Spatial positioning method, device and equipment

Abstract

The invention discloses a spatial positioning method, device and equipment, and relates to the technical field of computers, and the spatial positioning method comprises the steps: carrying out the feature point extraction of a target object region of a current image frame, and obtaining a plurality of depth feature points of the target object region; performing feature matching on each depth feature point and a plurality of map points existing in at least one frame of map point image to obtain a feature point matching pair corresponding to the map point image; and when the number of the matching pairs of the feature point matching pairs corresponding to the map point image is greater than a preset number, performing pose calculation according to the feature point matching pairs corresponding to the map point image, and determining a flight body pose corresponding to the current image frame. Through the above mode, the plurality of depth feature points of the target object area are matched with the map points in the map point image constructed in advance, and when the matching result meets the condition, the flight body posture of the current image frame is determined through feature point matching, so that accurate positioning of the flight body in a complex environment is realized.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a spatial positioning method, device and equipment. Background Art

[0002] At present, due to the limitations of feature point detection and description, the inaccuracy of feature point matching and other factors, the application effect of traditional methods in complex environments is restricted, resulting in low accuracy of tasks requiring high precision and robustness. Summary of the invention

[0003] The main purpose of this application is to provide a spatial positioning method, device and equipment, aiming to solve the technical problem that the existing technology cannot accurately position the flying object.

[0004] To achieve the above purpose, the present application proposes a spatial positioning method, which includes:

[0005] Extracting feature points from a target object region of a current image frame to obtain a plurality of depth feature points of the target object region;

[0006] Perform feature matching on each depth feature point with a plurality of map points existing in at least one frame of map point image to obtain a feature point matching pair corresponding to the map point image;

[0007] When the number of matching pairs of feature point matching pairs corresponding to the map point image is greater than a preset number, posture calculation is performed according to the feature point matching pairs corresponding to the map point image to determine the posture of the flying object corresponding to the current image frame.

[0008] In one embodiment, the step of performing feature matching on each depth feature point with a plurality of map points existing in at least one frame of map point image to obtain a feature point matching pair corresponding to the map point image comprises:

[0009] Perform feature matching on each depth feature point and a plurality of map points existing in at least one frame of map point image to obtain an initial matching pair corresponding to the map point image;

[0010] Performing matching verification on the initial matching pairs corresponding to the map point images according to the target homography transformation model to obtain a matching verification result;

[0011] Initial matching pairs corresponding to the map point image are filtered according to the matching verification result to obtain feature point matching pairs corresponding to the map point image.

[0012] In one embodiment, the step of calculating the posture of the flying object corresponding to the current image frame according to the matching pairs of feature points corresponding to the map point image comprises:

[0013] Extracting a plurality of map points and a plurality of depth feature points from the feature point matching pairs corresponding to the map point image;

[0014] Calculate the target rotation matrix and target translation vector based on the three-dimensional coordinates of each map point and the pixel coordinates of each depth feature point;

[0015] The position and posture of the flying object corresponding to the current image frame is determined according to the target rotation matrix and the target translation vector.

[0016] In one embodiment, before the step of performing feature matching between each depth feature point and a plurality of map points existing in at least one frame of map point image to obtain a feature point matching pair corresponding to the map point image, the step further includes:

[0017] Performing image segmentation on the flying object image frames corresponding to different acquisition conditions to determine the flying object region of each flying object image frame;

[0018] According to the target deep learning model, feature points of the flying body region of each flying body image frame are extracted to obtain multiple depth feature point data of each flying body region;

[0019] Map points are constructed according to a plurality of depth feature point data of each flying body region and a reference image frame to obtain a plurality of map point images and a plurality of map points existing in each map point image.

[0020] In one embodiment, the step of constructing map points based on the multiple depth feature point data of each flying body region and the reference image frame to obtain multiple map point images and multiple map points existing in each map point image includes:

[0021] Perform binocular stereo matching on the flying body area of ​​each flying body image frame, and generate multiple monocular map points of multiple monocular image frames according to the matching results;

[0022] Retaining multiple depth feature point data of multiple flying body regions to obtain multiple monocular feature points of multiple monocular image frames;

[0023] Map points are constructed according to a plurality of monocular map points of each monocular image frame, a plurality of monocular feature points of each monocular image frame and a reference image frame to obtain a plurality of map point images and a plurality of map points existing in each map point image.

[0024] In one embodiment, the step of performing binocular stereo matching on the flying body region of each flying body image frame and generating multiple monocular map points of multiple monocular image frames according to the matching results includes:

[0025] Perform same-frame binocular matching on multiple flying object image frames to determine multiple same-frame binocular image pairs;

[0026] Perform stereo matching on the flying object regions of multiple binocular image pairs in the same frame, and determine the initial matching pairs of each binocular image pair in the same frame;

[0027] Perform mismatch filtering on the initial matching pairs of each binocular image pair in the same frame, and determine the target matching pairs of each binocular image pair in the same frame;

[0028] A triangulation calculation is performed based on the target matching pairs of each binocular image pair in the same frame to determine multiple monocular map points of multiple monocular image frames.

[0029] In one embodiment, the step of constructing map points according to the multiple monocular map points of each monocular image frame, the multiple monocular feature points of each monocular image frame, and the reference image frame to obtain the multiple map point images and the multiple map points existing in each map point image includes:

[0030] Performing deep feature matching on multiple monocular feature points of each monocular image frame to determine feature point matching pairs between each monocular image frame;

[0031] Determine the relative position and posture of the flying object corresponding to each monocular image frame according to the feature point matching pairs between each monocular image frame and the reference image frame;

[0032] The relative posture of the flying object corresponding to each monocular image frame and multiple monocular map points of each monocular image frame are globally optimized, and multiple map point images and multiple map points existing in each map point image are determined according to the optimization results.

