the candidate positions in the first candidate position set 55 and the second candidate position set 56 to determine the target corresponding position belonging to the same fire source spatial part.
[0103] After the flame outline oscillates, a candidate position can determine whether the current pixel is inside or at the boundary of the fire source area. However, cross-view correspondence still needs to be determined in conjunction with the observation direction. The positions, optical axis directions, and imaging parameters of the first observation unit 20 and the second observation unit 30 have been determined. Therefore, any candidate position can be back-projected into a spatial observation direction. The spatial observation direction formed by two candidate positions belonging to the same actual spatial location should both point to the same spatial area within the allowable range of equipment positioning error.
[0104] Further, step S2 includes:
[0105] S21: Determine the device attitude information and spatial geometric relationship corresponding to the first and second observation perspectives.
[0106] In one embodiment, the first observation unit 20 and the second observation unit 30 are fixedly installed. After installation, the camera's extrinsic parameters are calibrated using control points on the bridge deck with known measurement coordinates. The coordinates of the control points are provided by bridge engineering survey data.
[0107] The pixel coordinates of the same control point in the first color imaging unit 21 and the second color imaging unit 31 are recorded respectively. The rotation parameters and position parameters of the first observation unit 20 and the second observation unit 30 relative to the bridge spatial coordinate system are obtained using the camera extrinsic parameter calibration method. In the fixed installation state, the rotation parameters and position parameters are read as the equipment attitude information during operation.
[0108] In other embodiments, at least one of the first observation unit 20 and the second observation unit 30 is disposed on a rotatable support platform. The position parameters of the observation unit mounting base relative to the bridge's spatial coordinate system are determined during the deployment phase. During operation, the current horizontal rotation angle and current pitch angle output by the support platform are read, and the rotation parameters of the mounting base, along with the current horizontal rotation angle and current pitch angle, are converted into the current observation unit equipment attitude information.
[0109] Both implementations ultimately yield the three-dimensional positions of the optical centers of the first observation unit 20 and the second observation unit 30 in the bridge's spatial coordinate system, and simultaneously obtain the directions of the corresponding optical axes relative to the bridge's spatial coordinate system. The camera intrinsic parameters corresponding to the first color imaging unit 21 and the second color imaging unit 31 include focal length, principal point position, and lens distortion parameters. These camera intrinsic parameters are obtained during the equipment deployment phase using camera calibration methods.
[0110] S22: Form candidate position combinations based on the first candidate position set 55 and the second candidate position set 56, and determine the spatial observation direction corresponding to each candidate position combination.
[0111] Further, step S22 includes:
[0112] S221: Filter candidate positions that can form combinations according to candidate position category and polar relationship.
[0113] Candidate position combinations are established by combining the boundary candidate positions in the first candidate position set 55 with the boundary candidate positions in the second candidate position set 56, and by combining the internal candidate positions in the first candidate position set 55 with the internal candidate positions in the second candidate position set 56, so that the two image positions within the same candidate position combination have the same position category. For the current first candidate position in the first candidate position set 55, the corresponding epipolar line of the current first candidate position in the second observation view is obtained according to the dual-target positioning parameters of the first observation unit 20 and the second observation unit 30.
[0114] The allowable epipolar distance is determined during the deployment phase through repeated calibration data. Specifically, control points with known spatial coordinates are deployed at the near, middle, and far ends of the target working distance range. At least 20 sets of binocular images are continuously acquired at each control point. For each set of binocular images, the control point in the first observation view is projected onto the corresponding epipolar line in the second observation view, and the vertical pixel distance from the actual control point pixel position in the second observation view to the corresponding epipolar line is calculated.
[0115] A set of epipolar residual samples is formed by combining the vertical pixel distances obtained from all working distances, all control points, and all duplicate images. The 95th percentile value of this set of epipolar residual samples is taken as the epipolar allowable distance. The epipolar residual sample set covers camera calibration residuals, pixel quantization errors, control point extraction errors, and imaging position differences at different working distances. During runtime, second candidate positions in the second candidate position set 56 whose vertical pixel distance to the current corresponding epipolar line is less than or equal to the epipolar allowable distance, and whose candidate position category is the same as the current first candidate position, are retained.
[0116] S222: Convert candidate locations into space observation directions.
[0117] Lens distortion correction is performed on the current first candidate position. Based on the camera intrinsic parameters of the first color imaging unit 21, the corrected pixel position is transformed to the normalized imaging plane to obtain the imaging direction in the first camera coordinate system. Using the device attitude information corresponding to the first observation unit 20, the imaging direction in the first camera coordinate system is transformed to the bridge spatial coordinate system to obtain the first spatial observation direction 61. The first spatial observation direction 61 is jointly represented by the three-dimensional coordinates of the optical center observed by the first observation unit 20 and the first unit direction vector.
[0118] The same method is used to obtain the second spatial observation direction 62 for the current second candidate position. A candidate position combination record is formed from the current first candidate position and the current second candidate position. The candidate position combination record shall at least store the first candidate position number, the second candidate position number, the candidate position category, the first spatial observation direction 61, the second spatial observation direction 62, and the fire identification number.
[0119] S223: Calculate the spatial intersection distance and candidate intersection positions between the first spatial observation direction 61 and the second spatial observation direction 62.
[0120] When the first spatial observation direction 61 and the second spatial observation direction 62 are affected by pixel discretization error, camera extrinsic parameter error, and fire source contour extraction error, they will form spatial straight lines with finite spacing. The method of calculating the closest point of two spatial straight lines is used to obtain the first closest point 63 on the first spatial observation direction 61 and the second closest point 64 on the second spatial observation direction 62.
