Target detection method and system based on improved image matching

Abstract

The application relates to a target detection method and system based on improved image matching, and belongs to the technical field of optical detection, which solves the problem of low image matching accuracy and difficulty in realizing accurate target detection in the prior art. The detection method comprises the following steps: image matching is performed on N camera images, a virtual-real fusion method is used to fuse the N camera images with a pre-constructed target imaging background digital model; a digital model containing a target is used to update each non-edge feature point, and the two-dimensional coordinates of each feature point in the corresponding camera image, the three-dimensional coordinates of each camera position, the shooting parameters of the camera and the two-dimensional coordinates of all the pixel points of each image corresponding to each non-edge feature point after the update are obtained; the three-dimensional coordinates of all the non-edge feature points are fused with a target preliminary model to obtain a target three-dimensional topography, so that the target is detected. The accurate detection of the three-dimensional topography of the target to be detected is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of optical detection technology, and in particular to a target detection method and system based on improved image matching. BACKGROUND

[0002] With the development of artificial intelligence and other technologies, the perception, recognition and measurement of targets are increasingly applied in many application fields such as transportation, security, medical treatment and construction.

[0003] Current target detection involves various technical means such as optical imaging, laser detection and vibration analysis. However, when the similarity of the feature points of the target is high, the phenomenon of feature point matching error is prone to occur, which ultimately leads to inaccurate target detection.

[0004] Therefore, there is an urgent need for a technical solution for accurate detection of the three-dimensional topography of a target. SUMMARY

[0005] In view of the above analysis, the embodiments of the present application aim to provide a target detection method and system based on improved image matching to solve the problem of low image matching accuracy in the prior art, which makes it difficult to achieve accurate target detection.

[0006] In one aspect, the embodiments of the present application provide a target detection method based on improved image matching, which comprises:

[0007] synchronously acquiring N camera images obtained by imaging a target by a camera array composed of N cameras; performing image matching on the N camera images to determine H camera images corresponding to each target feature point and H feature points in the H camera images that match the target feature point;

[0008] fusing the N camera images with a pre-constructed target imaging background digital model using a virtual-real fusion method, establishing a unified world coordinate system of the imaging scene of the N camera images and the target imaging background digital model, and determining the three-dimensional coordinates of the N camera positions according to the unified world coordinate system;

[0009] dividing the target feature points to obtain edge feature points and non-edge feature points; obtaining a preliminary model of the target based on the edge feature points;

[0010] fusing the preliminary model of the target with the pre-constructed target imaging background digital model to obtain a digital model containing the target; updating the H feature points corresponding to each non-edge feature point and the H camera images using the digital model containing the target to obtain updated M feature points and M camera images;

[0011] According to the two-dimensional coordinates of the M feature points corresponding to each non-edge feature point in the corresponding M camera images, the three-dimensional coordinates of the M camera positions corresponding to the M camera images, the shooting parameters of the M cameras, and the two-dimensional coordinates of all pixel points in each of the M camera images, the three-dimensional coordinates of each non-edge feature point are obtained; the three-dimensional coordinates of all non-edge feature points are fused with the target preliminary model to obtain a target three-dimensional topography, so that the target is detected.

[0012] Based on the further improvement of the above method, the M feature points and the M camera images are updated by using the digital model containing the target for each non-edge feature point and the H camera images, including:

[0013] For each non-edge feature point, the following steps are performed:

[0014] Any one of the H feature points is taken, and any one of the remaining H-1 camera images other than the camera image corresponding to the feature point is taken to form a combination;

[0015] H-1 combinations are traversed, and for each combination, it is determined by camera imaging principle whether there is a successfully matched feature point of the feature point in another camera image, if so, the other camera image is retained, and the corresponding feature point in the camera image is updated;

[0016] Otherwise, the other camera image is removed from the H camera images corresponding to the non-edge feature point, and the corresponding feature point in the other camera image is removed from the H feature points corresponding to the non-edge feature point.

[0017] Based on the further improvement of the above method, for each combination, it is determined by camera imaging principle whether there is a successfully matched feature point of the feature point in another camera image, including:

[0018] According to the two-dimensional coordinates of the feature point in the corresponding camera, the shooting parameters of the corresponding camera, and the three-dimensional coordinates of the corresponding camera position, the three-dimensional coordinates of the feature point are determined;

[0019] A screening straight line is determined through the three-dimensional coordinates of the feature point and the three-dimensional coordinates of the corresponding camera position of the feature point, and the three-dimensional coordinates of the intersection point of the screening straight line and the digital model containing the target are calculated;

[0020] A intersecting straight line is determined through the three-dimensional coordinates of the intersection point and the three-dimensional coordinates of the camera position corresponding to the other camera image, and it is calculated whether there is an image intersection point between the intersecting straight line and the camera imaging surface region corresponding to the other camera image;

[0021] If there is no image intersection point, there is no matching successful feature point in the other camera image; otherwise, there is a matching successful feature point.

[0022] Based on the further improvement of the above method, the updating the corresponding feature point in the camera image comprises:

[0023] In the other camera image, a matching region is demarcated around the image intersection point, and a pixel point matched with the feature point in the matching region is found as the updated feature point by using a preset matching algorithm.

[0024] Based on the further improvement of the above method, the obtaining the three-dimensional coordinates of each non-edge feature point according to the two-dimensional coordinates of each non-edge feature point in the corresponding M camera images, the three-dimensional coordinates of the corresponding M camera positions, the shooting parameters of the M cameras, and the two-dimensional coordinates of all pixel points in each of the M camera images comprises:

[0025] For each of the M feature points, the following steps are performed:

[0026] Converting the two-dimensional coordinates of the feature point in the corresponding camera image into three-dimensional coordinates; converting the two-dimensional coordinates of all pixel points in each of the remaining M-1 camera images into three-dimensional coordinates; obtaining the target preselected three-dimensional coordinates corresponding to the feature point based on the three-dimensional coordinates of the feature point, the three-dimensional coordinates of the camera position corresponding to the feature point, and the three-dimensional coordinates of all pixel points in each of the remaining M-1 camera images, and the three-dimensional coordinates of the remaining M-1 camera positions.

[0027] Obtaining the three-dimensional coordinates of each non-edge feature point by averaging the M target preselected three-dimensional coordinates corresponding to the M feature points.