[0033] In one embodiment, the step of extracting feature points from a target object region of the current image frame to obtain a plurality of depth feature points of the target object region includes:

[0034] Performing image preprocessing on the current image frame to obtain a processed current image frame;

[0035] Segmenting the current image frame through a target segmentation network to determine a target object region of the current image frame;

[0036] The target object area is input into a target deep learning model to obtain a plurality of depth feature points of the target object area and a descriptor of each depth feature point.

[0037] In addition, to achieve the above-mentioned purpose, the present application also proposes a spatial positioning device, which includes:

[0038] An extraction module, used to extract feature points from a target object region of a current image frame to obtain a plurality of depth feature points of the target object region;

[0039] A matching module, used for performing feature matching between each depth feature point and a plurality of map points existing in at least one frame of map point image, to obtain a feature point matching pair corresponding to the map point image;

[0040] A calculation module is used to perform posture calculation based on the feature point matching pairs corresponding to the map point image when the number of matching pairs of feature point matching pairs corresponding to the map point image is greater than a preset number, so as to determine the posture of the flying object corresponding to the current image frame.

[0041] In addition, to achieve the above-mentioned purpose, the present application also proposes a spatial positioning device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the spatial positioning method as described above.

[0042] In addition, to achieve the above-mentioned purpose, the present application also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the spatial positioning method described above are implemented.

[0043] In addition, to achieve the above-mentioned purpose, the present application also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, the steps of the spatial positioning method described above are implemented.

[0044] The present application obtains multiple depth feature points of the target object area by extracting feature points from the target object area of ​​the current image frame; performs feature matching on each depth feature point with multiple map points existing in at least one frame of map point image to obtain feature point matching pairs corresponding to the map point image; when the number of feature point matching pairs corresponding to the map point image is greater than a preset number, performs posture calculation based on the feature point matching pairs corresponding to the map point image to determine the posture of the flying object corresponding to the current image frame. In the above manner, multiple depth feature points of the target object area in the current image frame are matched with map points in a map point image constructed in advance. When the matching result meets the conditions, the posture of the flying object corresponding to the current image frame is determined through the feature point matching pairs, thereby achieving accurate positioning of the flying object in a complex environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0046] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0047] Figure 1 A schematic diagram of a flow chart provided for the first embodiment of the spatial positioning method of the present application;

[0048] Figure 2 A schematic diagram of a flow chart provided for the second embodiment of the spatial positioning method of the present application;

[0049] Figure 3 A schematic diagram of a flow chart provided for Embodiment 3 of the spatial positioning method of the present application;

[0050] Figure 4 A schematic diagram of a brief flow of the spatial positioning method provided in Example 3 of the present application;

[0051] Figure 5 This is a schematic diagram of the module structure of the spatial positioning device according to an embodiment of the present application;

[0052] Figure 6 Schematic diagram of the device structure of the hardware operating environment involved in the spatial positioning method in the embodiment of the present application.

[0053] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0054] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present application and are not used to limit the present application.

[0055] In order to better understand the technical solution of the present application, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0056] The main solution of the embodiment of the present application is: extract feature points from the target object area of ​​the current image frame to obtain multiple depth feature points of the target object area; perform feature matching on each depth feature point with multiple map points existing in at least one frame of map point image to obtain feature point matching pairs corresponding to the map point image; when the number of matching pairs of feature point matching pairs corresponding to the map point image is greater than a preset number, perform posture calculation based on the feature point matching pairs corresponding to the map point image to determine the posture of the flying object corresponding to the current image frame.

[0057] Since traditional feature point extraction relies on manually designed rules to extract feature points in images, such as ORB, SIFT, SURF, etc.; such methods mainly determine whether the current pixel is a key point by the difference in grayscale values ​​between the current pixel and its neighboring pixels. If the pixel is determined to be a key point, its feature vector is extracted, that is, a feature point is obtained; but traditional image key point extraction has poor robustness (it may not be able to stably detect the same key point again when the image is rotated, scaled, or the lighting conditions change); because such methods only use the local grayscale gradient information of the image to determine whether the point is a key point. When the image changes, the grayscale information of the point will also change, resulting in unstable key point extraction.

[0058] At the same time, traditional key point features are also generated by artificial rules, encoding the grayscale information of the neighborhood area of ​​the key point to generate the feature vector of this key point. However, this method will also generate large changes in the feature points after the image is rotated, scaled, tilted, and the lighting changes (although ORB\SIFT\SURF and other methods have taken the aforementioned changes into consideration when generating features). When the traditional mutual matching method is used to calculate the similarity of the feature vectors of the extracted feature points, there will be a mismatch. This is mainly because the feature vectors generated by this method only consider the change information around the key point, and do not capture complex image information, resulting in insufficient feature expression capabilities.

[0059] In addition, when the descriptors generated by traditional feature points are faced with large changes in perspective or environment, they are often unable to obtain effective results for downstream use when matching through similarity calculation. Therefore, when generating map points with features, since only continuous frames of images can stably complete the matching of feature points, the generated map points have limited observations, making the matching results unstable during subsequent use.

[0060] The present application provides a solution, which uses deep learning and deep matching to extract and match the feature points of a flying object, and triangulates the feature points extracted and matched from images of different angles and distances using a binocular camera to generate map points with feature information. When in use, only the image frame of a monocular camera needs to be matched with the map points with features. That is, accurate positioning of a flying object in a complex environment is achieved through the feature point extraction using deep learning and the feature point matching method using deep learning.