[0121] Using the three-dimensional coordinates of the optical center observed by the first observation unit 20, the first unit direction vector, the three-dimensional coordinates of the optical center observed by the second observation unit 30, and the second unit direction vector as inputs for calculating the closest point between two straight lines in space, the first closest point 63 and the second closest point 64 are determined according to the analytical solution relationship of the closest point between two straight lines in space. The Euclidean distance between the first closest point 63 and the second closest point 64 is calculated and recorded as the spatial intersection distance. The midpoint of the line connecting the first closest point 63 and the second closest point 64 is taken as the candidate intersection position 66. The midpoint of the closest line segment can provide the corresponding spatial position when two spatial observation directions form skew lines, and the spatial intersection distance reflects the proximity between the spatial observation directions corresponding to the two candidate positions.
[0122] S23: Select candidate position combinations based on spatial intersection conditions.
[0123] Further, step S23 includes:
[0124] S231: Perform the first screening based on the spatial intersection distance.
[0125] The spatial intersection distance threshold is determined using repeated positioning data from control points during the deployment phase. Specifically, test points with known spatial coordinates are set at the near, middle, and far ends of the target working distance range. At least 20 sets of binocular images are continuously acquired at each test point. For each set of binocular images, the pixel positions of the test points are extracted from both the first and second observation perspectives. Two spatial observation directions are formed according to step S222, and the spatial intersection distance is calculated according to step S223.
[0126] A spatial intersection residual sample set is formed by representing all spatial intersection distances corresponding to all test points. The 95th percentile value of this sample set is used as the spatial intersection distance threshold. For example, after device deployment, the spatial intersection distance threshold is 0.30 meters. If the spatial intersection distance corresponding to the current candidate location combination is less than or equal to 0.30 meters, the candidate location combination is moved to the next stage of screening; if the spatial intersection distance is greater than 0.30 meters, the candidate location combination is deleted from the current candidate location combination set.
[0127] S232: Perform a second screening based on the positional relationship between candidate intersection location 66 and the bridge surface entity space.
[0128] The bridge engineering data pre-provides the spatial outlines of the bridge deck 11, guardrail 15, and key bridge components 16. The bridge deck 11 is represented as a bridge deck area with spatial elevation in the bridge spatial coordinate system, and the first lane 12, the second lane 13, and the third lane 14 are respectively represented as lane boundary areas located in the bridge deck area.
[0129] The candidate intersection position 66 is compared with the corresponding spatial area of the bridge deck 11. If the candidate intersection position 66 is located in the open space above the bridge deck or in the open space adjacent to the bridge's critical component 16, the current candidate position combination is retained. If the candidate intersection position 66 falls inside the bridge deck entity, in the space below the bridge deck, or in an area far from the actual bridge structure, the candidate position combination is deleted from the current candidate position combination set. The spatial location of the flames corresponding to vehicle fires is concentrated in the open areas above the bridge deck and adjacent to bridge components; therefore, the spatial boundary of the bridge entity can be used to further limit the range of candidate intersection positions.
[0130] S233: Form a set of valid candidate correspondences.
[0131] The candidate position combinations filtered through steps S231 and S232 are written into the valid candidate correspondence set. Each valid candidate correspondence record includes at least the first candidate position, the second candidate position, the candidate position category, the spatial intersection distance, and the candidate intersection position 66.
[0132] For example, the first internal candidate position P1 in the first candidate position set 55 and the second internal candidate position Q1 in the second candidate position set 56 form a first candidate position combination. According to step S223, the spatial intersection distance is calculated to be 0.18 meters, and the candidate intersection position is located 1.7 meters above the second lane 13. The first internal candidate position P2 and the second internal candidate position Q2 form a second candidate position combination, with a corresponding spatial intersection distance of 0.24 meters, and the candidate intersection position is located 2.0 meters above the second lane 13.
[0133] The first internal candidate position P3 and the second internal candidate position Q3 form a third candidate position combination with a spatial intersection distance of 1.16 meters. When a spatial intersection distance threshold of 0.30 meters is used, the combination of P1 and Q1 and the combination of P2 and Q2 are included in the set of valid candidate combinations, while the combination of P3 and Q3 is deleted from the set of candidate position combinations.
[0134] S24: Determine the target corresponding position from the set of valid candidate corresponding positions.
[0135] When the set of valid candidate correspondences contains only one valid candidate correspondence record, the first candidate position and the second candidate position in that valid candidate correspondence record are determined as the target correspondence position. When the set of valid candidate correspondences contains two or more valid candidate correspondence records, they are arranged in ascending order of spatial intersection distance.
[0136] To distinguish whether the spatial intersection distance difference between different valid candidate corresponding records exceeds the repeated positioning fluctuation range of the device, this embodiment sets a resolvable threshold for spatial intersection distance. The resolvable threshold for spatial intersection distance is determined during the device deployment phase. For the same control point with a known spatial location, no less than 20 sets of binocular images are continuously acquired, and the spatial intersection distance corresponding to each set of binocular images is obtained according to step S223, thereby forming a repeated spatial intersection distance sequence.
[0137] Calculate the median of the repeating spatial intersection distance sequence, and then calculate the absolute difference between each spatial intersection distance in the sequence and the median, forming a repeating location absolute deviation set. Take the 95th percentile of the repeating location absolute deviation set as the resolvable threshold for spatial intersection distance.
[0138] When the difference in spatial intersection distance between the first and second valid candidate corresponding records after sorting is greater than the resolvable threshold for spatial intersection distance, the valid candidate corresponding record with the smaller spatial intersection distance is selected. When the difference in spatial intersection distance between two valid candidate corresponding records is less than or equal to the resolvable threshold for spatial intersection distance, for non-first positioning cycles, the spatial distance between the two candidate intersection positions and the three-dimensional position of the valid fire source in the previous positioning cycle is calculated, and the valid candidate corresponding record with the smaller spatial distance to the three-dimensional position of the valid fire source in the previous positioning cycle is selected.
