WO2023092997A1 - 图像特征点选取方法及装置、设备、存储介质和程序产品 - Google Patents

图像特征点选取方法及装置、设备、存储介质和程序产品 Download PDF

Info

Publication number
WO2023092997A1
WO2023092997A1 PCT/CN2022/098194 CN2022098194W WO2023092997A1 WO 2023092997 A1 WO2023092997 A1 WO 2023092997A1 CN 2022098194 W CN2022098194 W CN 2022098194W WO 2023092997 A1 WO2023092997 A1 WO 2023092997A1
Authority
WO
WIPO (PCT)
Prior art keywords
node
feature
feature points
storage space
information
Prior art date
Application number
PCT/CN2022/098194
Other languages
English (en)
French (fr)
Inventor
翟尚进
Original Assignee
上海商汤智能科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 上海商汤智能科技有限公司 filed Critical 上海商汤智能科技有限公司
Publication of WO2023092997A1 publication Critical patent/WO2023092997A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/211—Selection of the most significant subset of features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/51—Indexing; Data structures therefor; Storage structures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
    • G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
    • G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/901—Indexing; Data structures therefor; Storage structures
    • G06F16/9027—Trees

Definitions

  • the embodiments of the present disclosure are based on the Chinese patent application with the application number 202111395862.5, the application date is November 23, 2021, and the application name is "Image feature point selection method and related devices, equipment and storage media", and requires the Chinese patent application
  • the priority of the Chinese patent application, the entire content of this Chinese patent application is hereby incorporated into the embodiments of the present disclosure as a reference.
  • the present disclosure relates to the technical field of image processing, in particular to an image feature point selection method, device, equipment, storage medium and program product.
  • Feature extraction of images is an important underlying technology in the fields of computer vision, robotics, unmanned vehicles, 3D reconstruction and augmented reality. Considering the efficiency of feature point processing such as subsequent feature matching, feature points obtained after image feature extraction are often selected to reduce the number of feature points.
  • Embodiments of the present disclosure provide an image feature point selection method, device, device, storage medium, and program product.
  • the first aspect of the embodiments of the present disclosure provides a method for selecting image feature points, including: acquiring a number of feature points in an image frame; pre-configuring a fixed-size node storage space based on the number of feature points; The point is divided into at least one node, and the first information of the node is stored in the node storage space, wherein the first information of the node includes the feature related information of the feature point corresponding to the node; the feature is selected from each node obtained by the final division points to obtain the selection results of several feature points.
  • the second aspect of the embodiments of the present disclosure provides an image feature point selection method, including: acquiring multiple feature maps of an image frame, wherein each feature map has a different resolution; performing the following feature point selection on each feature map in parallel : Obtain several feature points from the feature map, and obtain the selection results of the several feature points, wherein the selection results of the several feature points are selected from the several feature points by using the method described in the first aspect above.
  • the device that implements the image feature point selection method of the embodiment of the present disclosure can simultaneously perform feature point selection on multiple feature maps with different resolutions, and obtain feature The result of point selection improves the overall execution efficiency of feature point selection for multiple feature maps with different resolutions.
  • the third aspect of the embodiment of the present disclosure provides an image feature point selection device, including: an acquisition part, a memory application part, a feature point division part and a feature point selection part; the acquisition part is configured to acquire several feature points in an image frame; The application part is configured to pre-configure a fixed-size node storage space based on several feature points; the feature point division part is configured to divide several feature points into at least one node by using tree division, and store the first information of the node in In the node storage space, the first information of the node includes the feature-related information of the feature point corresponding to the node; the feature point selection part is configured to select a feature point from each node obtained by the final division, so as to obtain a selection of several feature points result.
  • the fourth aspect of the embodiment of the present disclosure provides an image feature point selection device, including: an acquisition part, a feature point selection part; an acquisition part configured to acquire multiple feature maps of an image frame, wherein the resolution of each feature map is Different; the feature point selection part is configured to perform the following feature point selection in parallel for each feature map: obtain several feature points from the feature map, and use the method described in the first aspect above to obtain the selection results of several feature points.
  • the fifth aspect of the embodiment of the present disclosure provides an electronic device, including a memory and a processor coupled to each other, the processor is configured to execute the program instructions stored in the memory, so as to realize the image features described in the first aspect and the second aspect above Click to select a method.
  • the sixth aspect of the embodiments of the present disclosure provides a computer-readable storage medium, on which program instructions are stored.
  • program instructions are executed by a processor, the image feature point selection methods described in the first aspect and the second aspect are implemented.
  • the seventh aspect of the embodiments of the present disclosure provides a computer program product.
  • the computer program product includes a non-transitory computer-readable storage medium storing a computer program. When the computer program is read and executed by a computer, the foregoing method is implemented.
  • Fig. 1 is a schematic diagram of the first flow chart of the first embodiment of the image feature point selection method of the present disclosure
  • Fig. 2 is an embodiment of adopting a tree-shaped division mode to divide feature points in the feature point selection method of the present disclosure
  • Fig. 3 is a second schematic flow diagram of the first embodiment of the image feature point selection method of the present disclosure
  • FIG. 4 is a schematic diagram of a third flow chart of the first embodiment of the image feature point selection method of the present disclosure
  • Fig. 5 is a schematic diagram of adjusting the storage position of the second information in the feature storage space in the image feature point selection method of the present disclosure
  • FIG. 6 is another schematic diagram of adjusting the storage location of the second information in the feature storage space in the image feature point selection method of the present disclosure
  • FIG. 7 is a schematic diagram of a fourth flowchart of the first embodiment of the image feature point selection method of the present disclosure.
  • Fig. 8 is a schematic flow chart of the second embodiment of the image feature point selection method of the present disclosure.
  • FIG. 9 is a schematic frame diagram of an embodiment of an image feature point selection device of the present disclosure.
  • Fig. 10 is a schematic frame diagram of another embodiment of the image feature point selection device of the present disclosure.
  • Fig. 11 is a schematic frame diagram of an embodiment of an electronic device of the present disclosure.
  • FIG. 12 is a schematic diagram of an embodiment of a computer-readable storage medium of the present disclosure.
  • system and “network” are often used interchangeably herein.
  • the term “and/or” in this article is just an association relationship describing associated objects, which means that there can be three relationships, for example, A and/or B can mean: A exists alone, A and B exist simultaneously, and there exists alone B these three situations.
  • the character "/" in this paper generally indicates that the associated objects are an "or” relationship.
  • "many” herein means two or more than two.
  • the method for selecting image feature points in the embodiments of the present disclosure can be applied to technical fields such as computer vision, robotics, unmanned vehicles, three-dimensional reconstruction, and augmented reality.
  • the device used to implement the image feature point selection method of the embodiments of the present disclosure may be electronic devices such as computers, mobile phones, tablet computers, and smart glasses.
  • FIG. 1 is a schematic diagram of the first flow chart of the first embodiment of the image feature point selection method of the present disclosure.
  • the method may include the following steps:
  • Step S11 Obtain several feature points in the image frame.
  • the image frame may be an image captured by an electronic device such as a mobile phone, a tablet computer, or smart glasses, or an image captured by a surveillance camera, which is not limited herein.
  • an electronic device such as a mobile phone, a tablet computer, or smart glasses
  • an image captured by a surveillance camera which is not limited herein.
  • Other scenarios can be deduced in the same way, and examples are not given here.
  • the image frame may be an image obtained by removing edge regions from the original image. For example, after removing a certain number of pixels on the edge of the original image, the obtained image is an image frame.
  • feature extraction can be performed on the image frame to obtain several feature points.
  • the obtained feature points are stored in memory in the form of an array.
  • Feature extraction algorithms such as FAST (Features From Accelerated Segment Test, an algorithm for corner detection) algorithm, SIFT (Scale-Invariant Feature Transform, scale-invariant feature transformation) algorithm, ORB (Oriented FAST and Rotated BRIEF, a A fast feature point extraction and description algorithm) algorithm and so on.
  • the feature extraction algorithm is an ORB (Oriented FAST and Rotated BRIEF) algorithm.
  • a feature representation corresponding to each feature point can also be obtained, and the feature representation is, for example, a feature vector.
  • image acquisition and feature extraction may be performed by a device that executes the image feature point selection method of the embodiment of the present disclosure, so as to obtain several feature points.
  • the feature points obtained by feature extraction of image frames performed by other devices may also be directly acquired by the device executing the image feature point selection method of the embodiment of the present disclosure.
  • a feature storage space may be applied for the several feature points, and the second information of the several feature points is stored in the feature storage space.