[0061] It should be noted that the execution subject of this embodiment may be a computing service device with data processing, network communication and program running functions, such as a tablet computer, a personal computer, a mobile phone, etc., or a spatial positioning device capable of realizing the above functions, etc. The following takes the spatial positioning device as the execution subject as an example to illustrate this embodiment and the following embodiments.

[0062] Based on this, the present application embodiment provides a spatial positioning method, referring to Figure 1 , Figure 1 This is a schematic diagram of the flow chart of the first embodiment of the spatial positioning method of the present application.

[0063] In this embodiment, the spatial positioning method includes steps S10 to S30:

[0064] Step S10: extracting feature points from a target object region of a current image frame to obtain a plurality of depth feature points of the target object region.

[0065] It should be noted that the current image frame refers to the current monocular image acquired in real time, and the target object area refers to the flying body area in the current image frame. A feature extraction network is used to extract multiple deep feature points in the target object area. In this embodiment, the SuperPoint network is used, and other feature extraction networks can also be used. The SuperPoint network is a deep learning model for detecting local feature points in an image and generating descriptors. It can perform feature point detection and descriptor learning at the same time. It is an efficient tool in computer vision tasks, especially suitable for real-time applications and large-scale data sets.

[0066] In a feasible implementation, step S10 may include steps A11 to A13:

[0067] Step A11, performing image preprocessing on the current image frame to obtain a processed current image frame.

[0068] It should be noted that the preprocessing includes but is not limited to the following operations: resizing, color space conversion, normalization, and data enhancement, performing image preprocessing on the current image frame to obtain a processed current image frame.

[0069] Step A12, segmenting the current image frame through a target segmentation network to determine the target object area of ​​the current image frame.

[0070] It should be noted that a pre-trained target segmentation network, such as Mask R-CNN, U-Net, DeepLab, etc., is used to segment the current image frame, thereby obtaining the target object area existing in the current image frame.

[0071] Step A13: input the target object area into a target deep learning model to obtain multiple depth feature points of the target object area and a descriptor of each depth feature point.

[0072] It should be noted that in this embodiment, the target deep learning model refers to the SuperPoint network. The segmented target object area is input into the target deep learning model, which automatically detects a series of feature points in the target object area and calculates the corresponding descriptor for each feature point. In order to ensure that the feature points are evenly distributed and representative, a non-maximum suppression algorithm can be applied to the heat map to select the most significant feature points. For each detected feature point, the corresponding descriptor vector is extracted from the descriptor, which usually involves aggregating a small window around the feature point, and finally obtaining a set of deep feature points and their descriptors located in the target object area, each of which has a corresponding two-dimensional image coordinate.

[0073] Step S20 , performing feature matching on each depth feature point and a plurality of map points existing in at least one frame of map point image to obtain feature point matching pairs corresponding to the map point image.

[0074] It should be noted that a binocular camera is used to collect multiple images of the flying object at different angles and distances, and deep learning and deep matching are used to extract and match the feature points of the flying object. The binocular camera is combined with the feature points extracted and matched in the images at different angles and distances to triangulate and generate map points with feature information. In this embodiment, the map point image refers to a monocular image containing multiple map points and the target three-dimensional coordinates corresponding to the map points, and the coordinate system of the target three-dimensional coordinates is the coordinate system of the randomly selected reference image frame.

[0075] It can be understood that the extracted multiple deep feature points are matched with the multiple map points existing in each frame of the map point image respectively. In this embodiment, the LightGlue algorithm is used, and other matching algorithms can also be used. LightGlue is a lightweight feature matching algorithm that uses deep learning technology to improve the speed and accuracy of feature matching. Compared with traditional matching methods, LightGlue can process large-scale data sets more effectively and can provide high-quality matching results while maintaining high efficiency. It can learn how to select the best match based on the context, rather than relying solely on a simple feature distance metric.

[0076] In the specific implementation, for each frame of map point image: use the LightGlue algorithm to match multiple depth feature points with the map points in the image, so as to find the feature point matching pairs of depth feature points and map points in the frame of map point image.

[0077] In a feasible implementation, step S20 may include steps B11 to A13:

[0078] Step B11, performing feature matching on each depth feature point and a plurality of map points existing in at least one frame of map point image to obtain an initial matching pair corresponding to the map point image.

[0079] It should be noted that for each frame of map point image: the LightGlue algorithm is used to match multiple depth feature points with the map points in the image. This process involves comparing the descriptors of the feature points in the two sets, thereby obtaining the initial matching pairs of depth feature points and map points in the map point image.

[0080] Step B12, performing matching verification on the initial matching pairs corresponding to the map point images according to the target homography transformation model to obtain matching verification results.

[0081] It should be noted that the target homography transformation model is a set transformation model that describes the projection relationship within a plane point. The target homography transformation model is a homography transformation model with RANSAC (random sampling consensus algorithm). For the initial matching pairs corresponding to each frame of map point image, the target homography transformation model needs to be used for matching verification to ensure that each initial matching pair meets the expected geometric relationship, thereby obtaining the matching verification result of each initial matching pair.

[0082] Step B13: filtering the initial matching pairs corresponding to the map point image according to the matching verification result to obtain the feature point matching pairs corresponding to the map point image.

[0083] It should be noted that, for the initial matching pairs corresponding to each frame of map point image: the initial matching pairs whose matching verification results are consistent with the expected geometric relationship are selected and used as the feature point matching pairs corresponding to the map point image, and the initial matching pairs whose matching verification results are inconsistent with the expected geometric relationship are filtered out.