[0139] For the first positioning cycle, the target's location is determined starting with the records with the smallest spatial intersection distances. When spatial intersection distances are the same, records corresponding to internal candidate locations are prioritized over records corresponding to boundary candidate locations. When candidate location categories are the same, the target's location is determined according to the candidate location number order. See also Figure 4 When the first spatial observation direction 61 and the second spatial observation direction 62 correspond to candidate position combinations that satisfy the spatial intersection conditions, the candidate intersection position 66 is located in the area where the fire source 50 is located above the vehicle 40. The other set of candidate positions corresponds to the first incorrect observation direction 67 and the second incorrect observation direction 68. The corresponding intersection areas are deviated from the area where the fire source 50 is located, or the corresponding spatial intersection distance exceeds the spatial intersection distance threshold. Therefore, they are screened out in step S231.
[0140] Step S2 uses the device attitude, epipolar relationship, and spatial intersection relationship together for cross-view position determination. The epipolar relationship limits the corresponding range of candidate positions in the image, and the spatial intersection relationship further checks the consistency of corresponding positions in actual space, so that candidate position combinations formed by different visible parts of the flame can be filtered before 3D localization.
[0141] For example, the spatial intersection distance between P1 and Q1 is 0.18 meters, and the spatial intersection distance between P2 and Q2 is 0.24 meters. The resolvable threshold for spatial intersection distance during equipment deployment is 0.03 meters. The difference in spatial intersection distance between the two records is 0.06 meters, which is greater than 0.03 meters; therefore, P1 and Q1 are selected as the target corresponding positions for the current positioning cycle.
[0142] S3: Determine the parallax of the fire source image based on the corresponding position of the target, calculate the three-dimensional position of the fire source based on the parallax of the fire source image, spatial geometric relationship and equipment attitude information, and re-determine the corresponding position of the target when the target correspondence changes.
[0143] Step S3 executes the final spatial solution after the target location is obtained through filtering in step S2. During continuous positioning, the flame shape and occlusion status continue to change. The target location corresponding to the previous positioning cycle may move to a nearby location in the current positioning cycle, or it may be replaced by a new visible fire source. Therefore, it is necessary to re-determine the correspondence based on the candidate locations of the current positioning cycle.
[0144] Further, step S3 includes:
[0145] S31: Determine the parallax of the fire source image based on the location of the first target and the location of the second target.
[0146] Specifically, for a fixed binocular setup supporting epipolar correction, standard binocular epipolar correction is performed using the camera intrinsic parameters and relative extrinsic parameters of the first observation unit 20 and the second observation unit 30, so that the two corrected virtual images have a common epipolar direction. The first target position is corrected to obtain the first corrected target position, and the second target position is corrected to obtain the second corrected target position. The absolute value of the difference between the abscissa of the first and second corrected target positions is calculated, and this absolute value is taken as the fire source image disparity d, in pixels. When the first observation unit 20 and the second observation unit 30 form a common effective corrected image area corresponding to the fixed binocular setup, the initial depth Z of the fire source target is calculated according to the following formula:
[0147]
[0148] in, This represents the initial depth of the fire source target relative to the first virtual camera coordinate system after epipolar correction, in meters; This represents the equivalent focal length of the virtual camera after epipolar correction, in pixels. This represents the baseline distance between the optical centers of the first observation unit 20 and the second observation unit 30, in meters. This represents the parallax of the fire source image, expressed in pixels. , And only in The initial depth calculation is performed at that time. Read from the virtual camera projection parameters generated during the dual-target localization and epipolar correction process. The optical center position is calculated from the corresponding extrinsic parameters of the first observation unit 20 and the second observation unit 30. It is calculated from the current target's location.
[0149] The dimensionless ratio is obtained by dividing the focal length per pixel by the parallax per pixel, and then multiplying it by the baseline distance B in meters. Therefore, the calculated Z is in meters. The initial depth obtained in step S31 is used for the consistency check in step S323, and the three-dimensional position of the fire source is calculated by the spatial observation direction intersection process in step S32.
[0150] S32: Calculate the three-dimensional position of the fire source based on the first target position, the second target position, and the equipment attitude information.
[0151] Further, step S32 includes:
[0152] S321: Reconstruct the target space observation direction.
[0153] The first target position is distorted according to the lens distortion parameters corresponding to the first color imaging unit 21, and converted into a unit direction vector in the first camera coordinate system based on the focal length and principal point position of the first color imaging unit 21. Using the current device attitude information of the first observation unit 20, the unit direction vector in the first camera coordinate system is transformed into the bridge spatial coordinate system to form the spatial observation direction of the first target.
[0154] Similarly, the spatial observation direction of the second target is formed based on the position of the second target. The spatial observation directions of the first and second targets are represented by the three-dimensional coordinates of the corresponding observation optical centers and the unit direction vector, respectively.
[0155] S322: Find the shortest line segment between the first target space observation direction and the second target space observation direction.
[0156] Using the same method as step S223 for calculating the closest point between two straight lines in space, the first closest point of the target in the spatial observation direction and the second closest point of the target in the spatial observation direction are obtained. The Euclidean distance between the first closest point of the target and the second closest point of the target is calculated, and the obtained distance is used as the target intersection distance.
[0157] When the target intersection distance is less than or equal to the spatial intersection distance threshold determined in step S231, the midpoint of the line connecting the nearest point of the first target and the nearest point of the second target is taken as the current three-dimensional position of the fire source. The current three-dimensional position of the fire source is stored using three floating-point numbers, corresponding to the longitudinal coordinate components of the bridge, the lateral coordinate components of the bridge deck, and the normal coordinate components of the bridge deck, respectively.
[0158] S323: Perform a conditional consistency check on the current three-dimensional position of the fire source based on the fixed binocular initial depth.