  • the feature storage space can be the memory space of memory. Therefore, by applying for the feature storage space, the storage location of the feature points can be determined, which facilitates subsequent division of the feature points.
  • the feature storage space may be an array in the memory, which is defined as the first array.
  • the first array contains a certain number of elements.
  • each element in the first array is used to store the second information of a feature point. Therefore, by determining the specific form of the feature storage space, it will be convenient to divide the feature points later.
  • the second information of the feature point includes at least one of the following: position information of the feature point in the image frame, responsivity of the feature point, and identification of the feature map to which the feature point belongs.
  • the position information of the feature point in the image frame may be the pixel coordinates of the feature point.
  • the responsivity of the feature point may be a specific value determined according to the feature extraction algorithm, for example, the response value of the feature point determined when the feature point is extracted by the FAST feature extraction algorithm.
  • the feature map to which the feature point belongs is one of a plurality of feature maps of different resolutions corresponding to the image frame, and the identification of the feature map to which the feature point belongs can be the feature map of the feature point to which the feature map belongs in multiple different resolutions number in .
  • Multiple feature maps with different resolutions may be obtained by feature extraction based on image pyramids of image frames, or different feature extractions may be performed on image frames to obtain multiple feature maps with different resolutions. Therefore, by determining the second information of the feature points, the information can be used for subsequent division of the feature points.
  • Step S12 Pre-configure a fixed-size node storage space based on several feature points.
  • Dividing the feature points may be to divide several feature points into at least one node in a tree-shaped manner. Therefore, before performing the step of dividing the feature points, a fixed-size node storage space may be pre-configured based on several feature points, and the fixed-size node storage space may be used to store related information of the divided feature points. In one embodiment, a memory space with a fixed size is applied in the memory as the node storage space. The tree division method can be selected as required. In one embodiment, the node storage space is the second array in memory.
  • Step S13 Divide several feature points into at least one node by means of tree division, and store the first information of the node in the node storage space.
  • the node storage space After the node storage space is determined, several feature points can be divided into at least one node by means of tree division.
  • the number of nodes can be set according to needs, and there is no limit here.
  • Divide several feature points into at least one node in a tree-shaped division method For an image frame, the image frame is divided into several areas, each area corresponds to a node, and the feature points contained in each area are determined. Each area contains The feature points of are the feature points corresponding to the nodes corresponding to the region.
  • the tree division manner includes at least one of a quadtree division manner and an octree division manner. By determining the specific way of tree division, the feature points can be divided into quadtrees or octrees.
  • the feature points on the boundaries may be divided into several image regions as required.
  • the first information of the nodes may be stored in the node storage space.
  • the first information of the node may include feature related information of the feature point corresponding to the node.
  • the feature related information of the feature point corresponding to the node can be understood as the related information of the feature point corresponding to the node, such as the position information of the feature point in the image, the storage location information of the feature point in the memory, and the feature point’s feature Represent information (such as feature vectors) and so on.
  • the feature information corresponding to the node includes the storage location of the feature point corresponding to the node in the feature storage space. For example, when the feature storage space is an array, the storage position of the feature point corresponding to the node in the feature storage space is the subscript of the element stored in the feature point in the array (feature storage space).
  • the node storage space is a second array in memory, and each element in the second array is used to store the first information of a node.
  • the node storage space is a one-way linked list, and each node in the one-way linked list is used to store the first information of a node. Therefore, by determining the specific storage form of the node storage space, it can be used for subsequent division of feature points.
  • the first information of the node further includes at least one of the following: the position information of the region corresponding to the node in the image frame, whether the node is a preset node, the location of the next node pointed to by the node in the node storage space Store location information and the layer identifier of the node in the tree structure obtained by tree division.
  • the location information of the region corresponding to the node in the image frame may be pixel coordinates of four vertices of the region corresponding to the node in the image frame.
  • the preset nodes may be leaf nodes.
  • the storage position information of the next node pointed to by the node in the node storage space, the storage position information of the next node in the node storage space may be the subscript of the position of the next node in the second array. Therefore, by determining the specific information included in the first information of the node, the specific information can be used for subsequent division of feature points.
  • FIG. 2 is an embodiment of dividing feature points in a tree-shaped division in the image feature point selection method of the present disclosure.
  • the tree division method is a quadtree division method.
  • the image frame 10 is divided, four regions are obtained in the first division, namely the region 11 , the region 12 , the region 13 and the region 14 , and the layer identifier of these four regions is 1.
  • the area 11 can be divided to obtain four areas, namely the area 111 , the area 112 , the area 113 , and the area 114 .
  • Step S14 Select feature points from each node obtained by the final division to obtain a selection result of several feature points.
  • a certain number of nodes can be obtained, and the area of the image frame corresponding to the nodes will contain a certain number of feature points.
  • feature points can be selected from the nodes obtained by the final division to obtain the selection results of several feature points. That is, a certain number of feature points are selected from each node obtained by the final division, which is used as the selection result of the feature points.
  • the second number of feature points may be selected from each final division node, so as to obtain a selection result of several feature points.
  • the second quantity can be set according to, which is not limited here. For example, one feature point may be selected from each node obtained in the final division, and the feature point selected from each node is the selection result of several feature points.
  • FIG. 3 is a schematic diagram of a second flow chart of the first embodiment of the image feature point selection method of the present disclosure.
  • This embodiment is a further extension of the "pre-configure a fixed-size node storage space based on several feature points" mentioned in the above steps, including the following steps S121 and S122.
  • Step S121 Based on the number of target selected feature points, determine the number of target node divisions.
  • the number of feature points for target selection is the number of feature points that need to be selected from several feature points, and the specific number of feature points for target selection can be set as required.
  • the number of target node divisions is the number of nodes obtained through final division. In one embodiment, the number of target node divisions may be the same as the number of target selected feature points. For example, if 10 feature points need to be selected from the number of feature points selected from several feature points as target feature points for selection, then the number of target node divisions may be 10 nodes.
  • Step S122 Apply for a node storage space whose size matches the number of target node divisions for several feature points.
  • the area of the image frame corresponding to these nodes will contain several feature points extracted from the image frame. Therefore, you can apply for a number of feature points to store nodes whose size matches the number of target node divisions.
  • the space is used to store the first information storage of the node.
  • the size of the first information of the node that a node needs to store in the memory can be determined first, and then apply for a node storage space in the memory whose size matches the number of target node divisions.
  • a memory space equal to the space of the second information of several feature points in the memory may be applied for in the memory, and then according to the size of the first information of the nodes stored in each node in the memory.
  • the node storage space can store the first information of all the nodes that are finally divided, so that there is no need to apply for a new storage space in the future, thereby reducing the need for storage
  • the spatial operation can improve the execution efficiency of feature point selection.
  • FIG. 4 is a schematic diagram of a third flowchart of the first embodiment of the image feature point selection method of the present disclosure.
  • This embodiment is a further extension of the step of "dividing several feature points into at least one node by means of tree division and storing the first information of the node in the node storage space" mentioned in the above steps, including:
  • Step S131 Divide some or all of the feature points into a node, and store the first information of the node in the node storage space.
  • the first choice is to divide all the feature points that need to be divided into one node.
  • some or all of the feature points may be divided into one node.
  • a node is determined first, and this node can be regarded as a root node. At this time, it can be considered that all feature points are classified into one node, or part of the feature points are classified into the root node.
  • an area in the image frame other than the preset boundary area is used as the root area.
  • the preset boundary area can be considered as an edge area of the image frame, such as an area within a certain number of pixels from the edge of the image. Then, the feature points in the root area are divided into one node. By removing the edge area of the image frame, the feature points on the edge area can be removed, the number of feature points can be reduced, and the execution speed of the image feature point selection method can be accelerated.
  • the feature corresponding to the currently divided node is detected Executed when the number is greater than the number of target selected feature points. That is, after some or all of the feature points are divided into one node, it is determined whether it is necessary to perform subsequent sequential node pairing by judging whether the number of features corresponding to the currently divided node is greater than the number of target selected feature points.
  • Each node that holds the space performs the partitioning step.
  • the number of features corresponding to the currently divided node is not greater than the number of target selected feature points, it means that all the feature points corresponding to the node can be directly determined as the selected feature points, so there is no need to execute Subsequent sequentially divided steps. Therefore, by judging whether the feature number corresponding to the currently divided node is greater than the target selected feature point number, it can be determined whether it is necessary to continue performing the subsequent step of dividing each node in the node storage space in sequence.
  • Step S132 perform the division step on each node in the node storage space in sequence until the node storage space no longer meets the first requirement; wherein, the division step includes: determining the node as a parent node, and dividing the feature points corresponding to the parent node to at least one new node, and store the first information of the new node in the node storage space.