[0084] Step S30, when the number of matching pairs of feature point matching pairs corresponding to the map point image is greater than a preset number, posture calculation is performed according to the matching pairs of feature point matching pairs corresponding to the map point image to determine the posture of the flying object corresponding to the current image frame.

[0085] It should be noted that for each frame of map point image: compare whether the number of matching pairs of feature point matching pairs corresponding to the image is greater than the preset number. If the number of matching pairs of feature point matching pairs corresponding to the map point image is greater than the preset number, the posture calculation is performed based on the feature point matching pairs corresponding to the map point image to determine the posture of the flying object corresponding to the current image frame. At this time, the posture of the flying object is the absolute posture of the flying object, which includes the posture information and position information of the flying object. In this embodiment, the preset number can be set to 15, and it can also be set according to the needs. This embodiment does not limit this.

[0086] It is understandable that when there are more than a preset number of feature point matching pairs corresponding to multiple frames of map point images, the feature point matching pairs corresponding to all the above qualified map point images are summarized, and a pnp (Perspective-n-Point Problem) calculation is performed based on the successfully matched depth feature points and map points to determine the position and posture of the flying object. After determining the position and posture of the flying object, subsequent filtering operations can be performed based on the position and posture of the flying object to ensure the accuracy of the operation. In this embodiment,

[0087] In a specific implementation, when the number of matching pairs of feature point matching pairs that do not correspond to the map point image is greater than a preset number, the current image frame is discarded.

[0088] This embodiment extracts feature points from the target object area of ​​the current image frame to obtain multiple depth feature points of the target object area; performs feature matching on each depth feature point with multiple map points existing in at least one frame of map point image to obtain feature point matching pairs corresponding to the map point image; when the number of feature point matching pairs corresponding to the map point image is greater than a preset number, performs posture calculation based on the feature point matching pairs corresponding to the map point image to determine the posture of the flying object corresponding to the current image frame. In the above manner, multiple depth feature points of the target object area in the current image frame are matched with map points in the map point image constructed in advance. When the matching result meets the conditions, the posture of the flying object corresponding to the current image frame is determined through the feature point matching pairs, thereby achieving accurate positioning of the flying object in complex environments.

[0089] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be repeated in the following. Figure 2 , step S30 in the spatial positioning method further includes steps S31 to S33:

[0090] Step S31: extracting a plurality of map points and a plurality of depth feature points from the feature point matching pairs corresponding to the map point image.

[0091] Step S32, performing calculations based on the three-dimensional coordinates of each map point and the pixel coordinates of each depth feature point to determine a target rotation matrix and a target translation vector.

[0092] It should be noted that for a map point image with a feature point matching pair greater than a preset number: obtain the feature point matching pairs corresponding to the image, extract all map points and depth feature points from all feature point matching pairs, and obtain the three-dimensional coordinates of each map point and the pixel coordinates of each depth feature point. At this time, the coordinate system where the three-dimensional coordinates of the map point are located is the coordinate system where the randomly selected reference image frame is located, and the pixel coordinates are also the two-dimensional image coordinates.

[0093] It can be understood that the problem of estimating the posture of the camera corresponding to the current image frame is solved based on the 3D coordinates of all map points and the pixel coordinates of each depth feature point, thereby obtaining the target rotation matrix and target translation vector of the camera relative to the world coordinate system.

[0094] Step S33, determining the posture of the flying object corresponding to the current image frame according to the target rotation matrix and the target translation vector.

[0095] It should be noted that, by combining the target rotation matrix and target translation vector of the camera relative to the world coordinate system, a complete six-degree-of-freedom posture corresponding to the flying object in the current image frame can be formed, including but not limited to three rotation components and three translation components, and finally the flying object posture corresponding to the current image frame is obtained.

[0096] This embodiment extracts multiple map points and multiple depth feature points from the feature point matching pairs corresponding to the map point image; calculates the target rotation matrix and the target translation vector based on the three-dimensional coordinates of each map point and the pixel coordinates of each depth feature point; and determines the position and posture of the flying object corresponding to the current image frame based on the target rotation matrix and the target translation vector. In the above manner, the position and posture of the flying object corresponding to the current image frame can be accurately obtained.

[0097] Based on the first embodiment and / or the second embodiment of the present application, in the third embodiment of the present application, the same or similar contents as those in the first embodiment and / or the second embodiment can refer to the above description and will not be described in detail later. Figure 3 , step S20 in the spatial positioning method further includes steps S21 to S23:

[0098] Step S21 , performing image segmentation on the flying object image frames corresponding to different acquisition conditions, and determining the flying object region of each flying object image frame.

[0099] It should be noted that different acquisition conditions include but are not limited to different acquisition angles and different acquisition distances. Under different acquisition conditions, a binocular camera is used to acquire flying object image frames, image segmentation is performed on all acquired flying object image frames, and masks of flying objects in flying object image frames are extracted, thereby obtaining flying object regions of each flying object image frame.

[0100] Step S22: extract feature points of the flying body region of each flying body image frame according to the target deep learning model to obtain multiple depth feature point data of each flying body region.

[0101] It should be noted that the target deep learning model is used to extract depth feature points from the flying body areas of all flying body image frames, and the depth feature points and their descriptors corresponding to all flying body areas are determined, thereby obtaining multiple depth feature point data for each flying body area.

[0102] Step S23 , constructing map points based on the multiple depth feature point data of each flying body region and the reference image frame to obtain multiple map point images and multiple map points existing in each map point image.

