A binocular-based long-distance flame positioning method and device

Abstract

The present invention relates to the technical field of flame detection, and particularly to a binocular-based long-distance flame positioning method and a long-distance flame positioning device to which the positioning method is applied. The present invention uses a high-resolution binocular camera and increases the baseline of the two cameras, so that when observing a long-distance flame, the targets in the left and right views still have a certain parallax, reducing the coordinate error. The present invention first performs rough positioning on the high-resolution image to determine the approximate range, and then intercepts the high-resolution image, and performs fine positioning under the condition of reducing the amount of calculation data, thereby reducing the calculation overhead and ensuring the calculation accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of flame detection, and particularly to a binocular long-distance flame positioning method and a long-distance flame positioning device using the positioning method. Background Art

[0002] Most of the current binocular flame positioning studies are for short-distance scenarios less than ten meters, which have limitations in practical applications - large errors will occur when applied to long distances (greater than 100 meters). This is because the estimated distance of depth is inversely proportional to the parallax. Therefore, when positioning a long-distance target, since the parallax of the target object in the left and right images is very small, when mapping to the three-dimensional space to calculate the world coordinates, a small error generated by image matching will be amplified into a large world coordinate error.

[0003] To reduce the error, it is necessary to increase the parallax of the target object. To increase the parallax, it is necessary to increase the baseline length, improve the image resolution, and increase the camera focal length. However, a large parallax will make the search space of stereo matching too large, significantly increasing the computational cost of stereo matching, and even exceeding the upper limit of the parallax search range of the stereo matching model (the current upper limit of the parallax of the mainstream model is 192 pixels), resulting in problems where correct matching cannot be achieved. At the same time, a large parallax depends on the image also having a large resolution, which further increases the computational cost. Summary of the Invention

[0004] Based on this, in view of the problem of high computational cost of the existing binocular flame positioning method when positioning long-distance flames, a binocular long-distance flame positioning method and device are provided.

[0005] The present invention is implemented by the following technical solutions:

[0006] In a first aspect, the present invention discloses a binocular long-distance flame positioning method, which performs long-distance flame positioning through a binocular camera acting on the same target flame. The binocular camera includes a high-resolution camera one on the left and a high-resolution camera two on the right.

[0007] The binocular long-distance flame positioning method includes the following steps:

[0008] Step 1, calibrate the binocular camera; then perform alignment correction on the binocular camera to make the images output by the two cameras parallel;

[0009] Step 2, obtain the images PL_0 and PR_0 taken by the high-resolution camera one and the high-resolution camera two at the same time, and form a high-resolution picture pair; wherein, the sizes of PL_0 and PR_0 are both W*H, W represents the length of PL_0 and PR_0, and H represents the width of PL_0 and PR_0;

[0010] Perform preliminary flame recognition on the image PL_0 of the high-resolution camera 1 to obtain the approximate pixel coordinate area A of the flame in PL_0; where A extends from u1 on the left to u2 on the right, from v2 at the top to v1 at the bottom;

[0011] Step three, downsample the high-resolution image pair to obtain PL_1 and PR_1, forming a low-resolution image pair; where the sizes of PL_1 and PR_1 are both w*h, w represents the length of PL_1 and PR_1, and h represents the width of PL_1 and PR_1;

[0012] Then perform stereo matching on the low-resolution image pair to calculate the low-resolution disparity disp1 of the flame;

[0013] Step four, calculate the center point (u0, v0) of A, where u0 = (u1 + u2) / 2 and v0 = (v1 + v2) / 2; then, with (u0, v0) as the center, intercept the area B in PL_0 and the area C in PR_0;

[0014] Among them, B extends from u0 - w / 2 on the left to u0 + w / 2 on the right, from v0 + h / 2 at the top to v0 - h / 2 at the bottom;

[0015] C extends from u0 - w / 2 - (N - M)*disp1 on the left to u0 + w / 2 - (N - M)*disp1 on the right, from v0 + h / 2 at the top to v0 - h / 2 at the bottom; in the formula, N is the downsampling multiple in the length direction, N = W / w; M is the compensation coefficient;

[0016] Then perform stereo matching on B and C to calculate the high-resolution disparity information disp2 of the flame;

[0017] Step five, calculate the precise positioning disparity disp3 = (N - M)*disp1 + disp2, and then calculate the flame depth based on disp3, and further calculate the precise coordinates of the flame.