[0028] Based on the further improvement of the above method, the obtaining the target preselected three-dimensional coordinates corresponding to the feature point based on the three-dimensional coordinates of the feature point, the three-dimensional coordinates of the camera position corresponding to the feature point, and the three-dimensional coordinates of all pixel points in each of the remaining M-1 camera images, and the three-dimensional coordinates of the remaining M-1 camera positions comprises:

[0029] Selecting any one of the remaining M-1 camera images as a feature point camera combination together with the three-dimensional coordinates of the feature point and the three-dimensional coordinates of the camera position corresponding to the feature point.

[0030] Iterating through M-1 feature point camera combinations, and based on each feature point camera combination, the following steps are performed to determine the target combined three-dimensional coordinates corresponding to the feature point:

[0031] determining a feature point matching straight line of the three-dimensional coordinates of the feature point and the three-dimensional coordinates of the camera position corresponding to the feature point;

[0032] connecting the three-dimensional coordinates of all pixel points in the camera image and the three-dimensional coordinates of the camera position to form a plurality of camera matching straight lines;

[0033] obtaining the intersection point or the midpoint of the shortest distance between the feature point matching straight line and the plurality of camera matching straight lines as the target combined three-dimensional coordinates of the feature point;

[0034] averaging the M-1 target combined three-dimensional coordinates obtained by the M-1 feature point camera combinations to obtain the target preselected three-dimensional coordinates corresponding to the feature point.

[0035] Based on the further improvement of the above method, the detection method further comprises:

[0036] When performing image matching on the N camera images, the color information corresponding to each target feature point is obtained and stored;

[0037] Fusing the color information corresponding to each target feature point with the three-dimensional coordinates of each target feature point to obtain a target three-dimensional topography with color.

[0038] Based on the further improvement of the above method, the image matching on the N camera images, the determination of the H camera images corresponding to each target feature point and the H feature points in the H camera images matched with the target feature point comprises:

[0039] randomly selecting two camera images from the N camera images as a camera combination; traversing N(N-1) / 2 camera combinations, based on each combination, using a preset matching algorithm to match the two camera images, obtaining the feature points matched successfully by the two camera images, and taking all the feature points matched successfully by the two camera images in the N(N-1) / 2 camera combinations as target feature points;

[0040] For each target feature point, the camera image containing the feature point matched with the target feature point is taken as the camera image corresponding to the target feature point, thereby obtaining the H camera images corresponding to each target feature point and the H feature points matched with each target feature point.

[0041] Based on the further improvement of the above method, the preset matching algorithm is SIFT matching algorithm, SURF matching algorithm, template matching algorithm or SuperGlue matching algorithm.

[0042] On the other hand, an embodiment of the present application provides a target detection system based on improved image matching, the detection system comprising:

[0043] A camera array comprising N cameras is used to synchronously capture target images;

[0044] A switch connected to the camera array is used to convert the target images sent by the camera array into single-channel network signals and transmit them to a computer device;

[0045] The computer device is used to read the single-channel network signals and extract the images synchronously captured by all cameras of the camera array therefrom, and based on the images synchronously captured by all cameras, the following processing is performed to realize target detection:

[0046] Image matching is performed on the N camera images to determine the H camera images corresponding to each target feature point and the H feature points in the H camera images matched with the target feature point;

[0047] A virtual-real fusion method is used to fuse the N camera images with a pre-constructed target imaging background digital model, a unified world coordinate system of the imaging scene of the N camera images and the target imaging background digital model is established, and the three-dimensional coordinates of the N camera positions are determined according to the unified world coordinate system;

[0048] The target feature points are divided to obtain edge feature points and non-edge feature points; and a preliminary target model is obtained based on the edge feature points;

[0049] The preliminary target model is fused with the pre-constructed target imaging background digital model to obtain a digital model containing the target; and the digital model containing the target is used to update the H feature points corresponding to each non-edge feature point and the H camera images to obtain updated M feature points and M camera images;

[0050] The three-dimensional coordinates of each non-edge feature point are obtained according to the two-dimensional coordinates of the M feature points corresponding to each non-edge feature point in the corresponding M camera images, the three-dimensional coordinates of the M camera positions corresponding to the M camera images, the shooting parameters of the M cameras, and the two-dimensional coordinates of all pixel points of each image in the M camera images; the three-dimensional coordinates of all non-edge feature points are fused with the preliminary target model to obtain the three-dimensional topography of the target, so that the target is detected.

[0051] Compared with the prior art, the present application can achieve at least one of the following beneficial effects:

[0052] 1、The present application can quickly obtain the three-dimensional topography of the target by using the camera array comprising N cameras to image the target in combination with the target imaging background digital model, and then realize synchronous detection of the three-dimensional topography of the target.

[0053] 2. This invention updates each non-edge feature point using a digital model containing the target to obtain the three-dimensional coordinates of each non-edge feature point; it then fuses the three-dimensional coordinates of all non-edge feature points with the initial target model to obtain the three-dimensional shape of the target, thereby detecting the target, improving the accuracy of target detection, and achieving precise detection of the three-dimensional shape of the target to be tested.

[0054] 3. This invention obtains and stores the color information corresponding to each target feature point when performing image matching on camera images, and fuses the color information corresponding to each target feature point with the three-dimensional coordinates of each target feature point to obtain a colored three-dimensional shape of the target.

[0055] 4. This invention employs a camera array combined with a digital model to establish a unified world coordinate system between the real-time scene and the digital model. This allows for the acquisition of world coordinates, 3D topography, and the 3D topography of the target at any location within the real-time scene. The digital model achieves centimeter-level accuracy and a modeling range reaching kilometers. The use of a camera array ensures measurement accuracy, preventing the overall measurement results from being affected by the measurement deviation of a single sensor. By acquiring 3D information of the real-time scene through a unified world coordinate system, rapid 3D modeling of real-time targets is achieved, resulting in high speed and accuracy.

[0056] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0057] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0058] Figure 1 This is a flowchart illustrating the target detection method based on improved image matching provided in an embodiment of the present invention.

[0059] Figure 2 A schematic diagram illustrating the fusion of a camera image and a pre-constructed digital model of the target imaging background, provided in an embodiment of the present invention;

[0060] Figure 3 A schematic diagram of non-edge feature points of a target provided in an embodiment of the present invention;

[0061] Figure 4 This is a schematic diagram illustrating the updating of non-edge feature points according to an embodiment of the present invention;

[0062] Figure 5 FIG. 1 is a schematic diagram of calculating a target combined three-dimensional coordinate corresponding to a feature point according to a feature point camera combination provided by an embodiment of the present application;

[0063] Figure 6 FIG. 2 is a structural schematic diagram of a target detection system based on an improved image matching provided by an embodiment of the present application. DETAILED DESCRIPTION

[0064] The preferred embodiments of the present application will be described in detail below with reference to the drawings, wherein the drawings constitute a part of this application and are used to explain the principles of the embodiments of the present application, but are not used to limit the scope of the present application.