  • the nodes can be divided sequentially, and after the division, the first information corresponding to the divided nodes is stored in the node storage space. In this way, the division of the feature points in the feature point selection process can be realized.
  • the root node is divided first to obtain four nodes, and then these four nodes are respectively used as parent nodes, and the four parent nodes are divided sequentially. After the division is completed, the newly divided nodes are used as the parent nodes, and these parent nodes are divided in order again.
  • Sequential division that is, division according to a certain order.
  • the nodes obtained by dividing the parent node are the child nodes of the parent node.
  • the execution order of the child nodes of the child nodes can be determined according to the number of feature points corresponding to the child nodes.
  • the step of dividing the child nodes may be performed according to the execution order of the child nodes, wherein the child nodes are nodes obtained by dividing the parent node. That is, for child nodes belonging to the same parent node, the order in which the division steps are performed depends on the number of feature points corresponding to the child nodes.
  • the node with the larger number of feature points the earlier the execution order of the division steps, that is, the order of execution is determined according to the number of feature points corresponding to the node.
  • a node is divided into four sub-nodes, which are A sub-node, B sub-node, C sub-node and D sub-node.
  • the relationship between the number of feature points corresponding to the node is: A>B>C>D, and the The A child node performs the division step, then the B child node performs the division step, and so on. Therefore, by preferentially dividing the sub-nodes corresponding to a large number of feature points, the nodes corresponding to a large number of feature points can be preferentially divided, and the uniformity of the divided feature points can be improved.
  • the order in which the division steps are performed may also depend on the maximum responsiveness of the feature points corresponding to the nodes. Nodes with greater maximum responsiveness are executed first. Also taking the above four nodes as an example, the relationship between the maximum responsiveness of the feature points corresponding to the nodes is: A>B>C>D, the division step will be performed on node A first, and then the division step on node B , and so on.
  • the order in which the division steps are performed may also be based on the sum of the responsivity of all the feature points corresponding to the node, and the node with the larger sum of the responsivity is the first to be classified. implement.
  • the relationship between the sum of the responsivity of the feature points corresponding to the nodes is: A>B>C>D, the division step will be performed on the A node first, and then the B node will be divided steps, and so on.
  • the newly divided nodes (the nodes obtained by dividing the A node, the B node, the C node and the D node) will be divided. In this way, by determining that the nodes belonging to the same parent node are preferentially divided, the uniformity of the divided feature points is improved, and the rationality of subsequent feature selection is improved.
  • node when a node is divided, it means that the node no longer stores and node storage space.
  • the current number of storage nodes in the node storage space is the undivided nodes included in the node storage space, because the divided nodes are no longer stored in the node storage space.
  • the first requirement is that the number of currently stored nodes in the node storage space is less than the number of target node divisions. At this time, it means that the current number of storage nodes is equal to or greater than the target number of nodes, so the division can no longer be continued. Therefore, by determining that the first requirement is that the number of currently stored nodes in the node storage space is less than the target number of node divisions, the step of performing node division can be stopped after the number of nodes meeting the required number is obtained.
  • the "determining the node as the parent node" mentioned in this step means that the node will be used as the parent node only when the number of feature points corresponding to the node meets the second requirement. That is, when the number of feature points corresponding to a node satisfies the second requirement, the node will be used as a parent node, and the feature points corresponding to the node will be divided into at least one new node.
  • the second requirement is, for example, when the number of feature points corresponding to a node is greater than the first threshold, the node will be used as the parent node.
  • the first threshold is, for example, 3, 4 or 5, etc., which can be set according to needs during implementation.
  • the number of nodes to be divided can be reduced. Therefore, by judging whether the number of feature points corresponding to a node meets the second requirement, the node that does not meet the second requirement can be used as a parent node to improve the uniformity of feature point division, and also reduce the number of nodes that need to be divided. The execution efficiency of image feature point selection.
  • the storage position of the second information in the feature storage space may be adjusted, so that the second information of the feature point corresponding to the same node is stored in an adjacent position.
  • the storage location of the second information in the feature storage space can be adjusted so that the corresponding to A node The second information of the feature points is stored in adjacent positions.
  • FIG. 5 is a schematic diagram of adjusting the storage position of the second information in the feature storage space in the image feature point selection method of the present disclosure.
  • the area 10 is a certain area of the image frame corresponding to a node Y.
  • the storage locations of the second information of the nine feature points in the feature storage space 20 are shown in part a of FIG. 5 .
  • node Y is determined as a parent node, and node Y is divided into four new nodes to obtain node A, node B, node C and node D respectively.
  • area 10 will also be divided into four new areas, namely area 11, area 12, area 13 and area 14.
  • the feature points corresponding to node A are feature point 1 and feature point 5
  • the feature points corresponding to node B are feature point 2 and feature point 6
  • the feature points corresponding to node C are feature point 7 and feature point 9.
  • the feature points corresponding to node D are feature point 3, feature point 4, and feature point 8.
  • the storage position of the second information in the feature storage space 20 can be adjusted to obtain the adjusted feature storage space 21.
  • the storage positions of the second information of the nine feature points in the feature storage space 21 are shown in part b of FIG. 5 . It can be seen that at this time, the second information of the feature points corresponding to the same node is stored in adjacent positions.
  • two pointers can be set to adjust the storage position of the second information in the feature storage space.
  • the two pointers are pointer A and pointer B respectively, and the second information sorting step of feature points is performed by using these two pointers.
  • the two pointers point to the starting position of sorting respectively.
  • the starting position is the first element of the second information of the corresponding feature point of the divided parent node in the feature storage space.
  • pointer A hereinafter referred to as the feature point pointed to by the pointer
  • pointer A points to the next feature point.
  • the end position is the last element in the feature storage space of the second information of the feature point corresponding to the divided parent node.
  • the second information of the feature points corresponding to the same node can be stored in adjacent positions.
  • the starting position will be adjusted, that is, the second information that has been adjusted as the feature point stored in the adjacent position
  • the element where the is located excludes the first element after that.
  • FIG. 6 is another schematic diagram of adjusting the storage location of the second information in the feature storage space in the image feature point selection method of the present disclosure.
  • node Y is determined as the parent node for division, and A node, B node, C node and D node are obtained.
  • the storage position of the second information of the feature point corresponding to node A may be adjusted first.
  • the starting position at this time is the position of the second information corresponding to feature point 1 in the feature storage space
  • the end position is the position of the second information corresponding to feature point 9 in the feature storage space
  • the pointer A and pointer B point to feature point 1. Execute the above sorting steps, and stop sorting when the feature point pointed to by pointer A exceeds feature point 9.
  • the position of each element in the feature storage space is shown in part b of FIG. 6 .
  • the starting position at this time is the position of the second information corresponding to the feature point 3 in part b of Figure 6 in the feature storage space, and the end position It is still the position of the second information corresponding to the feature point 9 in the feature storage space.
  • the second information of the feature points corresponding to the same node can be stored in adjacent positions. Because the first information of the node stored in the node storage space will include the feature-related information of the feature point corresponding to the node, at this time the feature-related information of the feature point corresponding to the node can include the starting point of the feature point corresponding to the node in the feature storage space start storage position and end storage position.
  • the feature-related information of the feature point corresponding to node D may include the initial storage position of feature point 3 in the feature storage space, and the end storage position of feature point 8 in the feature storage space.
  • FIG. 7 is a schematic diagram of a fourth flowchart of the first embodiment of the image feature point selection method of the present disclosure.
  • This embodiment is a further extension of "dividing the feature points corresponding to the parent node into at least one new node" mentioned in the above steps, including:
  • Step S1321 Divide the parent area corresponding to the parent node into a first number of sub-areas.
  • the parent area corresponding to the parent node may be divided into a first number of sub-areas, the first number being 4 or 8, for example.
  • the image frame is also divided into several areas, so that the node corresponding to a certain area can be determined, and the feature point corresponding to the node can be determined. Therefore, the parent area can be considered as the area in the image frame of the feature point corresponding to the parent node. For example, in part b of FIG. 5 , if node D is determined as the parent node, then the feature point 3, feature point 4, and feature point 8 corresponding to node D are the parent area in the image frame area 14.
  • Step S1322 Determine whether the number of feature points in the sub-region meets the third requirement.
  • the third requirement is that the number of feature points in the sub-region is greater than the second threshold.
  • the second threshold may be zero. Therefore, by judging whether the number of feature points in the sub-area is greater than the second threshold, the area with the number of feature points less than the second threshold can be excluded.
  • Step S1323 When the number of feature points in the sub-region meets the third requirement, generate corresponding child nodes for the sub-region.