[0018] The implementation of this binocular long-distance flame positioning method is based on the method or process of the embodiments of the present disclosure.

[0019] The present invention uses high-resolution binocular cameras and increases the baseline of the two cameras to ensure that there is still a certain disparity between the targets in the left and right views when observing long-distance flames, reducing the coordinate error; the present invention first uses rough positioning of high-resolution images to determine the approximate range, and then intercepts the high-resolution images, and performs fine positioning while reducing the amount of calculation data, thereby reducing the calculation overhead and ensuring the calculation accuracy, and solving the problem of large calculation overhead of the existing binocular flame positioning method when positioning long-distance flames.

[0020] In a second aspect, the present invention discloses a binocular long-distance flame positioning device, which includes a flame image acquisition unit and a data processing and calculation unit.

[0021] The flame image acquisition unit includes a stage and a binocular camera that acts on the same target flame; the binocular camera includes a high-resolution camera one and a high-resolution camera two. The high-resolution camera one is located on the left side of the stage, and the high-resolution camera two is located on the right side of the stage.

[0022] The data processing and calculation unit is used to process the images collected by the flame image acquisition unit according to the binocular long-distance flame positioning method disclosed in the first aspect to obtain the accurate coordinates of the flame.

[0023] The implementation of this binocular long-distance flame positioning device is based on the method or process of the embodiments disclosed in this disclosure.

[0024] In a third aspect, the present invention discloses a readable storage medium. Computer program instructions are stored in this readable storage medium. When the computer program instructions are read and run by a processor, they execute the binocular long-distance flame positioning method disclosed in the first aspect.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] The present invention uses high-resolution binocular cameras and increases the baseline of the two cameras, so as to ensure that there is still a certain parallax between the targets in the left and right views when observing long-distance flames, reducing the coordinate error; the present invention first uses rough positioning of high-resolution pictures to determine the general range, and then intercepts the high-resolution pictures, and performs fine positioning while reducing the amount of calculation data, thereby reducing the calculation overhead and ensuring the calculation accuracy. Description of the Drawings

[0027] Figure 1 It is a structural diagram of the binocular long-distance flame positioning device proposed in Embodiment 1 of the present invention;

[0028] Figure 2 It is a flowchart of the binocular long-distance flame positioning method provided in Embodiment 1 of the present invention;

[0029] Figure 3 It is the PL_1 diagram obtained from the experiment in Embodiment 2 of the present invention.

[0030] Figure 4 It is the low-resolution parallax diagram obtained from the experiment in Embodiment 2 of the present invention;

[0031] Figure 5 It is the B diagram obtained from the experiment in Embodiment 2 of the present invention.

[0032] Figure 6 It is the high-resolution parallax diagram obtained from the experiment in Embodiment 2 of the present invention;

[0033] Figure 7 This is the final result diagram of the experiment in Embodiment 2 of the present invention;

[0034] Figure 8 is Figure 7 a partially enlarged view of.

[0035] In the attached drawings, the list of components represented by each label is as follows:

[0036] 1. First high-resolution camera, 2. Second high-resolution camera, 3. Carriage, 4. Server, 5. External screen. Specific implementation manner

[0037] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0038] It should be noted that when a component is referred to as being "installed on" another component, it can be directly on the other component or there may also be an intermediate component. When a component is considered to be "set on" another component, it can be directly set on the other component or there may be an intermediate component at the same time. When a component is considered to be "fixed to" another component, it can be directly fixed to the other component or there may be an intermediate component at the same time.