[0065] One specific embodiment of the present application discloses a target detection method based on an improved image matching, as shown in FIG. 1, the detection method comprises the following steps. Figure 1

[0066] Step S1: synchronously acquiring N camera images obtained by imaging a target by a camera array composed of N cameras; performing image matching on the N camera images to determine H camera images corresponding to each target feature point and H feature points in the H camera images matched with the target feature point;

[0067] Step S2: fusing the N camera images with a target imaging background digital model constructed in advance by using a virtual-real fusion method, establishing a unified world coordinate system of an imaging scene of the N camera images and the target imaging background digital model, and determining three-dimensional coordinates of N camera positions according to the unified world coordinate system;

[0068] Step S3: dividing the target feature points to obtain edge feature points and non-edge feature points; and obtaining a target preliminary model based on the edge feature points;

[0069] Step S4: fusing the target preliminary model with the target imaging background digital model constructed in advance to obtain a digital model containing the target; and updating H feature points corresponding to each non-edge feature point and H camera images by using the digital model containing the target to obtain updated M feature points and M camera images;

[0070] Step S5: obtaining three-dimensional coordinates of each non-edge feature point according to two-dimensional coordinates of M feature points corresponding to each non-edge feature point in corresponding M camera images, three-dimensional coordinates of M camera positions corresponding to the M camera images, shooting parameters of the M cameras, and two-dimensional coordinates of all pixel points of each image in the M camera images; fusing the three-dimensional coordinates of all non-edge feature points with the target preliminary model to obtain a target three-dimensional topography, thereby detecting the target.

[0071] ​Specifically, in step S1, the N camera images are synchronously acquired by the camera array composed of the N cameras, and each camera in the camera array can acquire a corresponding camera image. It should be noted that synchronous acquisition refers to the acquisition of imaging of the target by each camera in the camera array at the same time point.

[0072] The number N of cameras in the camera array satisfies N≥2. The larger the number N of cameras is, the more target feature points can be acquired, and the higher the granularity of the three-dimensional topography of the target is, and the clearer the detection of the target is. Meanwhile, the number N of cameras can be specifically designed according to the target imaging scene and the size of the target.

[0073] Preferably, the placement positions of the N cameras in the camera array need to satisfy that the imaging of the target by two adjacent cameras has overlapping fields of view, and the imaging of the target by the N cameras needs to completely cover the target from different angles.

[0074] Specifically, in step S1, the image matching is performed on the N camera images, and the target feature points of the target can be obtained. It can be understood that, since the N camera images are target images acquired by shooting the target from different angles, the feature points included in the N camera images can all be one of the target feature points, and different target feature points exist in multiple different camera images due to different imaging angles.

[0075] After the image matching is performed on the N camera images, the distribution of each target feature point in the N camera images is obtained, that is, the H camera images corresponding to each target feature point are determined. It should be noted that the number of the corresponding H camera images can be the same or different for different target feature points. It can be understood that, for one target feature point, it is distributed in at least two camera images and at most N camera images. Meanwhile, the feature points in the H camera images that match the target feature points are regarded as the H feature points in the H camera images that match the target feature points, and it can be understood that only one feature point in each camera image can match the target feature point.

[0076] Preferably, in step S1, the image matching is performed on the N camera images, the H camera images corresponding to each target feature point are determined, and the H feature points in the H camera images that match the target feature points, and the method comprises the following steps.

[0077] Any two camera images are taken from the N camera images as a camera combination; N(N-1) / 2 camera combinations are traversed, based on each combination, a preset matching algorithm is used to match the two camera images, feature points that match the two camera images successfully are acquired, and all the feature points that match the two camera images successfully in the N(N-1) / 2 camera combinations are regarded as target feature points.

[0078] For each target feature point, a camera image containing a feature point matched with the target feature point is taken as a camera image corresponding to the target feature point, so as to obtain H camera images corresponding to each target feature point and H feature points matched with each target feature point.

[0079] Specifically, when performing image matching on the N camera images, two camera images are randomly selected from the N camera images as a camera combination, that is, each camera combination includes two camera images. It can be understood that when performing image matching on the N camera images, N(N-1) / 2 camera combinations can be obtained.

[0080] For the N(N-1) / 2 camera combinations, a preset matching algorithm is used to match the two camera images in each camera combination in a preset order.

[0081] Preferably, the preset matching algorithm is a SIFT matching algorithm, a SURF matching algorithm, a template matching algorithm or a SuperGlue matching algorithm.

[0082] Specifically, the SIFT matching algorithm refers to a Scale Invariant Feature Transform (SIFT) matching algorithm, also known as a scale-invariant feature transform matching algorithm. The SIFT feature is invariant to rotation and scale, and has good robustness to noise, view angle change and illumination change.

[0083] Specifically, the SURF matching algorithm refers to a Speeded Up Robust Features (SURF) accelerated robust feature matching algorithm. The SURF feature is a local feature of an image. When a target image is rotated, scaled, and brightness changed, the SURF feature has invariance, and has a certain degree of stability to view angle change, affine transformation and noise. The flow of the SURF feature extraction algorithm mainly includes three parts: feature point detection, feature point description and feature point matching. The feature point detector based on Hessian matrix is used, which is superior to the detector based on Harris in stability and repeatability. The Haar wavelet is used as a feature descriptor. The biggest feature of the Harr feature is fast speed, which can reduce the calculation time and increase the robustness. The second-order Gaussian filter is replaced by a box filter, and the integral image is used to speed up the convolution, so as to reduce the complexity of time calculation and improve the calculation speed.

[0084] Specifically, for the template matching algorithm, the template is a known small image, and the template matching is to search for a target in a large image, where the target to be found is known in the large image, and the target has the same size, direction and image elements as the template. When the target is found in the image through a certain algorithm, its coordinate position is determined.

[0085] Specifically, the SuperGlue matching algorithm is a feature matching algorithm based on a graph convolutional neural network, which is used to complete feature point matching.