  • the feature points corresponding to the sub-nodes are the feature points in the sub-region.
  • Step S1324 When the number of feature points in the sub-region does not meet the third requirement, no corresponding child node is generated for the sub-region.
  • the "store the first information of the new node in the node storage space" mentioned in the above step S132 may be: overwrite the first information of the parent node with the first information of a new node For storage, the first information of the remaining new nodes is stored in the remaining storage space in the node storage space.
  • the remaining storage space can be understood as no storage space for storing the first information of the node in the node storage space. It can be understood that after the parent node is divided, it can be considered that the node no longer exists. Therefore, the first information of one of the new nodes obtained from the division can be stored over the first information of the parent node, Then store the first information of the remaining new nodes in the remaining storage space in the node storage space. Therefore, by using the first information of a new node to overwrite the first information of the parent node for storage, the storage space can be reused and the storage space can be saved.
  • the image feature point selection method of the disclosed embodiment may further include the step: The nodes of the nodes are linked to each other, and the node at the end of the link points to the node pointed to by the parent node, so that the node pointing to the parent node is updated to point to the node at the head of the link; among them, the node The pointers of are used to determine the execution order of the nodes.
  • the storage position information of the pointed next node in the node storage space may be stored in the first information of each node, so as to perform link pointing between nodes.
  • node A points to node B
  • node C points to node A
  • Node A is divided into node A1, node A2, node A3, and node A4.
  • Link-pointing between nodes belonging to the same parent node, and make the node at the end of the link point to the node pointed to by the parent node, and update the node pointing to the parent node to point to the node at the head of the link which may be that node C points to node A1, node A1 points to node A2, node A2 points to node A3, node A3 points to node A4, and node A4 points to node B.
  • the above scheme by applying for a fixed-size node storage space, and storing the first information of the node in the node storage space, makes it unnecessary to repeatedly apply for memory space to the memory when using the tree-shaped division method to divide several feature points To store the first information of the node, reducing the operation on the memory, so as to improve the execution efficiency of the feature point division method.
  • FIG. 8 is a schematic flowchart of a second embodiment of an image feature point selection method according to an embodiment of the present disclosure. This embodiment includes the following steps:
  • Step S21 Obtain multiple feature maps of the image frame, where each feature map has a different resolution.
  • Multiple feature maps with different resolutions may be obtained by feature extraction based on the image pyramid of the image frame, or different feature extractions may be performed on one image frame to obtain multiple feature maps with different resolutions.
  • Step S22 Perform the following feature point selection in parallel for each feature map: obtain several feature points from the feature map, and obtain the selection results of several feature points.
  • the selection result of obtaining several feature points of each feature map is obtained by selecting feature points according to the technical solution described in the first embodiment of the image feature point selection method above.
  • the device that implements the image feature point selection method of the embodiment of the present disclosure can simultaneously perform feature point selection on multiple feature maps with different resolutions, and obtain feature The result of point selection improves the overall execution efficiency of feature point selection for multiple feature maps with different resolutions.
  • the more feature points the feature map contains when executed in parallel for multiple feature maps with different resolutions, the more processing resources are used to perform feature point selection on the feature map. It can be understood that the more the total number of feature points contained in the feature map, the more the number of feature points that need to be divided, and the more computing power is required. Therefore, by determining when the feature map contains more total feature points, the The more processing resources are required to perform feature point selection on the feature map, so as to improve the overall execution efficiency of feature point selection.
  • the overall execution efficiency is improved by performing the image feature point selection method on multiple feature maps with different resolutions.
  • the writing order of each step does not mean a strict execution order and constitutes any limitation on the implementation process.
  • the specific execution order of each step should be based on its function and possible
  • the inner logic is OK.
  • FIG. 9 is a schematic frame diagram of an embodiment of an image feature point selection device of the present disclosure.
  • the image feature point selection device 90 includes an acquisition part 91 , a memory application part 92 , a feature point division part 93 and a feature point selection part 94 .
  • the acquisition part 91 is configured to obtain some feature points in the image frame; the memory application part 92 is configured to pre-configure a fixed-size node storage space based on several feature points; the feature point division part 93 is configured to divide several feature points into Divide into at least one node, and store the first information of the node in the node storage space, wherein, the first information of the node includes the feature-related information of the feature point corresponding to the node; the feature point selection part 94 is configured to obtain from the final division Feature points are selected from each node to obtain a selection result of several feature points.
  • the above-mentioned memory application part 92 is configured to pre-configure a fixed-size node storage space based on several feature points, including: selecting the number of feature points based on the target, and determining the number of target node divisions.
  • the number of feature points selected in , the number of target node divisions is the number of nodes obtained by the final division; apply for a node storage space whose size matches the number of target node divisions for several feature points.
  • the above-mentioned feature point division part 93 is configured to divide several feature points into at least one node by means of tree division, and store the first information of the node in the node storage space, including: part or all of the several feature points
  • the feature points are divided into a node, and the first information of the node is stored in the node storage space; the division step is performed on each node in the node storage space in sequence, until the node storage space no longer meets the first requirement; wherein, the division step The method includes: determining the node as the parent node, dividing the feature points corresponding to the parent node into at least one new node, and storing the first information of the new node in the node storage space.
  • the execution order of other nodes belonging to the same parent node as the currently executed node is prior to the execution order of the nodes obtained by dividing the currently executed node;
  • the node executes the division step, including: for the child nodes belonging to the same parent node, the sub-node execution order of the sub-nodes is determined according to the number of feature points corresponding to the sub-nodes, and the sub-nodes are divided according to the sub-node execution order, wherein the sub-nodes are The node obtained by dividing the parent node; and/or, after the feature point dividing part 93 divides the feature point corresponding to the parent node into at least one new node, the feature point dividing part 93 can also be configured to divide the nodes belonging to the same parent node Link pointing between links, and make the node at the end of the link point to the node pointed to by the parent node, and make the node pointing to the parent node point to the node at the chain head
  • the above-mentioned feature point dividing part 93 is configured to store the first information of the new node in the node storage space, including: storing the first information of a new node overwriting the first information of the parent node, and the remaining new node The first information of the nodes is stored in the remaining storage space in the node storage space.
  • the above-mentioned feature point dividing part 93 is configured to determine the node as the parent node, including: when the number of feature points corresponding to the node meets the second requirement, using the node as the parent node; and/or, the feature point dividing part 93 It is configured to divide the feature point corresponding to the parent node into at least one new node, including: dividing the parent area corresponding to the parent node into a first number of sub-areas, wherein the parent area is the feature point corresponding to the parent node in the image frame area; when the number of feature points in the sub-area meets the third requirement, generate corresponding sub-nodes for the sub-area, where the feature points corresponding to the sub-nodes are the feature points in the sub-area; the number of feature points in the sub-area If the third requirement is not met, no corresponding child node is generated for the sub-area.
  • the first requirement in the above scheme is that the number of currently stored nodes in the node storage space is less than the number of target node divisions; the second requirement is that the number of feature points corresponding to the node is greater than the first threshold; the third requirement is that the number of feature points in the sub-region The number is greater than a second threshold; the first number is 4 or 8.
  • the above-mentioned feature point division part 93 is configured to divide some or all of the feature points into a node, including: taking the area in the image frame other than the preset boundary area as the root area; Points are divided into a node; and/or, the step of dividing each node in the node storage space is performed in sequence until the node storage space no longer meets the first requirement.
  • the step is to detect the feature number corresponding to the currently divided node Executed when the number of selected feature points is greater than the target.
  • the device 90 also includes a feature point space application part.
  • the feature point space application part is configured to apply for a feature storage space for the several feature points, and store the second information of the several feature points in the feature storage space; wherein,
  • the feature information corresponding to the node includes the storage position of the feature point corresponding to the node in the feature storage space.
  • the feature-related information of the feature point corresponding to the node includes the start storage position and the end storage position of the feature point corresponding to the node in the feature storage space.
  • the feature point division part 93 is also configured to adjust the storage position of the second information in the feature storage space, so that the second information of the feature points corresponding to the same node is stored in the adjacent position.
  • the feature storage space is the first array, and each element in the first array is configured to store the second information of a feature point; and/or, the second information of the feature point includes at least one of the following: The position information of the point in the image frame, the responsivity of the feature point, and the identification of the feature map to which the feature point belongs, wherein the feature map to which the feature point belongs is one of a plurality of feature maps with different resolutions corresponding to the image frame.