[0103] It should be noted that the reference image frame is a fixed image frame randomly selected from multiple flying body image frames. Map points are constructed based on multiple depth feature point data of each flying body area and the reference image frame to obtain multiple map point images and multiple map points in each map point image.

[0104] In a feasible implementation, step S23 may include steps C11 to C13:

[0105] Step C11 , performing binocular stereo matching on the flying body region of each flying body image frame, and generating a plurality of monocular map points of a plurality of monocular image frames according to the matching results.

[0106] It should be noted that for the flying body area of ​​all flying body image frames: identify the binocular images of the same frame, perform left and right stereo matching on all binocular images of the same frame, the above process involves finding corresponding feature points in the left eye image and the right eye image, and using the RT matrix of the binocular camera to ensure that the matching pairs meet the geometric constraints, thereby eliminating false matches, and performing triangulation calculations based on the matching pairs, thereby generating multiple monocular map points of the monocular image frame. In this embodiment, the monocular image frame is the left eye image selected from the multiple flying body image frames.

[0107] Step C12, retaining multiple depth feature point data of multiple flying body regions to obtain multiple monocular feature points of multiple monocular image frames.

[0108] It should be noted that, since the multiple flying body image frames include left-eye images and right-eye images, the depth feature point data of multiple monocular image frames are retained for the multiple depth feature point data of the multiple flying body areas, thereby obtaining multiple monocular feature points of the multiple monocular image frames.

[0109] Step C13, constructing map points according to the multiple monocular map points of each monocular image frame, the multiple monocular feature points of each monocular image frame and the reference image frame, to obtain multiple map point images and multiple map points existing in each map point image.

[0110] In a feasible implementation, step C11 may include steps D11 to D14:

[0111] Step D11, performing same-frame binocular matching on multiple flying object image frames to determine multiple same-frame binocular image pairs.

[0112] It should be noted that, for the flying body region of all flying body image frames: binocular images of the same frame are identified, thereby obtaining multiple binocular image pairs of the left eye image and the right eye image of the same frame collected at the same timestamp.

[0113] Step D12, stereo matching is performed on the flying object regions of multiple binocular image pairs in the same frame to determine initial matching pairs of each binocular image pair in the same frame.

[0114] It should be noted that stereo matching is performed on the flying object region of all binocular image pairs in the same frame. This process involves finding corresponding feature points in the left and right images and calculating depth information based on the disparity between them, thereby obtaining the initial matching pairs of each binocular image pair in the same frame.

[0115] Step D13, performing mismatch filtering on the initial matching pairs of each binocular image pair in the same frame, and determining the target matching pairs of each binocular image pair in the same frame.

[0116] It should be noted that in order to improve the accuracy of matching, geometric consistency verification (such as the RANSAC algorithm combined with the homography transformation matrix or the essential matrix) is used to eliminate false matches, and the initial matching pairs that pass the verification are used as the target matching pairs of each binocular image pair in the same frame.

[0117] Step D14, performing triangulation calculation based on target matching pairs of binocular image pairs in the same frame, and determining multiple monocular map points of multiple monocular image frames.

[0118] It should be noted that for multiple pairs of binocular images in the same frame: for verified target matching pairs, the triangulation principle is used to calculate their actual coordinates in three-dimensional space. Triangulation is based on the position information of the same point under two different perspectives, combined with the internal and external parameters of the camera, to solve the three-dimensional coordinates of the point. The calculated three-dimensional coordinates are projected back to the left-eye image plane to generate map points with feature information. These map points represent feature points at fixed positions in the environment, and only the corresponding positions in the left-eye image are retained. Finally, the left-eye image is used as a monocular image frame, and multiple monocular map points of multiple monocular image frames are obtained.

[0119] In a feasible implementation, step C13 may include steps E11 to E13:

[0120] Step E11, performing deep feature matching on a plurality of monocular feature points of each monocular image frame to determine feature point matching pairs between each monocular image frame.

[0121] It should be noted that a feature matching algorithm (such as the Lightglue algorithm) is used to perform deep feature matching between multiple monocular feature points in monocular image frames of different frames, and false matches are eliminated through homography transformation with RANSAC to obtain feature point matching pairs between all monocular image frames of different frames.

[0122] Step E12, determining the relative posture of the flying object corresponding to each monocular image frame according to the feature point matching pairs between each monocular image frame and the reference image frame.

[0123] It should be noted that, according to the feature point matching pairs between monocular image frames of different frames, the PNP algorithm is used to calculate the relative pose of each monocular image frame to the reference image frame, so as to obtain the relative pose of the flying object corresponding to each monocular image frame, providing an initial value for subsequent optimization.

[0124] Step E13, globally optimizing the relative posture of the flying object corresponding to each monocular image frame and multiple monocular map points of each monocular image frame, and determining multiple map point images and multiple map points existing in each map point image according to the optimization result.

[0125] It should be noted that the relative posture of the flying object corresponding to each monocular image frame and the multiple monocular map points of each monocular image frame are optimized by BA (Bundle Adjustment) to obtain more accurate data of the relative posture of the flying object and the map points. After the optimization, the three-dimensional coordinates of the multiple monocular map points of each monocular image frame are converted to the coordinate system of the reference image frame through the relative posture of the flying object corresponding to each monocular image frame, and finally the multiple monocular image frames are used as multiple map point images, and the monocular map points existing in the monocular image frames are used as multiple map points existing in each map point image.