[0086] It can be understood that, according to the two camera images included in each combination, the feature point matching is performed between all feature points in one of the camera images and the feature points in the other camera image, so that the matching feature points in the two camera images can be determined.

[0087] Specifically, all the feature points that are successfully matched in the two camera images in the N(N-1) / 2 camera combinations are taken as target feature points, that is, the target target feature points are distributed in at least two camera images.

[0088] Specifically, for each target feature point, the camera image containing the feature point matched with the target feature point is taken as the camera image corresponding to the target feature point, so that H camera images corresponding to each target feature point and H feature points matched with each target feature point are obtained.

[0089] Specifically, in step S2, as shown in Figure 2 N camera images are fused with a pre-constructed target imaging background digital model by using a virtual-real fusion method, an imaging scene of the N camera images and the target imaging background digital model are established in a unified world coordinate system, and three-dimensional coordinates of the N camera positions are determined according to the unified world coordinate system.

[0090] Specifically, the pre-constructed target imaging background digital model is constructed according to the target imaging background. In the target imaging background digital model, the target imaging background can be displayed. In implementation, pictures of the target imaging background can be obtained by a drone, and then the target imaging background can be constructed by using modeling software of the DJI drone to obtain the target imaging background digital model. Alternatively, other methods capable of constructing a three-dimensional digital model in the prior art are also feasible. The construction method of the digital model is not protected in the present application, and the prior art can be used to obtain it.

[0091] In Figure 2In this method, after fusing N camera images with a pre-constructed digital model of the target imaging background using a virtual-real fusion method, a unified world coordinate system is established between the imaging scene of the N camera images and the digital model of the target imaging background. The center of the digital model of the target imaging background can be used as the origin of the unified world coordinate system, thereby obtaining the three-dimensional coordinates of the N camera positions.

[0092] exist Figure 2 The camera in the middle can represent any one camera in the camera array. For each camera in the camera array, after the virtual-real fusion method is used to fuse the images of N cameras with the pre-constructed digital model of the target imaging background, the virtual camera image in the digital scene is exactly the same as the camera image taken by the real camera. At this time, the position of the virtual camera in the three-dimensional digital model is the three-dimensional position of the physical camera, thereby realizing the position positioning of N cameras.

[0093] Specifically, in step S3, the target feature points are divided into edge feature points and non-edge feature points, such as... Figure 3 As shown, target feature points P1, P2, P3, P4, P5, P7, and P8 can be considered as edge feature points, while the remaining target feature points are considered as non-edge feature points. Furthermore, points with sharp angles can also be considered as edge feature points. Based on these edge feature points, the approximate outline of the target can be described, serving as a preliminary target model.

[0094] Specifically, obtaining the preliminary target model based on edge feature points includes:

[0095] Determine the H camera images corresponding to each edge feature point, and the H feature points in the H camera images that match the target feature point;

[0096] The three-dimensional coordinates of each edge feature point are obtained by considering the two-dimensional coordinates of the H feature points corresponding to each edge feature point in the corresponding H camera images, the three-dimensional coordinates of the H camera positions corresponding to the H camera images, the shooting parameters of the H cameras, and the two-dimensional coordinates of all pixels in each of the H camera images.

[0097] The specific implementation method of the above process is the same as the method of calculating the three-dimensional coordinates of non-edge feature points in step S5, and will not be repeated here.

[0098] Specifically, in step S4, the preliminary target model is fused with the pre-constructed target imaging background digital model to obtain a digital model containing the target. The digital model containing the target is used to update the H feature points and H camera images corresponding to each non-edge feature point. After the update, each non-edge feature point corresponds to M feature points and M camera images.

[0099] Preferably, in order to further improve the accuracy of image matching, reduce the matching errors caused by high similarity of image feature points, the H feature points corresponding to each non-edge feature point and the H camera images need to be updated using the digital model containing the target in step S4.

[0100] Specifically, the updating of the H feature points corresponding to each non-edge feature point and the H camera images using the digital model containing the target to obtain the updated M feature points and M camera images comprises:

[0101] For each non-edge feature point, the following steps are performed:

[0102] Any one of the H feature points is taken, and any one of the remaining H-1 camera images other than the camera image corresponding to the feature point is taken to form a combination;

[0103] H-1 combinations are traversed, and for each combination, it is determined whether there is a successfully matched feature point in another camera image of the feature point through camera imaging principle, if there is, another camera image is retained, and the corresponding feature point in the camera image is updated;

[0104] Otherwise, another camera image is removed from the H camera images corresponding to the non-edge feature point, and the corresponding feature point in another camera image is removed from the H feature points corresponding to the non-edge feature point.

[0105] For example, when updating any one target feature point R in the non-edge feature point, the H cameras matched with the target feature point R are H1, H2…Hi…Hj…HH, and the H feature points matched are R1, R2…Ri…Rj…RH.

[0106] Specifically, as shown in Figure 4 For the non-edge feature point R, the matching point in the camera Hi is Ri, and there is a successfully matched feature point in the camera Hj, denoted as Rj; it can be understood that when image matching is performed in step S1, if two point features are similar but actually two different target feature points, it is possible to mistakenly match two similar feature points corresponding to different target feature points as corresponding to the same target feature point when matching the entire image. In order to further verify whether Ri and Rj are truly corresponding to the same target feature point, the H feature points corresponding to the non-edge feature point and the H camera images are updated, so as to obtain the camera images and feature points that are truly corresponding to the non-edge feature point, and remove the camera images and feature points that are not corresponding to the non-edge feature point.

[0107] Specifically, one feature point Ri is selected from the H feature points R1, R2...Ri...Rj...RH, and a combination is formed by the feature point Ri and any one of the remaining H-1 camera images other than the camera image corresponding to the feature point Ri. Whether there is a matching successful feature point of the feature point Ri in another camera Hj image is determined by camera imaging principle. If there is, another camera Hj image is retained, and the corresponding feature point Rj in the camera image is updated; otherwise, another camera Hj image is removed from the H camera images corresponding to the non-edge feature point, and the corresponding feature point Rj in another camera image is removed from the H feature points corresponding to the non-edge feature point.

[0108] Preferably, for each combination, whether there is a matching successful feature point of the feature point in another camera image is determined by camera imaging principle, including:

[0109] The three-dimensional coordinates of the feature point are determined according to the two-dimensional coordinates of the feature point in the corresponding camera, the shooting parameters of the corresponding camera, and the three-dimensional coordinates of the corresponding camera position.