  • the node storage space is the second array, and each element in the second array is configured to store the first information of a node; or, the node storage space is a one-way linked list, and each node in the one-way linked list The point is configured to store the first information of a node; and/or, the first information of the node also includes at least one of the following: the position information of the corresponding area of the node in the image frame, whether the node is a preset node, the node pointed to The storage position information of the next node in the node storage space, and the layer identifier of the node in the tree structure obtained by the tree division method.
  • the tree division method includes at least one of the quadtree division method and the octree division method; and/or, the feature point selection part 94 is configured to select a feature from each node obtained by the final division points to obtain a selection result of several feature points, including: respectively selecting a second number of feature points from each final division node to obtain a selection result of several feature points.
  • FIG. 10 is a schematic diagram of another embodiment of an image feature point selection device of the present disclosure.
  • the image feature point extraction device 10 includes an acquisition section 100 and a feature point extraction section 110 .
  • the acquiring part 100 is configured to acquire a plurality of feature maps of an image frame, wherein each feature map has a different resolution;
  • the feature point selecting part 110 is configured to perform the following feature point selection in parallel for each feature map: acquire several features from the feature map points, and the selection results of several feature points, wherein the selection results of several feature points are obtained by using the above-mentioned embodiment of the image feature point selection method.
  • the more the total number of feature points contained in the feature map in the above solution the more processing resources are used to perform feature point selection on the feature map.
  • FIG. 11 is a schematic frame diagram of an embodiment of the electronic device of the present disclosure.
  • the electronic device 11 includes a memory 111 and a processor 112 coupled to each other, and the processor 112 is configured to execute program instructions stored in the memory 111, so as to implement the steps of any one of the image feature point selection method embodiments described above.
  • the electronic device 11 may include, but is not limited to: a microcomputer and a server.
  • the electronic device 11 may also include mobile devices such as notebook computers and tablet computers, which are not limited herein.
  • the processor 112 is configured to control itself and the memory 111 to implement the steps of any one of the above embodiments of the image feature point selection method.
  • the processor 112 may also be called a CPU (Central Processing Unit, central processing unit).
  • the processor 112 may be an integrated circuit chip with signal processing capabilities.
  • the processor 112 can also be a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field-programmable gate array (Field-Programmable Gate Array, FPGA) or other Programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.
  • a general-purpose processor may be a microprocessor, or the processor may be any conventional processor, or the like.
  • the processor 112 may be jointly realized by an integrated circuit chip.
  • FIG. 12 is a schematic frame diagram of an embodiment of a computer-readable storage medium of the present disclosure.
  • the computer-readable storage medium 120 stores program instructions 121 that can be executed by the processor, and the program instructions 121 are used to implement the steps of any one of the above embodiments of the image feature point selection method.
  • the foregoing solution can improve the execution efficiency of the feature point division method.
  • the functions or parts included in the apparatus provided in the embodiments of the present disclosure can be configured to execute the methods described in the above method embodiments, and for the specific implementation, refer to the descriptions of the above method embodiments.
  • a computer readable storage medium may be a tangible device that holds and stores instructions for use by an instruction execution device, and may be a volatile storage medium or a nonvolatile storage medium.
  • the disclosed methods and devices may be implemented in other ways.
  • the device implementations described above are only illustrative.
  • the division of parts or units is only a logical function division.