[0126] This embodiment determines the flying body region of each flying body image frame by performing image segmentation on the flying body image frames corresponding to different acquisition conditions; extracts feature points of the flying body region of each flying body image frame according to the target deep learning model to obtain multiple depth feature point data of each flying body region; constructs map points according to the multiple depth feature point data of each flying body region and the reference image frame to obtain multiple map point images and multiple map points existing in each map point image. In the above manner, more accurate map points can be generated for use in the subsequent positioning stage, thereby improving the accuracy of subsequent positioning.

[0127] For example, to help understand the implementation process of the spatial positioning method obtained by combining this embodiment with the above-mentioned embodiment 1, please refer to Figure 4 , Figure 4 A brief flowchart of a spatial positioning method is provided, specifically: 1. binocular feature map generation; 2. using a binocular camera to collect images of a flying object at different angles and distances; 3. performing image segmentation on all collected binocular images to extract the mask of the flying object; 4. using a superpoint network to extract deep feature points from the flying object region of all binocular images and save the feature point data; 5. performing left and right stereo matching on all binocular images of the same frame, eliminating mismatches through the RT of the binocular camera, and retaining the generated left-eye map points and the feature point data extracted from the left eye; 6. matching the left-eye feature points of different frames Use lightglue for deep feature matching, and eliminate false matches through homography transformation with ransac to obtain the left-eye matching feature points between all different frames; 7. According to the left-eye matching feature points of different frames, use the pnp algorithm to calculate the relative pose of each frame to a fixed frame, and provide initial values ​​for subsequent optimization; 8. Perform BA optimization on the poses of all frames and the generated map points (where each map point observation comes from all left-eye and right-eye images that observe the map point) to obtain a more accurate pose of each frame and all map point data; 9. Save the optimized left-eye poses and map points of all frames for subsequent positioning. When calculating the posture of the flying object: 1. Monocular matching and positioning; 2. Collect the current monocular image at any time; 3. Use the segmentation network to segment the flying object area of ​​the image; 4. Use the superpoint network to extract the depth feature points of the segmented image area; 5. Use lightglue to match the extracted depth feature points with all map points, and determine the number of matches. If it is less than 15 points, the current frame is discarded; 6. The number of matching pairs>N; 7. Perform pnp calculation on the successfully matched feature points and map points to obtain the posture of the current frame; 8. Get the posture for subsequent filtering operations.

[0128] The method of this embodiment has the following advantages: 1. Compared with traditional feature points, the extraction of key points using superpoint depth feature points no longer relies solely on grayscale information for simple judgment, but extracts feature points based on the geometric structure and texture information of the image area; and at the same time generates feature vectors of key points based on local structure or texture information, avoiding the problem of insufficient feature expression when using grayscale information to generate feature vectors. 2. When using the lightglue deep matching network for image matching, the horizontal and vertical coordinates of the input feature points and the feature vector information are encoded at a higher level, and the attention matching mechanism is used for all feature points to obtain the global optimal matching result; compared with the traditional simple use of similarity calculation or minimum Hamming distance, it has better matching performance. 3. Because the depth feature points and the depth matching mechanism will bring better matching performance at different viewing angles and distances of the flying body, the scenes that cannot be matched by the traditional method can be matched, and the generated map points will be observed at different angles and scales, and more accurate map points will be generated for use in the subsequent positioning stage, improving the accuracy of the subsequent positioning.

[0129] It should be noted that the above examples are only used to understand the present application and do not constitute a limitation on the spatial positioning method of the present application. More simple transformations based on this technical concept are all within the scope of protection of the present application.

[0130] This application also provides a spatial positioning device, please refer to Figure 5 , the spatial positioning device comprises:

[0131] An extraction module 10 is used to extract feature points from a target object region of a current image frame to obtain a plurality of depth feature points of the target object region;

[0132] A matching module 20, configured to perform feature matching on each depth feature point with a plurality of map points existing in at least one frame of map point image, to obtain a feature point matching pair corresponding to the map point image;

[0133] The calculation module 30 is used to perform posture calculation according to the feature point matching pairs corresponding to the map point image when the number of matching pairs of feature point matching pairs corresponding to the map point image is greater than a preset number, so as to determine the posture of the flying object corresponding to the current image frame.

[0134] Optionally, the matching module 20 is further configured to:

[0135] Perform feature matching on each depth feature point and a plurality of map points existing in at least one frame of map point image to obtain an initial matching pair corresponding to the map point image;

[0136] Performing matching verification on the initial matching pairs corresponding to the map point images according to the target homography transformation model to obtain a matching verification result;

[0137] Initial matching pairs corresponding to the map point image are filtered according to the matching verification result to obtain feature point matching pairs corresponding to the map point image.

[0138] Optionally, the extraction module 10 is further used for:

[0139] Extracting a plurality of map points and a plurality of depth feature points from the feature point matching pairs corresponding to the map point image;

[0140] Calculate the target rotation matrix and target translation vector based on the three-dimensional coordinates of each map point and the pixel coordinates of each depth feature point;

[0141] The position and posture of the flying object corresponding to the current image frame is determined according to the target rotation matrix and the target translation vector.

[0142] Optionally, the matching module 20 is further configured to:

[0143] Performing image segmentation on the flying object image frames corresponding to different acquisition conditions to determine the flying object region of each flying object image frame;

[0144] According to the target deep learning model, feature points of the flying body region of each flying body image frame are extracted to obtain multiple depth feature point data of each flying body region;

[0145] Map points are constructed according to a plurality of depth feature point data of each flying body region and a reference image frame to obtain a plurality of map point images and a plurality of map points existing in each map point image.