[0039] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs. The terms used herein in the description of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "or / and" used herein includes any and all combinations of one or more of the related listed items.

[0040] Embodiment 1

[0041] Refer to Figure 1 , which is the structural diagram of the binocular long-distance flame positioning device proposed in Embodiment 1 of the present invention, including: a flame image acquisition unit and a data processing and calculation unit.

[0042] The flame image acquisition unit includes a carriage 3 and a binocular camera acting on the same target flame. The binocular camera includes a first high-resolution camera 1 and a second high-resolution camera 2. The first high-resolution camera 1 is located on the left side of the carriage 3, and the second high-resolution camera 2 is located on the right side of the carriage 3.

[0043] For the requirement of long - distance flame positioning, when the flame is more than 100m away from the binocular camera, the baseline between the binocular cameras should be at least greater than 60cm. Generally, the baseline is recommended to be set greater than 1m.

[0044] The high - resolution camera 1 and the high - resolution camera 2 have the same resolution. Generally, the resolution is not less than 20.2MP. In this embodiment 1, both cameras use the Hikrobot CE - 200 - 10UV industrial array cameras, and are additionally equipped with adjustable - focal - length lenses MVL - KF5028 - 12MP, with a maximum resolution of 5472*3648. The flatness requirement of the carrier 3 is relatively high, and it is necessary to ensure that the two installed cameras are parallel as much as possible. In this embodiment 1, the carrier 3 uses a 2.5m long flat rail, and there are two sliders with adjustable pitch angle scales on the rail. The two cameras are respectively installed on the two sliders, and their postures can be made consistent by adjusting the pitch angle.

[0045] The data processing and calculation unit is used to process the images collected by the flame image acquisition unit according to the binocular - based long - distance flame positioning method to obtain the accurate coordinates of the flame. In this embodiment 1, the data processing and calculation unit uses the built - in GPU server 4, which can run the software based on the binocular - based long - distance flame positioning method internally, or can be built - in with several functional modules. Considering the high resolution of the binocular cameras, the length and width of the images they collect are also based on the highest resolution standard - 5472*3648, and the file size can reach 60MB. Therefore, the binocular cameras and the GPU server 4 should ensure sufficient communication bandwidth - RJ45 network cable connection or USB3.0 data connection can be used.

[0046] If the GPU server 4 runs the software based on the binocular - based long - distance flame positioning method internally, see Figure 2 which is the flowchart of the binocular - based long - distance flame positioning method, and it is to perform long - distance flame positioning through the binocular cameras acting on the same target flame.

[0047] That is, this embodiment 1 also synchronously discloses the binocular - based long - distance flame positioning method, including the following steps:

[0048] Step 1, calibrate the binocular cameras; then perform alignment correction on the binocular cameras to make the images output by the two cameras parallel.

[0049] It should be noted that the binocular cameras include the high - resolution camera 1 on the left and the high - resolution camera 2 on the right, and both act on the same target flame.

[0050] Calibrate the binocular cameras. For example, use the general camera calibration process such as the Zhang Zhengyou method, and the internal and external parameters of the binocular cameras and other parameters can be obtained, which are used to calculate the accurate coordinates of the flame according to the obtained accurate positioning parallax disp3 later.

[0051] Aligning and calibrating a binocular camera can correct the spatial relationship error between the left and right cameras, making the pictures taken by the binocular camera parallel, that is, aligned in the horizontal direction.

[0052] A high-precision aluminum chessboard calibration plate can be selected to take synchronous pictures at different distances. By changing the inclination angle of the calibration plate, multiple groups of binocular picture pairs are taken, and the binocular camera is calibrated and aligned using the binocular picture pairs.

[0053] Step 2: Obtain images PL_0 and PR_0 taken by high-resolution camera 1 and high-resolution camera 2 at the same time, and form a high-resolution picture pair. Among them, the sizes of PL_0 and PR_0 are both W*H, where W represents the length of PL_0 and PR_0, and H represents the width of PL_0 and PR_0. In this embodiment 1, W = 5472 and H = 3648.