[0110] A screening straight line passing through the three-dimensional coordinates of the feature point and the three-dimensional coordinates of the corresponding camera position of the feature point is determined, and the three-dimensional coordinates of the intersection point of the screening straight line and the digital model containing the target are calculated.

[0111] An intersecting straight line passing through the three-dimensional coordinates of the intersection point and the three-dimensional coordinates of the corresponding camera position of another camera image is determined, and whether there is an image intersection point between the intersecting straight line and the camera imaging surface region corresponding to another camera image is calculated.

[0112] If there is no image intersection point, there is no matching successful feature point of the feature point in another camera image; otherwise, there is a matching successful feature point.

[0113] Specifically, for the combination of the feature point Ri and the camera Hj, the two-dimensional coordinates of the feature point Ri in the Hi camera image are determined, the two-dimensional coordinates of the feature point Ri in the Hi camera image are converted into three-dimensional coordinates of the feature point Ri in the unified world coordinate system by combining the shooting parameters of the camera Hi and the three-dimensional coordinates of the Hi camera position.

[0114] As Figure 4As shown, a screening straight line Si passing through the three-dimensional coordinates of the feature point Ri and the three-dimensional coordinates of the camera position of Hi is determined, the three-dimensional coordinates of the intersection point of the screening straight line Si and the digitized model containing the target are calculated, and are recorded as Rcam1; then an intersecting straight line Ij passing through the three-dimensional coordinates Ram1 of the intersection point and the three-dimensional coordinates of the camera position of Hj is determined, and whether there is an image intersection point in the camera imaging surface region corresponding to the camera image of Hj is calculated. If there is no image intersection point, the feature point Ri does not exist in the other camera Hj image as a matching successful feature point; otherwise, there is a matching successful feature point.

[0115] Preferably, the corresponding feature point in the camera image is updated, including:

[0116] In another camera image, a matching region is demarcated around the image intersection point, and a preset matching algorithm is used to find a pixel point matching the feature point in the matching region as an updated feature point.

[0117] Specifically, as Figure 4 shown, when the intersecting straight line Ij has an image intersection point Rappr' in the camera imaging surface region corresponding to the camera image of Hj, it is indicated that there is a matching successful feature point in the camera image of Hj corresponding to the feature point Ri. A matching region is set around the image intersection point Rappr', and a preset matching algorithm is used to find a pixel point matching the feature point Ri in the matching region as an updated feature point Rj.

[0118] Specifically, the matching region is centered on Rappr', and a suitable neighborhood is selected as the matching region. The matching region can be a circular region or a rectangular region, which is not limited here. The preset matching algorithm is a SIFT matching algorithm, a SURF matching algorithm, a template matching algorithm, or a SuperGlue matching algorithm.

[0119] Because the neighborhood range is small, the feature point that can truly match the feature point Ri must be within the neighborhood range, reducing the influence of other feature points that can match the feature point Ri, improving the matching accuracy, and making the updated Rj more accurate.

[0120] After the camera image and the feature point corresponding to each non-edge feature point are updated through step S4, M camera images and M feature points corresponding to each non-edge feature point are obtained.

[0121] Specifically, in step S5, the three-dimensional coordinates of each non-edge feature point are obtained according to the two-dimensional coordinates of the M feature points corresponding to each non-edge feature point in the corresponding M camera images, the three-dimensional coordinates of the M camera positions corresponding to the M camera images, the shooting parameters of the M cameras, and the two-dimensional coordinates of all pixel points in each of the M camera images; the three-dimensional coordinates of all non-edge feature points are fused with the target preliminary model to obtain a target three-dimensional topography, thereby detecting the target.

[0122] It can be understood that after step S4 ends, the positions of the N cameras are determined in the unified world coordinate system. The distribution of any one of the non-edge feature points in the M camera images is also determined, and further, the two-dimensional coordinates of the M feature points in the M camera images can be obtained. The three-dimensional coordinates of each non-edge feature point are calculated by the two-dimensional coordinates of the M feature points in the corresponding M camera images, the three-dimensional coordinates of the M camera positions corresponding to the M camera images, and the shooting parameters of the M cameras.

[0123] Specifically, the shooting parameters of the camera include the focal length, magnification, imaging surface size, and the like.

[0124] After the three-dimensional coordinates of each non-edge feature point are obtained, the three-dimensional coordinates of all non-edge feature points are fused with the target preliminary model to obtain a target three-dimensional topography, thereby detecting the target.

[0125] Preferably, in step S5, the three-dimensional coordinates of each non-edge feature point are obtained according to the two-dimensional coordinates of the M feature points corresponding to each non-edge feature point in the corresponding M camera images, the three-dimensional coordinates of the M camera positions corresponding to the M camera images, the shooting parameters of the M cameras, and the two-dimensional coordinates of all pixel points in each of the M camera images, including:

[0126] For each of the M feature points, the following steps are performed:

[0127] The two-dimensional coordinates of the feature point in the corresponding camera image are converted into three-dimensional coordinates; for the remaining M-1 camera images other than the camera image corresponding to the feature point, the two-dimensional coordinates of all pixel points in each of the M-1 camera images are converted into three-dimensional coordinates; based on the three-dimensional coordinates of the feature point, the three-dimensional coordinates of the camera position corresponding to the feature point, and the three-dimensional coordinates of all pixel points in each of the remaining M-1 camera images, the three-dimensional coordinates of the remaining M-1 camera positions, the target preselected three-dimensional coordinates corresponding to the feature point are obtained;

[0128] The M target preselected three-dimensional coordinates corresponding to the M feature points are averaged to obtain the three-dimensional coordinates of each non-edge feature point.

[0129] It can be understood that the three-dimensional coordinates of the camera position can be the position of the camera center point or the position of the camera lens center. When the shooting parameters of the camera and the three-dimensional coordinates of the camera are determined, the positional relationship of any feature point on the camera image relative to the camera is fixed. According to the shooting parameters, the three-dimensional coordinates of any feature point in the camera image can be obtained through corresponding coordinate conversion.

[0130] Specifically, as shown in Figure 2 for any one camera in the N cameras, the three-dimensional coordinates of the position are pos1, and the two-dimensional coordinates of any feature point P1 in the camera image are fixed. In combination with the shooting parameters of the camera, the two-dimensional coordinates of the feature point P1 are converted to obtain the three-dimensional coordinates of the feature point P1 in the unified world coordinate system.