[0146] Optionally, the matching module 20 is further configured to:

[0147] Perform binocular stereo matching on the flying body area of ​​each flying body image frame, and generate multiple monocular map points of multiple monocular image frames according to the matching results;

[0148] Retaining multiple depth feature point data of multiple flying body regions to obtain multiple monocular feature points of multiple monocular image frames;

[0149] Map points are constructed according to a plurality of monocular map points of each monocular image frame, a plurality of monocular feature points of each monocular image frame and a reference image frame to obtain a plurality of map point images and a plurality of map points existing in each map point image.

[0150] Optionally, the matching module 20 is further configured to:

[0151] Perform same-frame binocular matching on multiple flying object image frames to determine multiple same-frame binocular image pairs;

[0152] Perform stereo matching on the flying object regions of multiple binocular image pairs in the same frame, and determine the initial matching pairs of each binocular image pair in the same frame;

[0153] Perform mismatch filtering on the initial matching pairs of each binocular image pair in the same frame, and determine the target matching pairs of each binocular image pair in the same frame;

[0154] A triangulation calculation is performed based on the target matching pairs of each binocular image pair in the same frame to determine multiple monocular map points of multiple monocular image frames.

[0155] Optionally, the matching module 20 is further configured to:

[0156] Performing deep feature matching on multiple monocular feature points of each monocular image frame to determine feature point matching pairs between each monocular image frame;

[0157] Determine the relative position and posture of the flying object corresponding to each monocular image frame according to the feature point matching pairs between each monocular image frame and the reference image frame;

[0158] The relative posture of the flying object corresponding to each monocular image frame and multiple monocular map points of each monocular image frame are globally optimized, and multiple map point images and multiple map points existing in each map point image are determined according to the optimization results.

[0159] Optionally, the calculation module 30 is further used for:

[0160] Performing image preprocessing on the current image frame to obtain a processed current image frame;

[0161] Segmenting the current image frame through a target segmentation network to determine a target object region of the current image frame;

[0162] The target object area is input into a target deep learning model to obtain a plurality of depth feature points of the target object area and a descriptor of each depth feature point.

[0163] The spatial positioning device provided by the present application adopts the spatial positioning method in the above embodiment, which can solve the technical problem that the prior art cannot accurately position the flying object. Compared with the prior art, the beneficial effects of the spatial positioning device provided by the present application are the same as the beneficial effects of the spatial positioning method provided by the above embodiment, and the other technical features in the spatial positioning device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0164] The present application provides a spatial positioning device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the spatial positioning method in the above-mentioned embodiment one.

[0165] Reference below Figure 6 , which shows a schematic diagram of the structure of a spatial positioning device suitable for implementing the embodiment of the present application. The spatial positioning device in the embodiment of the present application may include but is not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 6 The spatial positioning device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0166] like Figure 6 As shown, the spatial positioning device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the spatial positioning device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following systems can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the spatial positioning device to communicate with other devices wirelessly or wired to exchange data. Although the figure shows a spatial positioning device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented or have alternatively.

[0167] In particular, according to the embodiments disclosed in the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present application are executed.

[0168] The spatial positioning device provided by the present application adopts the spatial positioning method in the above embodiment, which can solve the technical problem that the prior art cannot accurately position the flying object. Compared with the prior art, the beneficial effects of the spatial positioning device provided by the present application are the same as the beneficial effects of the spatial positioning method provided by the above embodiment, and the other technical features in the spatial positioning device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0169] It should be understood that the various parts disclosed in this application can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0170] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

[0171] The present application provides a computer-readable storage medium having computer-readable program instructions (ie, computer programs) stored thereon, wherein the computer-readable program instructions are used to execute the spatial positioning method in the above-mentioned embodiment.

[0172] The computer-readable storage medium provided in the present application may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, systems or devices, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, system or device. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0173] The computer-readable storage medium may be included in the spatial positioning device; or may exist independently without being assembled into the spatial positioning device.

[0174] The above-mentioned computer-readable storage medium carries one or more programs. When the above-mentioned one or more programs are executed by the spatial positioning device, the spatial positioning device: extracts feature points from the target object area of ​​the current image frame to obtain multiple depth feature points of the target object area; performs feature matching on each depth feature point with multiple map points existing in at least one frame of map point image to obtain feature point matching pairs corresponding to the map point image; when the number of matching pairs of feature point matching pairs corresponding to the map point image is greater than a preset number, performs posture calculation based on the feature point matching pairs corresponding to the map point image to determine the posture of the flying object corresponding to the current image frame.

[0175] Computer program code for performing the operations of the present application may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0176] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present application. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0177] The modules involved in the embodiments described in this application may be implemented by software or hardware, wherein the name of the module does not constitute a limitation on the unit itself in some cases.

[0178] The readable storage medium provided by the present application is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned spatial positioning method, and can solve the technical problem that the prior art cannot accurately position the flying object. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as the beneficial effects of the spatial positioning method provided by the above-mentioned embodiment, and will not be repeated here.

[0179] The present application also provides a computer program product, including a computer program, which implements the steps of the above-mentioned spatial positioning method when executed by a processor.

[0180] The computer program product provided by this application can solve the technical problem that the prior art cannot accurately locate the flying object. Compared with the prior art, the beneficial effects of the computer program product provided by this application are the same as the beneficial effects of the spatial positioning method provided by the above embodiment, which will not be repeated here.

[0181] The above descriptions are only some embodiments of the present application, and are not intended to limit the patent scope of the present application. All equivalent structural changes made using the contents of the present application specification and drawings under the technical concept of the present application, or direct / indirect applications in other related technical fields are included in the patent protection scope of the present application.