[0159] Before performing the consistency check, it is determined whether the current target position meets the usage conditions of the fixed binocular depth relationship. It is determined whether the first and second correction target positions are both located in the common effective image area of the two virtual images after epipolar correction, and the absolute difference between the ordinate of the first and second correction target positions is calculated.
[0160] When both the first and second correction target positions are within a common valid image area, and the absolute difference in their ordinates is less than or equal to the epipolar tolerance distance determined in step S221, a consistency check is performed using the fixed binocular initial depth from step S31. The initial depth Z calculated in step S31 is compared with the depth component of the current three-dimensional fire source position in the optical axis direction of the first virtual camera after epipolar correction, calculated in step S322.
[0161] The depth consistency threshold is determined during the equipment deployment phase using control points with known spatial distances. Control points with known distances are set at the near, middle, and far ends of the target working distance range. Two depths are obtained using the parallax depth calculation method in step S31 and the spatial observation direction intersection method in step S322, respectively. The absolute difference between the two depths is calculated to form a depth difference sample set. The 95th percentile value of the depth difference sample set is taken as the depth consistency threshold.
[0162] When the absolute difference between the two depths is less than or equal to the depth consistency threshold, the current three-dimensional position of the fire source obtained in step S322 is determined as the current valid three-dimensional position of the fire source. When the absolute difference between the two depths is greater than the depth consistency threshold, return to step S2 to redetermine the corresponding position of the target.
[0163] When either the first or second correction target location is outside the common valid image area, or when the absolute difference in the ordinate between the two correction target locations is greater than the epipolar allowable distance, the current positioning cycle uses the intersection result corresponding to the spatial observation direction in step S322. If the target intersection distance obtained in step S322 is less than or equal to the spatial intersection distance threshold, and the candidate intersection location satisfies the bridge entity spatial boundary conditions in step S232, then the current three-dimensional location of the fire source is determined as the current valid three-dimensional location of the fire source.
[0164] For observation methods with variable attitudes, the spatial observation direction intersection result and spatial intersection distance condition from step S322 are also used to determine the current effective three-dimensional position of the fire source. For example, after P1 and Q1 are determined as the corresponding positions of the targets, the first target spatial observation direction and the second target spatial observation direction are formed according to step S321. The target intersection distance is calculated to be 0.18 meters according to step S322. The coordinates of the midpoint of the line connecting the closest point of the first target and the closest point of the second target in the bridge spatial coordinate system are 124.6 meters longitudinally, 8.3 meters laterally, and 1.8 meters normal to the bridge surface. When the corresponding target position meets the fixed binocular initial depth check condition, and the difference between the initial depth obtained in step S31 and the corresponding depth of the three-dimensional position is less than or equal to the depth consistency threshold obtained in the deployment stage, the longitudinal 124.6 meters, lateral 8.3 meters, and bridge surface normal 1.8 meters are recorded as the current effective three-dimensional position of the fire source.
[0165] S33: Determine the continuity status of the target correspondence based on the nearest neighbor relationship between the candidate position in the current positioning cycle and the target corresponding position in the previous positioning cycle.
[0166] After re-executing step S1 in different positioning cycles, the length of the fire source boundary, the number of candidate locations, and the pixel coordinates of the candidate locations will all be updated according to the current fire source visibility status. Step S33 uses the target location corresponding to the previous positioning cycle as the nearest neighbor search center for the current positioning cycle to determine the continuing candidate locations from the current candidate location set.
[0167] For the first target position in the target location corresponding to the previous positioning cycle, a candidate position of the same category as the first target position candidate position in the current first candidate position set 55 is searched. The two-dimensional pixel Euclidean distance between each candidate position and the first target position in the previous positioning cycle is calculated, and the candidate position with the smallest distance is recorded as the first nearest continuing candidate position. When the first target position in the previous positioning cycle belongs to the boundary candidate position, the average sampling interval of the boundary corresponding to the current first fire source area 53 is used as the first continuing distance limit. When the pixel distance between the first nearest continuing candidate position and the first target position in the previous positioning cycle is less than or equal to the first continuing distance limit, the first nearest continuing candidate position is used as the continuing candidate position corresponding to the first observation viewpoint.
[0168] When the first target position in the previous positioning cycle is an internal candidate position, the interval between the candidate positions corresponding to the current first fire source region 53 is used as the first continuation distance limit. When the pixel distance between the first nearest continuation candidate position and the first target position in the previous positioning cycle is less than or equal to the first continuation distance limit, the first nearest continuation candidate position is used as the continuation candidate position corresponding to the first observation viewpoint.
[0169] The second observation perspective, using the same method, searches for the second most recent continuing candidate position in the current second candidate position set 56 that is of the same category as the second target position in the previous positioning cycle, and uses the average sampling interval of the current second fire source region 54's own boundary or the candidate position interval for continuity determination. When both the first and second observation perspectives obtain continuing candidate positions, the first most recent continuing candidate position and the second most recent continuing candidate position are used to form a continuing candidate position combination.
[0170] Steps S222 and S223 are re-executed for the combination of continuing candidate positions to calculate the new spatial observation direction, target intersection distance, and candidate intersection positions. When the new target intersection distance is less than or equal to the spatial intersection distance threshold, and the candidate intersection positions satisfy the spatial boundary conditions of the bridge entity, the first nearest continuing candidate position and the second nearest continuing candidate position are used as the new target corresponding positions for the current positioning cycle, and then steps S31 and S32 are entered to update the current effective three-dimensional position of the fire source.
[0171] If no candidate position that meets the continuation distance limit is obtained from any observation viewpoint, or if the target intersection distance formed by two nearest continuation candidate positions is greater than the spatial intersection distance threshold, the current target status is updated to a re-filtering status, and the process proceeds to step S34. The nearest neighbor determination allows the target corresponding position to move slightly along the visible outline of the flame within adjacent positioning cycles, while using the recalculated spatial intersection distance to check whether two current candidate positions still point to the same spatial region.