  • units or components can be combined or integrated. to another system, or some features may be ignored, or not implemented.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, and the indirect coupling or communication connection of devices or units may be in electrical, mechanical or other forms.
  • a unit described as a separate component may or may not be physically separated, and a component shown as a unit may or may not be a physical unit, that is, it may be located in one place, or may also be distributed to network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, each unit may exist separately physically, or two or more units may be integrated into one unit.
  • the above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.
  • the integrated unit is realized in the form of a software function unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.
  • the technical solution of the embodiment of the present disclosure is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage
  • several instructions are included to make a computer device (which may be a personal computer, server, or network device, etc.) or a processor (processor) execute all or part of the steps of the methods in various embodiments of the present disclosure.
  • the aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disc, etc., which can store program codes. .
  • the embodiment of the present disclosure discloses an image feature point selection method, device, device, storage medium, and program product, wherein the image feature point selection method includes: obtaining several feature points in an image frame; based on several feature points, preconfiguring a fixed size The node storage space; the tree-shaped division method is used to divide several feature points into at least one node, and the first information of the node is stored in the node storage space, wherein the first information of the node includes the feature correlation of the feature point corresponding to the node Information; Select a feature point from each node obtained by the final division, so as to obtain a selection result of several feature points.
  • the execution efficiency of feature point selection can be improved.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Databases & Information Systems (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Library & Information Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)

Abstract

本公开公开了一种图像特征点选取方法及装置、设备、存储介质和程序产品，其中，方法包括：获取图像帧中的若干特征点；基于若干特征点，预配置固定大小的节点存放空间；采用树形划分方式将若干特征点划分到至少一个节点，并将节点的第一信息存储至节点存放空间中，其中，节点的第一信息包括节点对应的特征点的特征相关信息；从最终划分得到的每个节点中选取特征点，以得到若干特征点的选取结果。

Description

图像特征点选取方法及装置、设备、存储介质和程序产品
相关申请的交叉引用
本公开实施例基于申请号为202111395862.5、申请日为2021年11月23日、申请名称为“图像特征点选取方法及相关装置、设备和存储介质”的中国专利申请提出，并要求该中国专利申请的优先权，该中国专利申请的全部内容在此引入本公开实施例作为参考。
技术领域
本公开涉及图像处理技术领域，特别是涉及一种图像特征点选取方法及装置、设备、存储介质和程序产品。
背景技术
对图像进行特征提取是计算机视觉、机器人、无人车、三维重建及增强现实等领域的重要底层技术。考虑后续特征匹配等对特征点处理的效率，往往会对图像进行特征提取后的得到特征点进行选取，以减少特征点的数量。
然而，对特征点进行选取的过程往往是对特征点进行划分后再进行选择，而现有的特征点划分过程往往需要动态不断申请空间和释放空间，从而影响特征点选取的执行效率，严重限制了是计算机视觉、机器人等技术领域的发展。
因此，如何提高特征点选取的执行效率，具有非常重要的意义。
发明内容
本公开实施例提供一种图像特征点选取方法及装置、设备、存储介质和程序产品。
本公开实施例第一方面提供了一种图像特征点选取方法，包括：获取图像帧中的若干特征点；基于若干特征点，预配置固定大小的节点存放空间；采用树形划分方式将若干特征点划分到至少一个节点，并将节点的第一信息存储至节点存放空间中，其中，节点的第一信息包括节点对应的特征点的特征相关信息；从最终划分得到的每个节点中选取特征点，以得到若干特征点的选取结果。
因此，通过申请固定大小的节点存放空间，并将节点的第一信息存储至节点存放空间中，使得在采用树形划分方式对若干特征点划分进行划分的过程中，无需重复申请存放空间来来存放节点的第一信息，减少了针对存放空间的操作，以此可以提高特征点选取的执行效率。
本公开实施例第二方面提供了一种图像特征点选取方法，包括：获取图像帧的多个特征图，其中，每个特征图的分辨率不同；对每个特征图并行执行以下特征点选取：从特征图获取若干特征点，并获取若干特征点的选取结果，其中，若干特征点的选取结果是利用上述第一方面描述的方法从若干特征点中选取得到的。
由此，通过对不同分辨率的特征图并行执行以下特征点选取，使得执行本公开实施例图像特征点选取方法的设备可以同步对多个不同分辨率的特征图进行特征点选取，并获得特征点的选取结果，提高了针对多个不同分辨率的特征图进行特征点选取的整体执行效率。
本公开实施例第三方面提供了一种图像特征点选取装置，包括：获取部分、内存申请部分、特征点划分部分和特征点选取部分；获取部分配置为获取图像帧中的若干特征点；内存申请部分，配置为基于若干特征点，预配置固定大小的节点存放空间；特征点划分部分，配置为采用树形划分方式将若干特征点划分到至少一个节点，并将节点的第一信息存储至节点存放空间中，其中，节点的第一信息包括节点对应的特征点的特征相关信息；特征点选取部 分，配置为从最终划分得到的每个节点中选取特征点，以得到若干特征点的选取结果。
本公开实施例第四方面提供了一种图像特征点选取装置，包括：获取部分、特征点选取部分；获取部分，配置为获取图像帧的多个特征图，其中，每个特征图的分辨率不同；特征点选取部分，配置为对每个特征图并行执行以下特征点选取：从特征图获取若干特征点，利用上述第一方面描述的方法获取若干特征点的选取结果。
本公开实施例第五方面提供了一种电子设备，包括相互耦接的存储器和处理器，处理器配置为执行存储器中存储的程序指令，以实现上述第一方面和第二方面描述的图像特征点选取方法。
本公开实施例第六方面提供了一种计算机可读存储介质，其上存储有程序指令，程序指令被处理器执行时实现上述第一方面和第二方面描述的图像特征点选取方法。
上述方案，通过申请固定大小的节点存放空间，并将节点的第一信息存储至节点存放空间中，使得在采用树形划分方式对若干特征点划分进行划分的过程中，无需重复申请存放空间来来存放节点的第一信息，减少了针对存放空间的操作，以此可以提高特征点选取的执行效率。
本公开实施例第七方面提供了一种计算机程序产品，所述计算机程序产品包括存储了计算机程序的非瞬时性计算机可读存储介质，所述计算机程序被计算机读取并执行时实现上述方法。
应当理解的是，以上的一般描述和后文的细节描述仅是示例性和解释性的，而非限制本公开。
附图说明
此处的附图被并入说明书中并构成本说明书的一部分，这些附图示出了符合本公开的实施例，并与说明书一起用于说明本公开的技术方案。
图1是本公开图像特征点选取方法第一实施例的第一流程示意图；
图2是本公开像特征点选取方法中采用树形划分方式划分特征点的一个实施例；
图3是本公开图像特征点选取方法第一实施例的第二流程示意图；
图4是本公开图像特征点选取方法第一实施例的第三流程示意图；
图5是本公开图像特征点选取方法中调整特征存放空间中第二信息的存放位置的一示意图；
图6是本公开图像特征点选取方法中调整特征存放空间中第二信息的存放位置的另一示意图；
图7是本公开图像特征点选取方法第一实施例的第四流程示意图；
图8是本公开图像特征点选取方法第二实施例的流程示意图；
图9是本公开图像特征点选取装置一实施例的框架示意图；
图10是本公开图像特征点选取装置另一实施例的框架示意图；
图11是本公开电子设备一实施例的框架示意图；