[0131] Specifically, as shown in Figure 5 in determining the three-dimensional coordinates of any target feature point Q in the non-edge feature points, the M cameras matched with the target feature point Q are M1, M2…Mi…Mj…MM, and the M feature points matched are Q1, Q2…Qi…Qj…QM. For each of the M feature points Q1, Q2…Qi…Qj…QM, the same steps are performed, and Q1, Q2…Qi…Qj…or QM is the feature point. As shown in Figure 3 for the feature point Qi in the camera Mi, and the feature point is Qj in the camera Mj.

[0132] Preferably, the conversion of the two-dimensional coordinates of the feature point in the corresponding camera image into three-dimensional coordinates comprises:

[0133] According to the three-dimensional coordinates of the camera position corresponding to the feature point and the shooting parameters of the camera corresponding to the feature point, the two-dimensional coordinates of the feature point in the corresponding camera image are converted into three-dimensional coordinates;

[0134] For the remaining M-1 camera images other than the camera image corresponding to the feature point, the two-dimensional coordinates of all pixel points in each of the M-1 camera images are converted into three-dimensional coordinates, comprising:

[0135] According to the three-dimensional coordinates of each camera position in the M-1 cameras and the shooting parameters of each camera, the two-dimensional coordinates of all pixel points in each of the M-1 camera images are converted into three-dimensional coordinates.

[0136] It can be understood that according to the three-dimensional coordinates of the camera position of each camera and the shooting parameters of the camera, the two-dimensional coordinates in the corresponding M camera images are converted to obtain the three-dimensional coordinates of the corresponding feature points Q1, Q2…Qi…Qj…QM in each camera, i.e. the three-dimensional coordinates of the feature point in the unified world coordinate system.

[0137] It is understandable that, for the M-1 cameras other than camera Mi corresponding to the feature point Qi, the two-dimensional coordinates of all pixels in each of the M-1 camera images are converted into three-dimensional coordinates in a unified world coordinate system. When the camera's shooting parameters and the three-dimensional coordinates of the camera position are determined, the positional relationship of any pixel in the camera image relative to the camera is fixed. By changing the coordinates according to the shooting parameters, the three-dimensional coordinates of any pixel in the camera image can be obtained.

[0138] Preferably, obtaining the target pre-selected three-dimensional coordinates corresponding to the feature point based on the three-dimensional coordinates of the feature point, the three-dimensional coordinates of the camera position corresponding to the feature point, and the three-dimensional coordinates of all pixels in each of the remaining M-1 camera images, and the three-dimensional coordinates of the remaining M-1 camera positions, includes:

[0139] Take any one camera image from the remaining M-1 camera images, and combine it with the 3D coordinates of the feature point and the 3D coordinates of the camera position corresponding to the feature point as a feature point camera combination;

[0140] Iterate through M-1 possible camera combinations with feature points. For each feature point camera combination, perform the following steps to determine the 3D coordinates of the target combination corresponding to that feature point:

[0141] A feature point matching line is determined by passing through the 3D coordinates of the feature point and the 3D coordinates of the camera position corresponding to the feature point;

[0142] For another camera image in the combination, connect the three-dimensional coordinates of all pixels in the camera image with the three-dimensional coordinates of the camera position to form multiple camera matching lines;

[0143] Obtain the intersection point or the midpoint of the shortest distance between the feature point matching line and multiple camera matching lines, and use it as the target combination three-dimensional coordinates of the feature point;

[0144] The average value of the three-dimensional coordinates of the M-1 target combinations obtained by the camera combination of M-1 feature points is used to obtain the pre-selected three-dimensional coordinates of the target corresponding to the feature point.

[0145] Specifically, such as Figure 5 As shown, the feature point is Qi, located in camera Mi; from the remaining M-1 camera images, the camera image taken by camera Mj is selected, and together with the three-dimensional coordinates of the feature point Qi in the unified world coordinate system and the three-dimensional coordinates of the camera position of camera Mi, a feature point camera combination is formed.

[0146] According to the three-dimensional coordinates of the feature point Qi and the three-dimensional coordinates of the camera position of the camera Mi, a feature point matching straight line Li is determined; according to the three-dimensional coordinates of the camera position of the camera Mj and the three-dimensional coordinates of all pixel points in the camera image of the camera Mj, a plurality of camera matching straight lines Mj1, Mj2, Mj3, …, Mj… are determined. The intersection point or the midpoint of the shortest distance of the feature point matching straight line Li and the plurality of camera matching straight lines Mj1, Mj2, Mj3, or Mj… is calculated, and the intersection point or the midpoint of the shortest distance is taken as the target combined three-dimensional coordinates of the feature point Qi.

[0147] It can be understood that, because of measurement errors, in actual engineering applications, the two rays rarely intersect, and are generally staggered. The midpoint of the shortest distance of the two rays can be taken as the preselected three-dimensional coordinates of the target feature point Q, which can minimize the influence of measurement errors and improve the accuracy of the three-dimensional coordinates of the target feature point.

[0148] Each of the M-1 feature point-camera combinations is traversed, and the target combined three-dimensional coordinates of the feature point Qi are calculated based on each feature point-camera combination, to obtain M-1 target combined three-dimensional coordinates corresponding to the feature point Qi. The average of the M-1 target combined three-dimensional coordinates corresponding to the feature point Qi is calculated as the target preselected three-dimensional coordinates corresponding to the feature point Qi.

[0149] For the target feature point Q, the matching M feature points are Q1, Q2, …, Qi, …, Qj, …, QM, and the M target preselected three-dimensional coordinates corresponding to the feature points Q1, Q2, …, Qi, …, Qj, …, QM are determined. The average of the M target preselected three-dimensional coordinates corresponding to the feature points Q1, Q2, …, Qi, …, Qj, …, QM is taken as the three-dimensional coordinates of the target feature point Q.

[0150] The above method traverses the M feature points corresponding to the target feature point, each feature point is solved to obtain a target preselected three-dimensional coordinate, and then the average of the M target preselected three-dimensional coordinates is calculated to obtain the three-dimensional coordinates of the final target feature point, so that the calculated three-dimensional coordinates of the target feature point are more accurate.

[0151] In order to make the three-dimensional appearance of the target more vivid and easier to identify, color information can be added to the three-dimensional appearance of the target. Preferably, in step S1, when the image matching is performed on the N camera images, the color information corresponding to each target feature point is obtained and stored.