[0172] S34: In response to the re-filtering state, return to step S2 and redetermine the target's corresponding position based on the first candidate position set 55 and the second candidate position set 56 of the current positioning cycle.
[0173] See Figure 6 , Figure 6 This represents the update process of the target's corresponding position under changing occlusion conditions. Figure 6 In (a), the first target position and the second target position in the previous positioning cycle are located within the fire source area corresponding to the first observation view and the second observation view, respectively. Figure 6 In (b), vehicle 40 creates greater occlusion of the lower part of the fire source in the second observation view. The current second candidate position set 56 is regenerated with the new visible outline, and the area corresponding to the original second target position is occluded. Figure 6 In (c), the target corresponding position that satisfies the spatial intersection condition is re-formed based on the current first candidate position set 55 and the current second candidate position set 56.
[0174] In the second state, step S33 searches for the nearest candidate position of the same category as the second target position in the previous positioning cycle from the current second candidate position set 56. The pixel distance is determined based on the interval between the candidate positions corresponding to the current second fire source area 54, and the spatial intersection distance is recalculated for the current combination of nearest candidate positions. When the continuation condition is met, the target's corresponding position is updated using the current nearest candidate position, and the process proceeds to step S31 to continue positioning; when entering the re-filtering state, step S34 is executed.
[0175] If valid candidate corresponding records are obtained after filtering in step S2 during the current positioning period, the new target corresponding location is determined according to step S24. If the set of valid candidate corresponding records is empty during the current positioning period, the current positioning period is recorded as a positioning pending confirmation state.
[0176] While the location is pending confirmation, the bridge monitoring interface retains the fire source location record from the previous valid positioning cycle, and the current positioning status field is recorded as "location pending confirmation". After obtaining valid candidate records again in subsequent positioning cycles, a new current valid three-dimensional fire source location is formed according to steps S31 to S32.
[0177] In other embodiments, the system sets up three observation perspectives. Candidate correspondences are established between the first and second observation perspectives, the first and third observation perspectives, and the second and third observation perspectives, respectively.
[0178] When the spatial distance between candidate intersection positions obtained from at least two sets of observation perspectives is less than or equal to the corresponding spatial intersection distance threshold, the average coordinate of the candidate intersection positions that meet the conditions is determined as the current three-dimensional position of the fire source. If a certain observation perspective is blocked by a vehicle, resulting in an unavailable candidate position, the other two observation perspectives that are in a candidate position available state are used to perform positioning according to the aforementioned two-view spatial intersection method.
[0179] The number of spatial observation directions participating in spatial intersection checks is increased by adding observation perspectives, while maintaining the same processing relationship in the processes of fire source area determination, candidate position formation, candidate correspondence screening, and correspondence re-screening. Step S3 incorporates the continuous validity of the target's corresponding position into the continuous positioning process. After changes in flame swaying and vehicle obstruction, the current candidate position is regenerated based on the new fire source area. Step S33 first searches for neighboring positions in the current candidate position set, and then checks the correspondence using spatial intersection conditions; when the spatial correspondence state changes, step S34 re-executes candidate screening, so that the 3D position update is always based on the actually visible fire source area in the current positioning cycle.
[0180] For example, in the next positioning cycle, the crosswind causes the flames to shift laterally towards the bridge surface, while the occlusion range of vehicles in the second observation view expands. The pixel distance between the current second internal candidate position near the previous positioning cycle Q1 and the previous positioning cycle Q1 is greater than the interval of the corresponding candidate position in the current second fire source area 54, and the current target's corresponding state enters the re-filtering state.
[0181] Step S34 repeats step S2. The first internal candidate position P2 and the second internal candidate position Q2 in the current positioning cycle form a candidate position combination, and the spatial intersection distance is recalculated to be 0.21 meters. The candidate intersection position is located above the second lane 13. P2 and Q2 are determined as the new target corresponding positions in the current positioning cycle.
[0182] Following step S32, the current effective three-dimensional position of the fire source is recalculated as 124.7 meters longitudinally, 8.5 meters laterally, and 2.0 meters normal to the bridge surface. When step S1 is executed again in the next positioning cycle, the candidate position number is regenerated along with the candidate position record of the corresponding cycle.
[0183] S4: Based on the bridge outline, lane lines, and the spatial location of key bridge components, the current effective three-dimensional location of the fire source is converted to the bridge spatial coordinate system to form the spatial positioning result of the bridge fire source.
[0184] Step S4 gives the three-dimensional positions calculated from multiple perspectives a spatial meaning related to bridge engineering. After adopting the bridge spatial coordinate system, the positioning results from different observation devices can be uniformly mapped to the longitudinal position of the bridge deck, the lateral lane position, and the spatial position of key bridge components, allowing bridge fire response equipment to read them directly.
[0185] Further, step S4 includes:
[0186] S41: Represent the current effective three-dimensional position of the fire source in the bridge spatial coordinate system.
[0187] When the first observation unit 20 and the second observation unit 30 have directly used the bridge spatial coordinate system to represent the device attitude, the current effective three-dimensional position of the fire source obtained in step S32 is already in the bridge spatial coordinate system, and the current effective three-dimensional position of the fire source is directly recorded as the fire source spatial coordinate 74. When the observation device uses an independent device coordinate system, the rigid body transformation parameters from the device coordinate system to the bridge spatial coordinate system are obtained using no less than three non-collinear bridge control points during the device deployment phase. The current effective three-dimensional position of the fire source obtained in step S32 is then transformed according to the rigid body transformation parameters to obtain the fire source spatial coordinate 74 in the bridge spatial coordinate system.
[0188] See Figure 5The first coordinate axis 71 of the bridge spatial coordinate system 70 is set along the longitudinal direction of the bridge, the second coordinate axis 72 is set along the transverse direction of the bridge deck, and the third coordinate axis 73 is set along the normal direction of the bridge deck. The fire source spatial coordinate 74 is located above the second lane 13. The key bridge component 16 has a pre-stored spatial outline 75.