图12为本公开计算机可读存储介质一实施例的框架示意图。
具体实施方式
下面结合说明书附图，对本公开实施例的方案进行详细说明。
以下描述中，为了说明而不是为了限定，提出了诸如特定系统结构、接口、技术之类的具体细节，以便透彻理解本公开。
本文中术语“系统”和“网络”在本文中常被可互换使用。本文中术语“和/或”，仅仅是一种描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况。另外，本文中字符“/”，一般表示前 后关联对象是一种“或”的关系。此外，本文中的“多”表示两个或者多于两个。
本公开实施例的图像特征点选取方法可以应用于计算机视觉、机器人、无人车、三维重建及增强现实等技术领域。用于执行本公开实施例图像特征点选取方法的设备可以是计算机、手机、平板电脑以及智能眼镜等电子设备。
请参阅图1，图1是本公开图像特征点选取方法第一实施例的第一流程示意图，该方法可以包括如下步骤：
步骤S11：获取图像帧中的若干特征点。
图像帧可以是诸如手机、平板电脑、智能眼镜等电子设备所拍摄到的图像，或是监控相机所拍摄到的图像，在此不做限定。其他场景可以以此类推，在此不再一一举例。
在一些实施例中，图像帧可以是原始图像去除边缘区域后的图像。例如，将原始图像的边缘一定数量的像素点去除后，得到的图像即为图像帧。
在获得图像帧以后，可以对图像帧进行特征提取，以此获得若干特征点。在一个实施方式中，得到的若干特征点存储在内存中，并且以数组的形式存储。特征提取算法例如是FAST(Features From Accelerated Segment Test，一种用于角点检测的算法)算法，SIFT(Scale-Invariant Feature Transform，尺度不变特征变换)算法，ORB(Oriented FAST and Rotated BRIEF，一种快速特征点提取和描述的算法)算法等等。在一个实施场景中，特征提取算法为ORB(Oriented FAST and Rotated BRIEF)算法。另外，在得到特征点以后，还可以得到与每个特征点对应的特征表示，特征表示例如是特征向量。
在一个实施例中，可以是由执行本公开实施例图像特征点选取方法的设备来进行图像采集并进行特征提取，以得到若干特征点。在另一个实施方式中，也可以是由执行本公开实施例图像特征点选取方法的设备直接获取由其他设备对图像帧进行特征提取得到的特征点。
在一个实施方式中，在获取图像帧中的若干特征点之后，可以为若干特征点申请特征存放空间，并将若干特征点的第二信息存储至特征存放空间中。特征存放空间可以内存的内存空间。因此，通过申请特征存放空间，可以确定特征点的存放位置，便于后续对特征点进行划分。
在一个实施方式中，特征存放空间在内存中可以是一个数组，定义为第一数组。第一数组中包含一定数量的元素。在一个实施方式中，第一数组中的每个元素用于存放一个特征点的第二信息。因此，通过确定特征存放空间的具体形式，会便于后续对特征点进行划分。
在一个实施方式中，特征点的第二信息包括以下至少之一：特征点在图像帧中的位置信息、特征点的响应度、特征点所属的特征图的标识。特征点在图像帧中的位置信息可以是特征点的像素坐标。特征点的响应度可以是根据特征提取算法确定的特定的值，例如由FAST特征提取算法提取特征点时确定的该特征点的响应值。特征点所属的特征图为图像帧对应的多个不同分辨率的特征图中的其中一个，特征点所属的特征图的标识可以是该特征点所属的特征图在多个不同分辨率的特征图中的编号。多个不同分辨率的特征图可以是基于图像帧的图像金字塔进行特征提取得到的，也可以是针对图像帧，进行不同的特征提取，得到多个不同分辨率的特征图。因此，通过确定特征点的第二信息，可以利用这些信息，用于后续对特征点进行划分。
步骤S12：基于若干特征点，预配置固定大小的节点存放空间。
在得到若干特征点以后，可以对这些特征点进行划分，以得到特征点的选取结果。对特征点进行划分，可以是采用树形划分方式将若干特征点划分到至少一个节点。因此，在执行对特征点进行划分的步骤前，可以基于若干特征点，预配置固定大小的节点存放空间，该固定大小的节点存放空间可以用于存储划分后的特征点的相关信息。在一个实施方式中，是在内存中申请固定大小的内存空间作为节点存放空间。树形划分方式可以根据需要进行选择。在一个实施方式中，节点存放空间为内存中的第二数组。
步骤S13：采用树形划分方式将若干特征点划分到至少一个节点，并将节点的第一信息存储至节点存放空间中。
在确定了节点存放空间以后，就可以采用树形划分方式将若干特征点划分到至少一个节 点中，节点的数量可以根据需要进行设置，此处不作限制。采用树形划分方式将若干特征点划分到至少一个节点，对于图像帧而言，即将图像帧分成若干个区域，每个区域对应一个节点，并确定每个区域包含的特征点，每个区域包含的特征点即为该区域对应的节点对应的特征点。由此，可以实现将若干特征点划分到至少一个节点中。在一些实施方式中，树形划分方式包括四叉树划分方式和八叉树划分方式中的至少一种。通过确定树形划分方式的具体方式，可以对特征点进行四叉树划分或者是八叉树划分。
在一个实施例中，若某些特征点的位置是若干个图像区域的边界，则可以根据需要将边界上的特征点划分至若干个图像区域中。
通过划分得到的节点后，可以将节点的第一信息存储至节点存放空间中。节点的第一信息可以包括节点对应的特征点的特征相关信息。节点对应的特征点的特征相关信息，可以理解为与该节点对应的特征点的相关信息，例如是特征点的在图像中的位置信息、特征点在内存中的存储位置信息、特征点的特征表示信息(如特征向量)等等。在一个实施方式中，节点对应的特征信息包括节点对应的特征点在特征存放空间的存放位置。例如，特征存放空间是数组时，节点对应的特征点在特征存放空间的存放位置即是特征点存储的元素在数组中(特征存放空间)的位置下标。
在一个实施方式中，节点存放空间为内存中的第二数组，第二数组中的每个元素用于存放一个节点的第一信息。在另一个实施方式中，节点存放空间为单向链表，单向链表中的每个结点用于存放一个节点的第一信息。因此，通过确定节点存放空间的具体存储形式，可以用于后续对特征点的划分。
在一个实施方式中，节点的第一信息还包括以下至少之一：节点在图像帧中对应的区域的位置信息、节点是否为预设节点、节点所指向的下一节点在节点存放空间中的存放位置信息、节点在树形划分方式得到的树形结构中所处的层标识。节点在图像帧中对应的区域的位置信息，可以是节点在图像帧中对应的区域的四个顶点的像素坐标。预设节点可以是叶节点。节点所指向的下一节点在节点存放空间中的存放位置信息，下一节点在节点存放空间中的存放位置信息可以是下一节点在第二数组的位置下标。因此，通过确定节点的第一信息包括的具体信息，可以根据这些具体信息用于后续对特征点的划分。
参阅图2，图2是本公开像特征点选取方法中采用树形划分方式划分特征点的一个实施例。在本实施例中，树形划分方式是四叉树划分方式。在对图像帧10进行划分时，第一次划分得到4个区域，分别是区域11、区域12、区域13和区域14，这4个区域的层标识为1。此时，可以对区域11进行划分，得到4个区域，分别是区域111、区域112、区域113、区域114。
步骤S14：从最终划分得到的每个节点中选取特征点，以得到若干特征点的选取结果。
经过划分后，可以得到一定数量的节点，节点对应的图像帧的区域中会包含一定数量的特征点，此时可以从最终划分得到的节点中选取特征点，以得到若干特征点的选取结果，即从最终划分得到的每个节点中，选择一定数量的特征点，以此作为特征点的选取结果。
在一个实施方式中，可以是分别从每个最终划分节点中选择第二数量个特征点，以得到若干特征点的选取结果。第二数量可以根据进行设置，此处不作限制。例如，可以在最终划分得到的每个节点中，从每个节点中选取1个特征点，从每个节点中选取的特征点即为若干特征点的选取结果。
因此，通过申请固定大小的节点存放空间，并将节点的第一信息存储至节点存放空间中，使得在采用树形划分方式对若干特征点划分进行划分时，无需重复地向内存申请内存空间来存放节点的第一信息，减少了针对内存的操作，以此可以提高特征点划分方法的执行效率。
请参阅图3，图3是本公开图像特征点选取方法第一实施例的第二流程示意图。本实施例是对上述步骤提及的“基于若干特征点，预配置固定大小的节点存放空间”进一步扩展，包括以下步骤S121和步骤S122。
步骤S121：基于目标选取特征点数，确定目标节点划分数量。
目标选取特征点数为需从若干特征点中选取的特征点数量，目标选取特征点数的具体数 量可以根据需要进行设置。目标节点划分数量为最终划分得到的节点的数量。在一个实施方式中，目标节点划分数量可以与目标选取特征点数相同。例如，需要从若干特征点中选取的特征点数量选取10个特征点作为目标选取特征点，则目标节点划分数量可以是10个节点。
步骤S122：为若干特征点申请大小与目标节点划分数量匹配的节点存放空间。
在确定最终划分得到的节点的数量后，在这些节点对应的图像帧的区域会包含从图像帧中提取得到的若干特征点，因此，可以若干特征点申请大小与目标节点划分数量匹配的节点存放空间，用于存放节点的第一信息存储。
在一个实施方式中，可以先确定一个节点需要存储的节点的第一信息在内存中的大小，然后在内存中申请大小与目标节点划分数量匹配的节点存放空间。在另一个实施方式中，可以在内存中申请中与若干特征点的第二信息在内存中的空间相等的内存空间，然后再根据每个节点存储的节点的第一信息在内存中的大小。
因此，通过申请大小与目标节点划分数量匹配的节点存放空间，可以使得节点存放空间能够存储全部最终划分得到的节点的第一信息，使得后续无需再申请新的存放空间，以此减少了针对存放空间的操作，以此可以提高特征点选取的执行效率。
请参阅图4，图4是本公开图像特征点选取方法第一实施例的第三流程示意图。本实施例是对上述步骤提及的“采用树形划分方式将若干特征点划分到至少一个节点，并将节点的第一信息存储至节点存放空间中”步骤的进一步扩展，包括：
步骤S131：将若干特征点的部分或全部特征点划分至一个节点，并将节点的第一信息存放在节点存放空间中。
在进行具体特征点的划分时，首选可以把需要划分的特征点都划入一个节点中。在实施中，可以将若干特征点的部分或全部特征点划分至一个节点。在一个实施方式中，在采用树形划分方式来对特征点进行划分时，会首先确定一个节点，该节点可以认为是根节点。此时，可以认为将全部特征点都划入至一个节点中，或者将部分特征点划分至根节点中。
在一个实施方式中，将图像帧中除预设边界区域以外的区域作为根区域。预设边界区域可以认为是图像帧的边缘区域，如距离图像边缘一定数量个像素点内的区域。然后，将根区域中的特征点划分至一个节点。通过将去除图像帧的边缘区域，可以去除边缘区域上的特征点，减少特征点的数量，加快图像特征点选取方法执行速度。
在一个实施方式中，在执行后续的依序对节点存放空间的每个节点执行划分步骤，直至节点存放空间不再满足第一要求的步骤时，是在检测到当前划分的节点所对应的特征数大于目标选取特征点数的情况下执行的。也即，在将若干特征点的部分或全部特征点划分至一个节点以后，是通过判断当前划分的节点所对应的特征数是否大于目标选取特征点数，来确定是否需要执行后续的依序对节点存放空间的每个节点执行划分步骤。可以理解的，如果当前划分的节点所对应的特征数没有大于目标选取特征点数，则意味着该节点对应的全部特征点，都可以直接被确定被选取的特征点了，因此也就不必再执行后续的依序划分的步骤。因此，通过判断当前划分的节点所对应的特征数是否大于目标选取特征点数，可以确定是否需要继续执行后续的依序对节点存放空间的每个节点执行划分步骤。