[0054] Here, to obtain the images taken by the binocular camera at the same time, the SDK provided by the camera hardware manufacturer can be used to develop the synchronization logic and use the semaphore mechanism to ensure that the synchronization error of the binocular camera during the frame acquisition process is at the millisecond level.

[0055] Perform preliminary flame recognition on the image PL_0 of high-resolution camera 1 to obtain the approximate pixel coordinate area A of the flame in PL_0; among them, A extends from u1 on the left to u2 on the right, from v2 on the top to v1 on the bottom. That is, A is rectangular, and the coordinates of its four corners are (u1, v1), (u1, v2), (u2, v1), and (u2, v2) respectively.

[0056] In other words, A = (u A , v A ), u1 < u A < u2, v1 < v A < v2; u1 is the minimum value of A in the u direction in PL_0, u2 is the maximum value of A in the u direction in PL_0, v1 is the minimum value of A in the v direction in PL_0, and v2 is the maximum value of A in the v direction in PL_0.

[0057] Among them, there are many methods for preliminary flame recognition. For example, a flame detection method based on color space, a flame detection method based on deep learning, a flame detection method based on multi-feature fusion, etc. can be selected according to actual needs.

[0058] It should be noted that the preliminary flame recognition can be set to be processed on the GPU server 4, or can be set in the flame image acquisition unit and executed by the binocular camera. Since the binocular camera shooting requires a debugging process, flames may not necessarily be captured in the initial stage. And due to the large file sizes of PL_0 and PR_0, if no flame is detected, it is meaningless to perform subsequent operations on the high-resolution picture pairs. Therefore, it can be set that if it is recognized that there is no flame in the currently acquired PL_0, the high-resolution picture pair of this time is discarded, and then the next acquisition and recognition are performed; if it is recognized that there is a flame in the currently acquired PL_0, the high-resolution picture pair of this time is subjected to the operation of Step 3.

[0059] Step 3: Downsample the high-resolution picture pair to obtain PL_1 and PR_1, which form a low-resolution picture pair; where the sizes of PL_1 and PR_1 are both w*h, w represents the length of PL_1 and PR_1, and h represents the width of PL_1 and PR_1.

[0060] Then perform stereo matching on the low-resolution picture pair to calculate the low-resolution disparity disp1 of the flame.

[0061] This step is to reduce the resolution of the high-resolution picture pair, thereby reducing the computational overhead of stereo matching. Here, the obtained disp1 is the disparity at low resolution, storing a large error.

[0062] Generally, the size of the low-resolution picture pair should be adapted to the requirements of the stereo matching model used. In this Embodiment 1, the stereo matching model used is the IGEV-Stereo model, and the size of the picture processed by this model can be 640*480 (i.e., w = 640, h = 480).

[0063] In this way, the downsampling multiple N in the length direction = W / w, which is correspondingly set to 8.55; the downsampling multiple Q in the width direction = H / h, which is correspondingly set to 7.6.

[0064] Of course, if other models are used, such as the ACVnet model, the CREStereo model, etc., the required low-resolution picture sizes will also be different, and N and Q should be adjusted accordingly.

[0065] Step 5: Calculate the center point (u0, v0) of A, where u0 = (u1 + u2) / 2 and v0 = (v1 + v2) / 2; then, with (u0, v0) as the center, intercept the region B in PL_0 and intercept the region C in PR_0;

[0066] Among them, B extends from u0 - w / 2 to u0 + w / 2 on the left and right, and from v0 + h / 2 to v0 - h / 2 on the top and bottom. That is, B is rectangular, and its coordinates at the four corners on PL_0 are (u0 - w / 2, v0 - h / 2), (u0 - w / 2, v0 + h / 2), (u0 + w / 2, v0 - h / 2), and (u0 + w / 2, v0 + h / 2) respectively.