[0189] S42: Determine the lane area where the fire source is located based on the inclusion relationship between the fire source spatial coordinates 74 and the lane spatial area.
[0190] Specifically, the spatial coordinates 74 of the fire source are projected onto the bridge deck 11 along the direction of the third coordinate axis 73, resulting in the fire source bridge deck projection point 76. The first lane 12, the second lane 13, and the third lane 14 are each recorded as a closed polygon in the bridge spatial coordinate system to define their respective lane boundaries. A geometric determination method based on whether a point is inside a polygon is used to determine whether the fire source bridge deck projection point 76 is located inside the corresponding lane polygon.
[0191] When the fire source projection point 76 is located inside the polygon of the lane corresponding to the second lane 13, the second lane 13 is determined as the lane area where the fire source is located. When the fire source projection point 76 is located on the common boundary line of the two lane areas, the two adjacent lane numbers and the lane boundary status are written into the fire source spatial positioning record.
[0192] For example, the current effective three-dimensional location of the fire source is 124.7 meters longitudinally, 8.5 meters laterally, and 2.0 meters normal to the bridge deck. Projected onto the bridge deck 11, the fire source bridge deck projection point 76 is obtained with a longitudinal distance of 124.7 meters and a lateral distance of 8.5 meters. When the lateral range of the second lane 13 at the corresponding longitudinal position is 6.0 meters to 9.5 meters, the lateral distance of 8.5 meters falls within the corresponding range of the second lane 13, and the lane area where the fire source is located is recorded as the second lane 13.
[0193] S43: Determine the positional relationship between the fire source and the bridge key component 16 based on the distance between the fire source spatial coordinates 74 and the actual spatial range of the bridge key component 16.
[0194] The spatial profile 75 of the key bridge component 16 is provided by the bridge engineering model or as-built measurement data. For steel beams, cable clamps, and bridge tower components that can be represented as spatial surfaces, the shortest spatial distance from the spatial coordinates 74 of the fire source to the actual spatial surface of the corresponding component is calculated.
[0195] For cable-stayed components such as main cables, suspenders, or stay cables with finite lengths and curved centerlines, the actual centerline of the cable is represented as a continuous sequence of finite line segments according to the discrete node order in the bridge engineering model. For the current finite line segment, the perpendicular position of the fire source spatial coordinate 74 to the line containing the current finite line segment is calculated.
[0196] When the perpendicular foot is located between the two actual endpoints of the current finite line segment, the distance from the fire source spatial coordinate 74 to the perpendicular foot is taken as the shortest distance from the fire source spatial coordinate 74 to the current finite line segment. When the perpendicular foot is located outside the endpoints of the current finite line segment, the distances from the fire source spatial coordinate 74 to the two endpoints of the current finite line segment are calculated respectively, and the endpoint distance with the smaller value is selected as the shortest distance from the fire source spatial coordinate 74 to the current finite line segment.
[0197] The process iterates through all finite line segments that make up the actual centerline of the cable, selects the shortest distance from all the corresponding finite line segments, and defines this distance as the shortest spatial distance from the fire source's spatial coordinates 74 to the current cable component. This process ensures that the shortest distance always corresponds to the coverage area of the actual centerline of the cable, and when the fire source is close to the cable end, the distance calculation can be naturally transferred to the actual end point position.
[0198] The identifier of the bridge critical component 16, the shortest spatial distance from the fire source spatial coordinates 74 to the bridge critical component 16, and the coordinate direction difference between the fire source spatial coordinates 74 and the bridge critical component 16 are written into the fire source spatial positioning record. When the bridge fire response system needs to determine the bridge critical component closest to the fire source, the bridge critical components 16 are arranged in ascending order of the shortest spatial distance, and the bridge critical component 16 with the smallest distance is recorded as the adjacent critical component.
[0199] S44: Generate spatial location results of bridge fire sources and provide these results to bridge fire response equipment.
[0200] The spatial location results of a bridge fire source should include at least the fire identifier, the current data collection timestamp, the spatial coordinates of the fire source (74), the lane area where the fire source is located, the identifiers of nearby critical components, the shortest spatial distance from the fire source to the nearby critical components, and the current location status. The current location status should include at least a valid location status and a location pending confirmation status.
[0201] Step S32 obtains the current valid 3D position of the fire source that satisfies the spatial intersection condition. If the check condition of the corresponding branch in step S323 is also satisfied, a valid positioning status is output. Step S34 outputs a positioning pending confirmation status when the set of valid candidates corresponding to the current positioning cycle is empty.
[0202] See Figure 7A jet fire extinguishing device 80 is installed on the side of the bridge deck 11. The jet fire extinguishing device 80 includes a rotating base 81 and a spray nozzle 82. The jet fire extinguishing controller reads the effective positioning status corresponding to the spatial coordinates 74 of the fire source, and simultaneously reads the current position of the nozzle of the jet fire extinguishing device 80 in the bridge spatial coordinate system. The spray target direction 83 is formed by pointing from the current position of the nozzle to the spatial coordinates 74 of the fire source. According to the existing angle conversion relationship of the jet fire extinguishing device 80, the jet fire extinguishing controller converts the spray target direction 83 into the horizontal rotation angle corresponding to the rotating base 81 and the pitch angle corresponding to the spray nozzle 82, and then controls the rotating base 81 and the spray nozzle 82 to change the spray direction. The spatial positioning result of the bridge fire source can also be provided to the bridge monitoring center for marking the lane where the fire source is located and the adjacent key bridge components in the bridge spatial display interface.
[0203] When the current positioning status is "positioning pending confirmation," the jet fire extinguishing controller maintains the current spray control status or enters a preset waiting state. After a valid positioning status is re-established in subsequent positioning cycles, the jet fire extinguishing controller updates the spray target direction 83 according to the new fire source spatial coordinates 74.