[0152] It can be understood that, for any one same target feature point of the target, the corresponding color information is the same, that is, the color information of the M feature points corresponding to the same target feature point in different cameras is also the same. The color information of the target feature point can be obtained in any one of the M camera images, and the obtained color information of the target feature point is stored.

[0153] Preferably, in step S5, the color information corresponding to each target feature point is fused with the three-dimensional coordinates of each target feature point to obtain a target three-dimensional topography with color.

[0154] Specifically, the color information corresponding to each target feature point can be fused with the three-dimensional coordinates of each target feature point to obtain a target three-dimensional topography with color, which can be displayed by a visualization display component.

[0155] In implementation, the size of the camera array N can be reasonably set so that the target is imaged on each camera as much as possible, which can increase the number of target feature points and further improve the accuracy of the target three-dimensional topography.

[0156] Compared with the prior art, the target detection method based on improved image matching provided by the embodiment of the application can quickly obtain a target three-dimensional topography by combining a target imaging background digital model, and then realize synchronous detection of the target three-dimensional topography; at the same time, the three-dimensional coordinates of each target feature point of the target in a unified world coordinate system can be determined according to the three-dimensional coordinates of different cameras in the unified world coordinate system and the two-dimensional coordinates of the plurality of feature points in different cameras, and the positioning of the target feature points is realized; and the color information corresponding to each target feature point can be obtained and stored when the camera images are image matched, and the color information corresponding to each target feature point is fused with the three-dimensional coordinates of each target feature point to obtain a target three-dimensional topography with color; and the three-dimensional coordinates of each non-edge feature point can be obtained by updating each non-edge feature point using a digital model containing the target; the three-dimensional coordinates of all non-edge feature points are fused with the target preliminary model to obtain a target three-dimensional topography, so that the target is detected, the accuracy of target detection is improved, and accurate detection of the three-dimensional topography of the target to be detected is realized.

[0157] One specific embodiment of the application discloses a target detection system based on improved image matching, as shown in Figure 6 The detection system comprises:

[0158] A camera array comprising N cameras is used to synchronously collect target images.

[0159] A switch connected with the camera array is used to convert the target images sent by the camera array into single-network signals and transmit them to a computer device.

[0160] The computer device is used for reading a single-channel network signal and taking out images synchronously collected by all cameras of a camera array from the single-channel network signal, and performing the following processing based on the images synchronously collected by all cameras to realize target detection.

[0161] Image matching is performed on the N camera images to determine H camera images corresponding to each target feature point and H feature points in the H camera images matched with the target feature point.

[0162] A virtual-real fusion method is used to fuse the N camera images with a pre-constructed target imaging background digital model, to establish a unified world coordinate system of an imaging scene of the N camera images and the target imaging background digital model, and to determine three-dimensional coordinates of N camera positions according to the unified world coordinate system.

[0163] The target feature points are divided to obtain edge feature points and non-edge feature points; and a target preliminary model is obtained based on the edge feature points.

[0164] The target preliminary model is fused with the pre-constructed target imaging background digital model to obtain a digital model containing the target; and the digital model containing the target is used to update H feature points corresponding to each non-edge feature point and H camera images to obtain updated M feature points and M camera images.

[0165] Three-dimensional coordinates of each non-edge feature point are obtained according to two-dimensional coordinates of the M feature points corresponding to each non-edge feature point in the corresponding M camera images, three-dimensional coordinates of M camera positions corresponding to the M camera images, shooting parameters of the M cameras, and two-dimensional coordinates of all pixel points of each image in the M camera images; and the three-dimensional coordinates of all non-edge feature points are fused with the target preliminary model to obtain a target three-dimensional topography, so that the target is detected.

[0166] Those skilled in the art can understand that all or part of the processes of the above-mentioned embodiments can be completed by a computer program instructing related hardware, and the program can be stored in a computer readable storage medium, such as a magnetic disk, an optical disk, a read-only memory or a random access memory.

[0167] The above description is only a preferred embodiment of the present application, but the protection scope of the present application is not limited thereto, and any changes or replacements within the technical scope disclosed by the present application can be easily thought by those skilled in the art, which should be covered within the protection scope of the present application.

Claims

1. A target detection method based on improved image matching, characterized in that, The detection method comprises: Synchronously acquiring N camera images obtained by imaging a target by a camera array composed of N cameras; performing image matching on the N camera images to determine H camera images corresponding to each target feature point and H feature points in the H camera images matched with the target feature point; Fusing the N camera images with a pre-constructed target imaging background digital model by a virtual-real fusion method to establish a unified world coordinate system of an imaging scene of the N camera images and the target imaging background digital model, and determining three-dimensional coordinates of N camera positions according to the unified world coordinate system; Dividing the target feature points to obtain edge feature points and non-edge feature points; obtaining a target preliminary model based on the edge feature points; Fusing the target preliminary model with the pre-constructed target imaging background digital model to obtain a digital model containing the target; updating H feature points corresponding to each non-edge feature point and H camera images by using the digital model containing the target to obtain updated M feature points and M camera images; Obtaining three-dimensional coordinates of each non-edge feature point according to two-dimensional coordinates of the M feature points corresponding to each non-edge feature point in the corresponding M camera images, three-dimensional coordinates of M camera positions corresponding to the M camera images, shooting parameters of the M cameras, and two-dimensional coordinates of all pixel points of each image in the M camera images; fusing the three-dimensional coordinates of all non-edge feature points with the target preliminary model to obtain a target three-dimensional topography, thereby detecting the target.

2. The detection method according to claim 1, characterized in that, The updating of the H feature points corresponding to each non-edge feature point and the H camera images by using the digital model containing the target to obtain the updated M feature points and the M camera images comprises: For each non-edge feature point, the following steps are performed: Randomly selecting one feature point from the H feature points and any one of the remaining H-1 camera images other than the camera image corresponding to the feature point to form a combination; Iterating through the H-1 combinations, for each combination, determining whether there is a successfully matched feature point of the feature point in another camera image by camera imaging principle, if there is, retaining the other camera image and updating the corresponding feature point in the camera image; Otherwise, removing the other camera image from the H camera images corresponding to the non-edge feature point and removing the corresponding feature point in the other camera image from the H feature points corresponding to the non-edge feature point.