[0067] C extends from u0 - w / 2 - (N - M)*disp1 to u0 + w / 2 - (N - M)*disp1 on the left and right, and from v0 + h / 2 to v0 - h / 2 on the top and bottom. That is, C is rectangular, and its coordinates at the four corners on PR_0 are (u0 - w / 2 - (N - M)*disp1, v0 - h / 2), (u0 + w / 2 - (N - M)*disp1, v0 + h / 2), (u0 - w / 2 - (N - M)*disp1, v0 - h / 2), and (u0 + w / 2 - (N - M)*disp1, v0 + h / 2) respectively.

[0068] In other words, B = (u B , v B ), where u0 - w / 2 < u B < u0 + w / 2, and v0 - h / 2 < v B < v0 + h / 2; C = (u C , v C ), where u0 - w / 2 - (N - M)*disp1 < u C < u0 + w / 2 - (N - M)*disp1, and v0 - h / 2 < v C < v0 + h / 2.

[0069] Then, perform stereo matching on B and C to calculate the high - resolution parallax information disp2 of the flame.

[0070] This step is to extract the part where the flame is located from the high - resolution picture pair based on the approximate pixel coordinate area A, so as to reduce the picture size while retaining the flame information, thereby reducing the computational overhead of stereo matching.

[0071] It should be noted that due to downsampling, parallax deviation may occur. Therefore, there is an offset of (N - M)*disp1 for PR_0. Among them, N*disp1 is to recover the impact caused by downsampling, and M*disp1 is for compensating parallax to prevent the situation where the right picture is more to the left than the left picture, resulting in a negative parallax. (N - M)*disp1 constitutes the parallax caused by step five. Among them, M is a compensation coefficient, and its value is adjusted according to the actual situation. It is recommended to take 1.

[0072] Step 6: Calculate the precise positioning parallax of the flame disp3 = (N - M) * disp1 + disp2, and then calculate the flame depth based on disp3, and further calculate the precise coordinates of the flame.

[0073] In this Step 6, first calculate the precise positioning parallax disp3 of the flame according to (N - M) * disp1 and disp2.

[0074] Since there is the following relationship between parallax and depth:

[0075]

[0076] In the formula, b represents the baseline length of the binocular camera, ΔX represents the parallax, f represents the focal length of the camera, and D is the depth of the observed target.

[0077] Then, based on disp3, the flame depth can be obtained

[0078] Since there is a projection relationship between the camera - taken photo and the real world, the world coordinates of the observed target can be calculated through the photo pixel coordinates, and the mathematical relationship can be expressed as:

[0079]

[0080] In the formula, D is the depth of the observed target, u and v represent the pixel coordinates of the observed target in the photo space; represents the internal parameter of the camera, represents the external parameter of the camera, and X, Y, Z represent the world coordinates of the observed target.

[0081] Then, based on the flame depth D fire , the internal and external parameters of the high - resolution camera 1 obtained in Step 1, and the flame pixel coordinates (u, v) in Region A, substitute them into the above formula to obtain the precise coordinates of the flame. It should be noted that the flame pixel coordinates (u, v) in Region A are taken at a position slightly below the center of Region A, corresponding to the flame core.

[0082] If the GPU server 4 has several built - in functional modules, the functional modules include: camera calibration module, camera correction module, high - resolution picture pair acquisition module, preliminary flame recognition module, low - resolution picture pair acquisition module, flame region intercepting module, stereo matching module, and coordinate calculation module.

[0083] Among them, the camera calibration module is used to calibrate the binocular camera. The camera correction module is used to perform alignment correction on the binocular camera to make the images output by the two cameras parallel.

[0084] The high-resolution image pair acquisition module is used to acquire images PL_0 and PR_0 taken by high-resolution camera 1 and high-resolution camera 2 at the same time, and form a high-resolution image pair. Among them, the sizes of PL_0 and PR_0 are both W*H, where W represents the length of PL_0 and PR_0, and H represents the width of PL_0 and PR_0.