[0204] Step S4 establishes a correspondence between the three-dimensional location of the same fire source and the actual lanes and key bridge components of the bridge. Lane determination uses the actual lane boundaries, and the relationship between key bridge components uses the actual finite spatial range of the components. Therefore, changes in the location of the fire source can be directly converted into changes in the bridge deck lane area and changes in the actual spatial distance between the fire source and the key bridge components.
[0205] This embodiment also provides a bridge fire source spatial positioning system based on multi-view stereo vision. The bridge fire source spatial positioning system includes a fire source area processing module, a candidate correspondence filtering module, a three-dimensional position calculation module, and a bridge spatial mapping module.
[0206] The fire source area processing module is configured to receive color and infrared images corresponding to the first observation unit 20 and the second observation unit 30, form a fire source area based on the color flame area and the infrared high temperature area within the same observation view, determine the candidate position interval and the number of boundary segments based on the size of the fire source area itself, and then form a first candidate position set 55 and a second candidate position set 56 from the boundary of the fire source area and the internal temperature distribution. The candidate correspondence filtering module is configured to read the device attitude information corresponding to the first observation unit 20 and the second observation unit 30, form a candidate position combination based on the candidate position category and the epipolar allowable distance, convert the candidate positions into spatial observation directions based on the camera parameters and device attitude information, calculate the spatial intersection distance corresponding to the candidate position combination and the candidate intersection position 66, form a valid candidate correspondence set based on the spatial intersection distance and the spatial boundary of the bridge entity, and then determine the target corresponding position from the valid candidate correspondence set.
[0207] The 3D position calculation module is configured to determine the disparity of the fire source image based on the target's corresponding position, calculate the current effective 3D position of the fire source based on the target's corresponding position, equipment attitude information, and spatial geometric relationships, perform a depth consistency check when the fixed binocular depth relationship usage conditions are met, and simultaneously determine the continuation status of the target correspondence relationship from the previous positioning cycle based on the nearest neighbor relationship of candidate positions in continuous positioning cycles. When the current target correspondence status enters a re-filtering state, the 3D position calculation module triggers the candidate correspondence filtering module to re-determine the target's corresponding position based on the candidate positions of the current positioning cycle.
[0208] The bridge spatial mapping module is configured to transform the current effective three-dimensional position of the fire source to the bridge spatial coordinate system. Based on the bridge lane boundaries and the actual spatial contours of the bridge's key components 16, it generates a spatial location result for the bridge fire source, which is then output via the data interface of the bridge fire response system. In the processor implementation, the fire source area processing module, candidate correspondence filtering module, three-dimensional position calculation module, and bridge spatial mapping module can be executed as program instructions stored in a computer-readable storage medium by the same processor, or they can be distributed across the edge computing unit of the observation equipment and the bridge monitoring server. Different deployment methods employ the same variable relationships and judgment rules for steps S1 to S4. In this embodiment, color images and infrared images are used to define the fire source area in the current observation view; different observation views form candidate positions based on the size of their own visible fire source areas; device attitude, epipolar relationship, and spatial intersection relationship are used to determine whether candidate positions in different observation views point to the same actual spatial area; the filtered target corresponding position is used to calculate the fire source image parallax and the three-dimensional position of the fire source; during continuous positioning, the correspondence status is determined based on the nearest neighbor relationship between the current candidate position and the target position in the previous positioning cycle, and the target corresponding position is re-formed based on the current candidate position set in the re-filtering state; the current effective three-dimensional position of the fire source is finally mapped to the actual spatial position of the bridge lane and key bridge components.
Claims
1. A method for spatially locating bridge fire sources based on multi-view stereo vision, characterized in that, include: A first set of candidate locations and a second set of candidate locations are formed based on the fire source area from each observation perspective; Based on the device attitude and spatial geometric relationship of the observation perspective, the candidate positions in the first candidate position set and the second candidate position set are paired, and the corresponding positions of the target are determined by filtering according to the spatial intersection distance of the corresponding spatial observation direction of each paired position. The fire source image parallax is determined from the location corresponding to the target, and the three-dimensional position of the fire source is calculated by combining the fire source image parallax, the spatial geometric relationship, and the device attitude. When the spatial intersection relationship of the target location does not meet the spatial intersection conditions, the target location is re-determined based on the first candidate location set and the second candidate location set of the current positioning cycle, the three-dimensional location of the fire source is recalculated, and the spatial positioning result of the bridge fire source is formed.
2. The method for spatial localization of bridge fire sources based on multi-view stereo vision according to claim 1, characterized in that, The fire source area includes: The infrared high-temperature region is mapped to the color image coordinate system according to the parameters corresponding to the image position. The minimum pixel distance from the pixel in the color flame region to the mapped infrared high-temperature region is calculated. Color flame pixels whose minimum pixel distance is not greater than the cross-band registration allowable distance are formed into the fire source region. The cross-band registration allowable distance is determined based on the registration residual distribution of the corresponding pixel positions of the same calibration target at different working distances.
3. The method for spatial localization of bridge fire sources based on multi-view stereo vision according to claim 1, characterized in that, The first candidate location set and the second candidate location set include: Contour tracing and internal temperature mapping are performed on the corresponding fire source regions to obtain the fire source boundary pixel sequence and temperature pixel set; The process of determining the interval of candidate locations is performed based on the size of the fire source area itself. Based on the candidate position interval, boundary candidate positions are extracted from the fire source boundary pixel sequence, and internal candidate positions are extracted from the temperature pixel set. The boundary candidate positions and internal candidate positions are then written into the corresponding candidate position set.