3. The detection method according to claim 2, characterized in that, The determination of whether there is a successfully matched feature point of the feature point in another camera image by camera imaging principle for each combination comprises: Determining the three-dimensional coordinates of the feature point according to the two-dimensional coordinates of the feature point in the corresponding camera, the shooting parameters of the corresponding camera, and the three-dimensional coordinates of the corresponding camera position; Determining a screening straight line of the three-dimensional coordinates of the feature point and the three-dimensional coordinates of the corresponding camera position of the feature point, and calculating the three-dimensional coordinates of the intersection of the screening straight line and the digital model containing the target. determining a straight line of intersection of the three-dimensional coordinate of the intersection point and the three-dimensional coordinate of the camera position corresponding to the other camera image, and calculating whether there is an image intersection point between the straight line of intersection and the imaging surface area of the camera corresponding to the other camera image; if there is no image intersection point, there is no matching successful feature point in the other camera image; otherwise, there is a matching successful feature point.

4. The detection method according to claim 3, characterized in that, the updating of the corresponding feature point in the camera image comprises: in the other camera image, a matching region is demarcated with the image intersection point as the center, and a preset matching algorithm is used to find a pixel point matched with the feature point in the matching region as the updated feature point.

5. The detection method according to any one of claims 1 to 4, characterized in that, the three-dimensional coordinate of each non-edge feature point is obtained according to the two-dimensional coordinates of the M feature points corresponding to each non-edge feature point in the M camera images, the three-dimensional coordinates of the M camera positions corresponding to the M camera images, the shooting parameters of the M cameras, and the two-dimensional coordinates of all pixel points in each of the M camera images, and the three-dimensional coordinate of each non-edge feature point is obtained, comprising: for each of the M feature points, the following steps are performed: the two-dimensional coordinates of the feature point in the corresponding camera image are converted into three-dimensional coordinates; for the remaining M-1 camera images other than the camera image corresponding to the feature point, the two-dimensional coordinates of all pixel points in each of the M-1 camera images are converted into three-dimensional coordinates; the target preselected three-dimensional coordinate corresponding to the feature point is obtained based on the three-dimensional coordinates of the feature point, the three-dimensional coordinates of the camera position corresponding to the feature point, and the three-dimensional coordinates of all pixel points in each of the remaining M-1 camera images and the three-dimensional coordinates of the remaining M-1 camera positions. the three-dimensional coordinates of each non-edge feature point are obtained by averaging the M target preselected three-dimensional coordinates corresponding to the M feature points.

6. The detection method according to claim 5, characterized in that, the target preselected three-dimensional coordinate corresponding to the feature point is obtained based on the three-dimensional coordinates of the feature point, the three-dimensional coordinates of the camera position corresponding to the feature point, and the three-dimensional coordinates of all pixel points in each of the remaining M-1 camera images and the three-dimensional coordinates of the remaining M-1 camera positions, comprising: any one of the remaining M-1 camera images is taken as a feature point camera combination together with the three-dimensional coordinates of the feature point and the three-dimensional coordinates of the camera position corresponding to the feature point; M-1 feature point camera combinations are traversed, and for each feature point camera combination, the following steps are performed to determine the target combined three-dimensional coordinate corresponding to the feature point; a feature point matching straight line is determined through the three-dimensional coordinates of the feature point and the three-dimensional coordinates of the camera position corresponding to the feature point; for another camera image in the combination, the three-dimensional coordinates of all pixel points in the camera image are connected with the three-dimensional coordinates of the camera position to form a plurality of camera matching straight lines; the intersection point or the midpoint of the shortest distance between the feature point matching straight line and the plurality of camera matching straight lines is obtained as the target combined three-dimensional coordinate of the feature point; the M-1 target combined three-dimensional coordinates obtained from the M-1 feature point camera combinations are averaged to obtain the target preselected three-dimensional coordinate corresponding to the feature point.

7. The method of claim 1, wherein, the detection method further comprises: when image matching is performed on the N camera images, the color information corresponding to each target feature point is obtained and stored; The color information corresponding to each target feature point is fused with the three-dimensional coordinates of each target feature point to obtain a target three-dimensional appearance with color.

8. The method of claim 1, wherein, The image matching on the N camera images is performed to determine H camera images corresponding to each target feature point and H feature points in the H camera images matched with the target feature point, and the image matching includes: Two camera images are randomly selected from the N camera images as a camera combination; N(N-1) / 2 camera combinations are traversed, and based on each combination, a preset matching algorithm is used to match the two camera images to obtain feature points matched successfully by the two camera images; and all feature points matched successfully by the two camera images in the N(N-1) / 2 camera combinations are taken as target feature points. For each target feature point, a camera image containing a feature point matched with the target feature point is taken as a camera image corresponding to the target feature point, so as to obtain H camera images corresponding to each target feature point and H feature points matched with each target feature point.

9. The detection method according to claim 3 or 8, characterized in that, The preset matching algorithm is a SIFT matching algorithm, a SURF matching algorithm, a template matching algorithm or a SuperGlue matching algorithm.

10. An object detection system based on improved image matching, characterized by, The detection system includes: a camera array including N cameras, configured to synchronously collect target images; a switch connected with the camera array, configured to convert the target images sent by the camera array into single-path network signals and transmit the single-path network signals to a computer device; the computer device is configured to read the single-path network signals and take out images synchronously collected by all cameras of the camera array from the single-path network signals, and perform the following processing based on the images synchronously collected by all cameras to realize target detection: image matching on the N camera images is performed to determine H camera images corresponding to each target feature point and H feature points in the H camera images matched with the target feature point; a virtual-real fusion method is used to fuse the N camera images with a pre-constructed target imaging background digital model to establish a unified world coordinate system of an imaging scene of the N camera images and the target imaging background digital model, and three-dimensional coordinates of positions of the N cameras are determined according to the unified world coordinate system; target feature points are divided to obtain edge feature points and non-edge feature points; a target preliminary model is obtained based on the edge feature points; the target preliminary model is fused with the pre-constructed target imaging background digital model to obtain a digital model containing the target; the digital model containing the target is used to update H feature points corresponding to each non-edge feature point and H camera images to obtain updated M feature points and M camera images; three-dimensional coordinates of each non-edge feature point are obtained according to two-dimensional coordinates of M feature points corresponding to each non-edge feature point in M camera images, three-dimensional coordinates of M camera positions corresponding to the M camera images, shooting parameters of the M cameras and two-dimensional coordinates of all pixel points of each image in the M camera images; three-dimensional coordinates of all non-edge feature points are fused with the target preliminary model to obtain a target three-dimensional appearance, so that the target is detected.

Images