[0085] The preliminary flame recognition module is used to perform preliminary flame recognition on the image PL_0 of high-resolution camera 1, and obtain the approximate pixel coordinate area A of the flame in PL_0. Among them, A extends from u1 on the left to u2 on the right, from v2 on the top to v1 on the bottom.

[0086] The low-resolution image pair acquisition module is used to downsample the high-resolution image pair to obtain PL_1 and PR_1, and form a low-resolution image pair. Among them, the sizes of PL_1 and PR_1 are both w*h, where w represents the length of PL_1 and PR_1, and h represents the width of PL_1 and PR_1.

[0087] The flame area intercepting module is used to calculate the center point (u0, v0) of A, where u0 = (u1 + u2) / 2 and v0 = (v1 + v2) / 2; then, with (u0, v0) as the center, intercept the area B in PL_0 and the area C in PR_0. Among them, B extends from u0 - w / 2 on the left to u0 + w / 2 on the right, from v0 + h / 2 on the top to v0 - h / 2 on the bottom;

[0088] C extends from u0 - w / 2 - (N - M)*disp1 on the left to u0 + w / 2 - (N - M)*disp1 on the right, from v0 + h / 2 on the top to v0 - h / 2 on the bottom; N = W / w.

[0089] The stereo matching module is used to perform stereo matching on the low-resolution image pair to calculate the low-resolution disparity disp1 of the flame; it is also used to perform stereo matching on B and C to calculate the high-resolution disparity information disp2 of the flame.

[0090] The coordinate calculation module is used to calculate the precise positioning disparity disp3 = (N - M)*disp1 + disp2 of the flame; then, based on disp3, calculate the flame depth, and further calculate the precise coordinates of the flame.

[0091] Of course, an external screen 5 can also be set up for the coordinate calculation module to output the precise coordinates of the flame to the external screen 5. The coordinate calculation module can also save the precise coordinates of the flame as a readable log file for convenient data collation.

[0092] Embodiment 2

[0093] Embodiment 2 discloses a readable storage medium in which computer program instructions are stored. When the computer program instructions are read and run by a processor, the method for long-distance flame positioning based on binocular vision disclosed in Embodiment 1 is executed.

[0094] When the method of Embodiment 1 is applied, it can be applied in the form of software, such as designed as a program that can run independently on a computer-readable storage medium. The computer-readable storage medium can be a USB flash drive, and a program for starting the entire method by external trigger is designed through the USB flash drive.

[0095] Embodiment 3

[0096] The inventor conducted experimental verification on the method of Embodiment 1. A long-distance flame positioning device based on binocular vision was set up in the evening of a certain day. Both the high-resolution camera 1 and the high-resolution camera 2 used the Hikrobot CE-200-10UV industrial area camera, and were also equipped with an adjustable focal length lens MVL-KF5028-12MP with a resolution of 5472*3648. The stage 3 used a 2.5m long straight guide rail, and the baseline of the binocular camera was 1m.

[0097] The flame was 150m away from the left camera, and the true coordinates [X’, Y’, Z’] were measured by laser as [-7758.1192, 3932.8852, 142752], with the coordinate unit being mm.

[0098] The method of Embodiment 1 was implemented using the alignment method of the binocular long-distance flame positioning device:

[0099] First, the high-resolution camera 1 and the high-resolution camera 2 were calibrated and alignment corrected to make the pictures taken by the binocular camera aligned in the horizontal direction.

[0100] Initial flame recognition was performed on the image PL_0 of the high-resolution camera 1 to obtain A=(u1, v1). Then, the high-resolution picture pair was downsampled to obtain a low-resolution picture pair, as shown in ( Figure 3 (only PL_1 is shown). Then, the low-resolution picture pair was stereoscopically matched using the IGEV-Stereo model to obtain a low-resolution disparity map as shown in ( Figure 3 and the low-resolution disparity disp1 of the flame was calculated. Figure 4 The low-resolution disparity disp1 of the flame was calculated.