4. The method for spatial localization of bridge fire sources based on multi-view stereo vision according to claim 3, characterized in that, The candidate location interval determination process includes: Calculate the equivalent diameter of the current fire source area based on the current fire source area area, and determine the initial candidate position interval based on the equivalent diameter of the current fire source area; Based on the initial candidate position interval, internal candidate positions are selected. When the number of internal candidate positions exceeds the upper limit, the candidate position interval is increased and internal candidate positions are reselected until the number of internal candidate positions meets the limit or the candidate position interval reaches the corresponding upper limit.
5. The method for spatial localization of bridge fire sources based on multi-view stereo vision according to claim 1, characterized in that, The filtering includes: Establish cross-viewpoint candidate position pairings according to candidate position categories, obtain the corresponding epipolar line of the first candidate position in another observation view based on the dual-target calibration parameters, and retain the pairings that are no more than the allowable distance of the epipolar line and have the same candidate position category as the corresponding epipolar line. The paired positions are converted into spatial observation directions, the nearest point and spatial intersection distance of the two spatial observation directions are calculated, and a valid candidate correspondence set is formed based on the spatial intersection distance and the candidate intersection positions.
6. The method for spatial localization of bridge fire sources based on multi-view stereo vision according to claim 5, characterized in that, The spatial rendezvous distance includes: The Euclidean distance between the first and second nearest points along two spatial observation directions is taken as the spatial intersection distance, and the midpoint of the line connecting the first and second nearest points is taken as the candidate intersection position. A spatial intersection residual sample set is formed based on the repeated positioning data of known spatial coordinate test points within the target working distance range, and the spatial intersection distance threshold is determined based on the quantile value of the spatial intersection residual sample set.
7. The method for spatial localization of bridge fire sources based on multi-view stereo vision according to claim 5, characterized in that, The target location includes: When the set of valid candidate correspondences contains more than two valid candidate correspondences, they are sorted according to spatial intersection distance. When the difference in spatial intersection distance between adjacent valid candidate corresponding records is greater than the spatial intersection distance resolvable threshold, the valid candidate corresponding record with the smaller spatial intersection distance is selected. When the difference in spatial intersection distance is not greater than the resolvable threshold of spatial intersection distance, the corresponding position of the target is determined based on the spatial distance between the candidate intersection position and the three-dimensional position of the effective fire source in the previous positioning cycle.
8. The method for spatial localization of bridge fire sources based on multi-view stereo vision according to claim 7, characterized in that, The spatial intersection distance resolvable threshold includes: Spatial intersection calculations are repeatedly performed on control points with the same known spatial location to form a repeated spatial intersection distance sequence; Calculate the median of the repeated spatial intersection distance sequence and the absolute deviation between each spatial intersection distance and the median, and determine the resolvable threshold of the spatial intersection distance based on the distribution of the absolute deviation.
9. The method for spatial localization of bridge fire sources based on multi-view stereo vision according to claim 1, characterized in that, The three-dimensional location of the fire source includes: The positions corresponding to the first and second observation angles in the target positions are determined as the first target position and the second target position, respectively. The first target spatial observation direction and the second target spatial observation direction are formed according to the first target position and the second target spatial observation direction, respectively. The closest point of the first target and the closest point of the second target in the two target spatial observation directions are obtained. The midpoint of the line connecting the closest point of the first target and the closest point of the second target is determined as the three-dimensional position of the fire source. In a fixed binocular setup, when the corrected first target position and second target position are located in a common effective image area and the longitudinal position difference is less than or equal to the epipolar allowable distance, the initial depth is obtained based on the parallax of the fire source image, and a consistency check is performed between the initial depth and the depth corresponding to the three-dimensional position of the fire source.
10. A method for spatially locating bridge fire sources based on multi-view stereo vision according to claim 3, characterized in that, The process of redetermining the target's location includes: Using the target location from the previous positioning cycle as the search center for the current positioning cycle, the nearest candidate location of the same category is determined in the current first candidate location set and second candidate location set respectively; The average sampling interval of the boundary is determined based on the total arc length of the fire source boundary in the current fire source area and the number of candidate positions of the boundary. The nearest candidate position is then determined to be a continuous candidate position based on the average sampling interval of the boundary or the candidate position interval. For pairs of candidate positions, the spatial intersection distance is recalculated. If the spatial intersection condition is met, the target position is updated. If the spatial intersection condition is not met, the candidate position filtering is re-executed.
11. The method for spatial localization of bridge fire sources based on multi-view stereo vision according to claim 1, characterized in that, The spatial location results of the bridge fire source include: The three-dimensional location of the fire source is transformed into the bridge spatial coordinate system to obtain the spatial coordinates of the fire source. The spatial coordinates of the fire source are then projected onto the bridge surface along the bridge surface normal. The lane area where the fire source is located is determined based on the inclusion relationship between the fire source projection point on the bridge surface and the lane boundary. The nearest key components are determined based on the shortest spatial distance between the spatial coordinates of the fire source and the actual spatial outline of the key components of the bridge, thus forming the spatial positioning result of the bridge fire source.
12. A bridge fire source spatial positioning system based on multi-view stereo vision, characterized in that, To implement the method according to any one of claims 1 to 11, comprising: The fire source area processing module is used to form a first candidate location set and a second candidate location set based on the fire source area from each observation perspective; The candidate correspondence filtering module is used to perform spatial intersection filtering on the candidate positions in the first candidate position set and the second candidate position set according to the device attitude and spatial geometric relationship of the observation perspective, so as to determine the target correspondence position; The three-dimensional position calculation module is used to determine the parallax of the fire source image based on the target's corresponding position, calculate the three-dimensional position of the fire source by combining the fire source image parallax, the spatial geometric relationship, and the device attitude, and re-determine the target's corresponding position when the spatial intersection relationship of the target's corresponding position does not meet the spatial intersection condition. The bridge spatial mapping module is used to generate spatial positioning results of the bridge fire source based on the three-dimensional location of the fire source.