Claims

1. A spatial positioning method, characterized in that: The spatial positioning method comprises: Extracting feature points from a target object region of a current image frame to obtain a plurality of depth feature points of the target object region; Perform feature matching on each depth feature point with a plurality of map points existing in at least one frame of map point image to obtain a feature point matching pair corresponding to the map point image; When the number of matching pairs of feature point matching pairs corresponding to the map point image is greater than a preset number, posture calculation is performed according to the feature point matching pairs corresponding to the map point image to determine the posture of the flying object corresponding to the current image frame.

2. The method according to claim 1, characterized in that The step of performing feature matching of each depth feature point with a plurality of map points existing in at least one frame of map point image to obtain a feature point matching pair corresponding to the map point image comprises: Perform feature matching on each depth feature point and a plurality of map points existing in at least one frame of map point image to obtain an initial matching pair corresponding to the map point image; Performing matching verification on the initial matching pairs corresponding to the map point images according to the target homography transformation model to obtain a matching verification result; Initial matching pairs corresponding to the map point image are filtered according to the matching verification result to obtain feature point matching pairs corresponding to the map point image.

3. The method according to claim 1, characterized in that The step of calculating the posture of the flying object corresponding to the current image frame by matching the feature points corresponding to the map point image comprises: Extracting a plurality of map points and a plurality of depth feature points from the feature point matching pairs corresponding to the map point image; Calculate the target rotation matrix and target translation vector based on the three-dimensional coordinates of each map point and the pixel coordinates of each depth feature point; The position and posture of the flying object corresponding to the current image frame is determined according to the target rotation matrix and the target translation vector.

4. The method according to any one of claims 1 to 3, characterized in that Before the step of performing feature matching between each depth feature point and a plurality of map points existing in at least one frame of map point image to obtain a feature point matching pair corresponding to the map point image, the step further includes: Performing image segmentation on the flying object image frames corresponding to different acquisition conditions to determine the flying object region of each flying object image frame; According to the target deep learning model, feature points of the flying body region of each flying body image frame are extracted to obtain multiple depth feature point data of each flying body region; Map points are constructed according to a plurality of depth feature point data of each flying body region and a reference image frame to obtain a plurality of map point images and a plurality of map points existing in each map point image.

5. The method according to claim 4, characterized in that The step of constructing map points based on the multiple depth feature point data of each flying body region and the reference image frame to obtain multiple map point images and multiple map points existing in each map point image comprises: Perform binocular stereo matching on the flying body area of ​​each flying body image frame, and generate multiple monocular map points of multiple monocular image frames according to the matching results; Retaining multiple depth feature point data of multiple flying body regions to obtain multiple monocular feature points of multiple monocular image frames; Map points are constructed according to a plurality of monocular map points of each monocular image frame, a plurality of monocular feature points of each monocular image frame and a reference image frame to obtain a plurality of map point images and a plurality of map points existing in each map point image.

6. The method according to claim 5, characterized in that The step of performing binocular stereo matching on the flying body area of ​​each flying body image frame and generating multiple monocular map points of multiple monocular image frames according to the matching results comprises: Perform same-frame binocular matching on multiple flying object image frames to determine multiple same-frame binocular image pairs; Perform stereo matching on the flying object regions of multiple binocular image pairs in the same frame, and determine the initial matching pairs of each binocular image pair in the same frame; Perform mismatch filtering on the initial matching pairs of each binocular image pair in the same frame, and determine the target matching pairs of each binocular image pair in the same frame; A triangulation calculation is performed based on the target matching pairs of each binocular image pair in the same frame to determine multiple monocular map points of multiple monocular image frames.

7. The method according to claim 5, characterized in that The step of constructing map points according to the multiple monocular map points of each monocular image frame, the multiple monocular feature points of each monocular image frame and the reference image frame to obtain multiple map point images and multiple map points existing in each map point image comprises: Performing deep feature matching on multiple monocular feature points of each monocular image frame to determine feature point matching pairs between each monocular image frame; Determine the relative position and posture of the flying object corresponding to each monocular image frame according to the feature point matching pairs between each monocular image frame and the reference image frame; The relative posture of the flying object corresponding to each monocular image frame and multiple monocular map points of each monocular image frame are globally optimized, and multiple map point images and multiple map points existing in each map point image are determined according to the optimization results.

8. The method according to any one of claims 1 to 3, characterized in that The step of extracting feature points from the target object area of ​​the current image frame to obtain a plurality of depth feature points of the target object area comprises: Performing image preprocessing on the current image frame to obtain a processed current image frame; Segmenting the current image frame through a target segmentation network to determine a target object region of the current image frame; The target object area is input into a target deep learning model to obtain a plurality of depth feature points of the target object area and a descriptor of each depth feature point.

9. A spatial positioning device, characterized in that: The spatial positioning device comprises: An extraction module, used to extract feature points from a target object region of a current image frame to obtain a plurality of depth feature points of the target object region; A matching module, used for performing feature matching between each depth feature point and a plurality of map points existing in at least one frame of map point image, to obtain a feature point matching pair corresponding to the map point image; A calculation module is used to perform posture calculation based on the feature point matching pairs corresponding to the map point image when the number of matching pairs of feature point matching pairs corresponding to the map point image is greater than a preset number, so as to determine the posture of the flying object corresponding to the current image frame.

10. A spatial positioning device, characterized in that: The spatial positioning device comprises: a memory, a processor, and a spatial positioning program stored in the memory and executable on the processor, wherein the spatial positioning program is configured to implement the spatial positioning method according to any one of claims 1 to 8.

Images