[0101] Next, the area B in PL_0 was intercepted, and the area C in PR_0 was intercepted to obtain B and C, as shown in ( Figure 5 (only B is shown). Among them, M was taken as 1. Then, B and C were stereoscopically matched to obtain a high-resolution disparity map as shown in ( Figure 5 and the high-resolution disparity information disp2 of the flame was calculated. Figure 6 The high-resolution disparity information disp2 of the flame was calculated.

[0102] Finally, calculate the precise positioning parallax disp3 of the flame and calculate it step by step. The specific process will not be elaborated here. Refer to Figure 7 、 Figure 8 , and finally obtain the calculated coordinates of the flame [X, Y, Z] as [-7735.61, 3921.4745, 142337.8228], with the coordinate unit of mm.

[0103] Compared with the true coordinates, the error between the calculated coordinates indicates that the method of Embodiment 1 can achieve better calculation accuracy.

[0104] It should be noted that the applicant has simultaneously submitted color drawings as other supporting documents.

[0105] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as these combinations of technical features do not conflict, they should be considered to be within the scope described in this specification.

[0106] The above-described embodiments merely represent several implementation manners of the present invention. Their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of the invention patent. It should be pointed out that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the appended claims.

Claims

1. A binocular long-distance flame positioning method, which performs long-distance flame positioning through binocular cameras acting on the same target flame; the binocular cameras include a high-resolution camera one on the left and a high-resolution camera two on the right, and is characterized in that, When the distance between the flame and the binocular camera exceeds 100 m, the baseline between the binocular cameras is greater than 60 cm; The binocular-based long-distance flame positioning method includes the following steps: Step 1: Calibrate the binocular camera; then perform alignment correction on the binocular camera to make the images output by the two cameras parallel; Step 2: Obtain images PL_0 and PR_0 captured by high-resolution camera 1 and high-resolution camera 2 at the same time, and form a high-resolution picture pair; where the sizes of PL_0 and PR_0 are both W*H, W represents the length of PL_0 and PR_0, and H represents the width of PL_0 and PR_0; Perform preliminary flame recognition on the image PL_0 of high-resolution camera 1 to obtain the approximate pixel coordinate area A of the flame in PL_0; where A extends from u1 on the left to u2 on the right, from v2 on the top to v1 on the bottom; Step 3: Downsample the high-resolution picture pair to obtain PL_1 and PR_1, and form a low-resolution picture pair; where the sizes of PL_1 and PR_1 are both w*h, w represents the length of PL_1 and PR_1, and h represents the width of PL_1 and PR_1; Then perform stereo matching on the low-resolution picture pair to calculate the low-resolution disparity disp1 of the flame; Step 4: Calculate the center point (u0, v0) of A, where u0 = (u1 + u2) / 2 and v0 = (v1 + v2) / 2; then, with (u0, v0) as the center, intercept the area B in PL_0 and intercept the area C in PR_0; Among them, B extends from u0 - w / 2 on the left to u0 + w / 2 on the right, from v0 + h / 2 on the top to v0 - h / 2 on the bottom; C to u0 - w / 2 - ( N - M ) * disp1, right to u0 + w / 2 - ( N - M ) * disp1, up to v0 + h / 2, down to v0 - h / 2; where N is the downsampling factor in the length direction, N = W / w; M is the compensation factor; Then perform stereo matching on B and C to calculate the high-resolution disparity information disp2 of the flame; Step 5, calculate the precise positioning parallax of the flame disp3 = (< N - M ) * disp1 + disp2, then calculate the flame depth based on disp3, and further calculate the precise coordinates of the flame.

2. The binocular long-distance flame positioning method according to claim 1, wherein The resolutions of high-resolution camera 1 and high-resolution camera 2 are the same.

3. The binocular long-distance flame positioning method according to claim 1, characterized in that In Step 1, the internal and external parameters of the two cameras are obtained using a general camera calibration process.

4. The binocular long-distance flame positioning method according to claim 1, characterized in that, The method for performing preliminary flame recognition in Step 2 includes, but is not limited to, using a flame detection method based on the RGB space, a flame detection method based on deep learning, and a flame detection method based on multi-feature fusion.

5. The binocular long-distance flame positioning method according to claim 1, wherein Stereo matching is performed in both Step 3 and Step 4 using a stereo matching model; The stereo matching model includes: IGEV-Stereo model, ACVnet model, CREStereo model.

6. The binocular long-distance flame positioning method according to claim 1, characterized in that M =1。 7. A binocular-based long-distance flame positioning device, characterized in that, Including: A flame image capturing unit, which includes a stage and a binocular camera that acts on the same target flame; the binocular camera includes high-resolution camera 1 and high-resolution camera 2, high-resolution camera 1 is located on the left side of the stage, and high-resolution camera 2 is located on the right side of the stage; and A data processing and calculation unit, which is used to process the images collected by the flame image capturing unit according to the binocular-based long-distance flame positioning method described in any one of claims 1-6 to obtain the accurate coordinates of the flame.

8. The binocular long-distance flame positioning device according to claim 7, wherein The data processing and calculation unit includes: A camera calibration module, which is used to calibrate the binocular camera; A camera correction module, which is used to perform alignment correction on the binocular camera to make the images output by the two cameras parallel; A high - resolution image pair acquisition module, which is used to acquire images PL_0 and PR_0 taken by high - resolution camera 1 and high - resolution camera 2 at the same time, and form a high - resolution image pair; wherein, the sizes of PL_0 and PR_0 are both W*H, W represents the length of PL_0 and PR_0, and H represents the width of PL_0 and PR_0. A preliminary flame recognition module, which is used to perform preliminary flame recognition on the image PL_0 of high - resolution camera 1 to obtain the approximate pixel coordinate area A of the flame in PL_0; wherein, A extends from u1 on the left to u2 on the right, from v2 on the top to v1 on the bottom. A low - resolution image pair acquisition module, which is used to downsample the high - resolution image pair to obtain PL_1 and PR_1, and form a low - resolution image pair; wherein, the sizes of PL_1 and PR_1 are both w*h, w represents the length of PL_1 and PR_1, and h represents the width of PL_1 and PR_1. A flame region intercepting module, which is used to calculate the center point (u0, v0) of A, where u0=(u1 + u2) / 2 and v0=(v1 + v2) / 2; then, with (u0, v0) as the center, intercept the region B in PL_0 and intercept the region C in PR_0. Wherein, B extends from u0 - w / 2 on the left to u0 + w / 2 on the right, from v0 + h / 2 on the top to v0 - h / 2 on the bottom. C to the left is u0 - w / 2 - ( N - M ) * disp1, C to the right is u0 + w / 2 - ( N - M ) * disp1, C to the top is v0 + h / 2, and C to the bottom is v0 - h / 2; where N is the downsampling factor in the length direction, N = W / w; M is the compensation factor; A stereo matching module, which is used to perform stereo matching on the low - resolution image pair to calculate the low - resolution disparity disp1 of the flame; is also used to perform stereo matching on B and C to calculate the high - resolution disparity information disp2 of the flame; and A coordinate calculation module, which is used to calculate the precise positioning parallax disp3 of the flame = ( N - M ) * disp1 + disp2, and then calculate the flame depth based on disp3, and further calculate the precise coordinates of the flame.

9. The binocular long-distance flame positioning device according to claim 8, characterized in that, The coordinate calculation module outputs the accurate coordinates of the flame to an external screen and saves them as a readable log file.

10. A readable storage medium, characterized in that, The readable storage medium stores computer program instructions. When the computer program instructions are read and run by a processor, they execute the binocular long - distance flame positioning method according to any one of claims 1 - 7.

Images