步骤S132：依序对节点存放空间的每个节点执行划分步骤，直至节点存放空间不再满足第一要求；其中，划分步骤包括：将节点确定为父节点，并将父节点对应的特征点划分到至少一个新的节点，将新的节点的第一信息存放在节点存放空间中。
在将特征点的部分或全部特征点划分至一个节点以后，就可以对节点进行依序划分，在划分后，将划分得到的节点对应的第一信息存放在节点存放空间中。以此，可以实现对特征点选取过程中的特征点的划分。
例如，当采用四叉树划分方式时，是先对根节点进行划分，得到四个节点，然后将这4个节点分别作为父节点，并依序对这4个父节点进行划分。划分完后，将新划分出的节点作为父节点，再次对这些父节点进行依序划分。依序划分，即依照一定的顺序进行划分。
在一个实施方式中，父节点划分得到的节点为该父节点的子节点，对于属于同一父节点的子节点，可以根据子节点对应的特征点数量来确定子节点的子节点执行顺序，后续便可根 据子节点执行顺序对子节点执行划分步骤，其中，该子节点为父节点划分得到的节点。也即，对于属于同一父节点的子节点，其执行划分步骤的顺序取决于子节点对应的特征点数。特征点数量越多的节点，其执行划分步骤的顺序越早，即按照节点对应的特征点数量的多少，决定执行顺序的先后。例如一个节点被划分为4个子节点，分别为A子节点、B子节点、C子节点和D子节点，节点对应的特征点数量大小关系为：A>B>C>D，则会先对A子节点执行划分步骤，然后是对B子节点执行划分步骤，以此类推。因此，通过优先对对应特征点数较多的子节点进行划分，可以使得对应特征点数量较多的节点优先被划分，可以提高划分特征点的均匀性。
在一个实施方式中，对于属于同一父节点的节点，其执行划分步骤的顺序也可以是取决于节点对应的特征点的最大响应度。最大响应度越大的节点，越先被执行。同样以上述的4个节点为例，节点对应的特征点的最大响应度的大小关系为：A>B>C>D，则会先对A节点执行划分步骤，然后是对B节点执行划分步骤，以此类推。在另一个实施方式中，对于属于同一父节点的节点，其执行划分步骤的顺序也可以是取决于节点对应的全部特征点的响应度之和，响应度之和越大的节点，越先被执行。还以上述的4个节点为例，节点对应的特征点的响应度之和的大小关系为：A>B>C>D，则会先对A节点执行划分步骤，然后是对B节点执行划分步骤，以此类推。
在一个实施方式中，可以是确定与当前执行的节点属于同一父节点的其他节点的执行顺序先于由当前执行的节点划分得到的节点的执行顺序。例如，已经划分得到4个节点，分别为A节点、B节点、C节点和D节点。此时，若将A节点确定为为父节点，并将父节点对应的特征点划分到至少一个新的节点后。会对B节点进行划分，然后是对C节点进行划分，以此类推。当A节点、B节点、C节点和D节点都被划分完以后，才会对新划分的节点(由A节点、B节点、C节点和D节点划分得到的节点)进行划分。以此，通过确定同属于一个父节点的节点优先被划分，提高了划分特征点的均匀性，进而提高后续特征选取的合理性。
可以理解地，当一个节点被划分后，意味着该节点不再存储与节点存放空间。节点存放空间中的当前存放节点数，是节点存放空间包含的未被划分的节点，因为被划分的节点已经不再存储与节点存放空间中。
在一个实施方式中，第一要求为节点存放空间中的当前存放节点数少于目标节点划分数量。此时，意味着当前存放节点数等于或大于目标节点数量，因此可以不再继续进行划分。因此，通过确定第一要求为节点存放空间中的当前存放节点数少于目标节点划分数量，可以在得到满足要求数量的节点以后，停止执行节点划分的步骤。
在一个实施方式中，本步骤提及的“将节点确定为父节点”，是在节点对应的特征点数量满足第二要求的情况下，才会将节点作为父节点。也即，当一个节点对应的特征点数量满足第二要求时，才会将该节点作为父节点，并将该节点对应的特征点划分到至少一个新的节点。第二要求例如是一个节点对应的特征点数量大于第一阈值时，才会将该节点作为父节点，例如第一阈值例如是3、4或5等，在实施中可以根据需要设置，此处不作限制，通过设置第一阈值，可以减少需要划分的节点的数量。因此，通过判断节点对应的特征点数量是否满足第二要求，可以将不满足第二要求的节点作为父节点，提高特征点划分的均匀性，而且也可减少了需要划分的节点数量，提高了图像特征点选取的执行效率。
在一个实施方式中，在每次将特征点划分到节点之后，可以调整特征存放空间中第二信息的存放位置，以使同一节点对应的特征点的第二信息存放在相邻位置上。在一个实施方式中，在将一个节点划分为四个新节点分别为A节点、B节点、C节点和D节点后，可以调整特征存放空间中第二信息的存放位置，使得对应于A节点的特征点的第二信息是存放在相邻位置上的。
参阅图5，图5是本公开图像特征点选取方法中调整特征存放空间中第二信息的存放位置的一示意图。请看图5的a部分，区域10是一节点Y对应的图像帧的某一区域。在区域10中，包含9个特征点，分别是特征点1至9，也即节点Y对应有9个特征点。这9个特征点的第二信息在特征存放空间20(内存中的数组)的存放位置如图5的a部分所示。此时， 会将节点Y确定父节点，并将节点Y划分为4个新节点，分别得到A节点、B节点、C节点和D节点。对应地，区域10也会被划分为4个新区域，分别是区域11、区域12、区域13和区域14。此时可以确定，A节点对应的特征点为特征点1和特征点5，B节点对应的特征点为特征点2和特征点6，C节点对应的特征点为特征点7和特征点9，D节点对应的特征点为特征点3、特征点4和特征点8。基于此，可以调整特征存放空间20中第二信息的存放位置，得到调整后特征存放空间21，这9个特征点的第二信息在特征存放空间21的存放位置如图5的b部分所示。可见，此时同一节点对应的特征点的第二信息存放在相邻位置上。
在一个实施方式中，可以设定两个指针来对特征存放空间中第二信息的存放位置进行调整。两个指针分别为指针甲和指针乙，利用这两个指针执行特征点的第二信息排序步骤。
第一，两个指针分别指向排序的起始位置。起始位置是被划分的父节点的对应的特征点的第二信息在特征存放空间第一个元素。
第二，在将父节点划分为不同的新节点以后，确定每个节点对应的图像帧区域，并判断指针甲指向的第二信息对应的特征点(后续简称为指针指向的特征点)是否在新节点对应的图像帧区域内。
第三，若在新节点对应的图像帧区域内，则交换指针甲和指针乙所指向的特征点(若指针甲和指针乙指向同一个特征点，则不用交换)。
第四，若指针甲指向的特征点不在新节点对应的图像帧区域内，则指针甲指向下一个特征点。
第五，重复执行上述第一至第四的步骤，直至指针甲超过结束位置。结束位置是被划分的父节点对应的特征点的第二信息在特征存放空间的最后一个元素。
以此，可以使得同一节点对应的特征点的第二信息存放在相邻位置上。
此时，如果需要继续调整另一个节点对应的特征点的第二信息存放在相邻位置上时，会调整起始位置，即将已经被调整为存放在相邻位置上的特征点的第二信息的所在的元素排除之后的第一个元素。
参阅图6，图6是本公开图像特征点选取方法中调整特征存放空间中第二信息的存放位置的另一示意图。结合查阅图5和图6，在图5中，节点Y被确定为父节点进行划分，得到A节点、B节点、C节点和D节点。在调整特征点的第二信息的存放位置时，可以先调整节点A对应的特征点的第二信息的存放位置。如图6的a部分所示，此时的起始位置为特征点1对应的第二信息在特征存放空间的位置，结束位置为特征点9对应的第二信息在特征存放空间的位置，指针甲和指针乙指向特征点1。执行上述的排序步骤，当指针甲指向的特征点超过特征点9以后，停止排序。此时特征存放空间中各元素的位置如图6的b部分所示。在继续调整B节点对应的特征点的第二信息的存放位置时，此时的起始位置为如图6的b部分中的特征点3对应的第二信息在特征存放空间的位置，结束位置依然是征点9对应的第二信息在特征存放空间的位置。
通过调整特征存放空间中第二信息的存放位置，可以使得同一节点对应的特征点的第二信息存放在相邻位置上。因为节点存放空间中存储的节点的第一信息，会包括节点对应的特征点的特征相关信息，此时节点对应的特征点的特征相关信息可以包括节点对应的特征点在特征存放空间中的起始存放位置和结束存放位置。例如，在图5中，节点D对应的特征点的特征相关信息就可以包括特征点3在特征存放空间中的起始存放位置，以及特征点8在在特征存放空间中的结束存放位置。以此，通过将节点对应的特征点的特征相关信息设置为可以包括节点对应的特征点在特征存放空间中的起始存放位置和结束存放位置，就可以存储该节点对应的全部的特征点的特征点的特征相关信息，而不必记录每一个特征点的特征信息，由此可以节省存放空间。
请参阅图7，图7是本公开图像特征点选取方法第一实施例的第四流程示意图。本实施例是对上述步骤提及的“父节点对应的特征点划分到至少一个新的节点”的进一步扩展，包括：
步骤S1321：将父节点对应的父区域划分为第一数量个子区域。
在确定父节点以后，可以将父节点对应的父区域划分为第一数量个子区域，第一数量例如是为4或8。在采用树形划分方式划分特征点时，也会将图像帧划分为若干个区域，以此可以确定某一区域对应的节点，并确定该节点对应的特征点。因此，父区域可以认为是父节点对应的特征点在图像帧中的区域。例如，在图5的b部分中，若将节点D确定为父节点，则节点D对应的特征点3、特征点4和特征点8在图像帧中区域14即为父区域。通过将每次划分子区域的数量限定为4或8，即相当于采用四叉树或八叉树方式实现树形划分。
步骤S1322：判断子区域的特征点数量是否满足第三要求。
在将父区域划分为第一数量个子区域，可以判断每个子区域包含的特征点的数量是否满足第三要求，以此来确定是否为该区域生成一个子节点。在一个实施方式中，第三要求为子区域中的特征点数量大于第二阈值。在一个实施方式中，第二阈值可以是0。因此，通过判断子区域的特征点数量是否大于第二阈值，可以将特征点数量少于第二阈值的区域排除。
步骤S1323：在子区域中的特征点数量满足第三要求的情况下，为子区域生成对应的子节点。
如果子区域中的特征点数量满足第三要求，意味着可以为该子区域生成对应的子节点。此时，子节点对应的特征点即为该子区域中的特征点。
步骤S1324：在所子区域的特征点数量不满足第三要求的情况下，不为子区域生成对应的子节点。
因此，通过将父节点对应的父区域划分为第一数量个子区域，并且判断子区域的特征点数量是否满足第三要求，可以将不满足要求的子区域排除从而可以不生成对应的子节点，能够针对子区域的特征数量灵活确定是否形成节点。
在一个公开实施例中，上述步骤S132中提及的“将新的节点的第一信息存放在节点存放空间中”可以是：将其中一新的节点的第一信息覆盖父节点的第一信息进行存放，剩余新的节点的第一信息存放在节点存放空间中的剩余存放空间中。
剩余存放空间可以理解为在节点存放空间中，还没有用于存储节点的第一信息的存放空间。可以理解的，因为父节点被划分后，可以认为该节点已经不再存在，因此，可以将划分得到的新节点中的其中一个新的节点的第一信息覆盖父节点的第一信息进行存放，然后将剩余新的节点的第一信息存放在节点存放空间中的剩余存放空间中。因此，通过利用一个新的节点的第一信息覆盖父节点的第一信息进行存放，实现对存储空间的重复利用，可以节省存放空间。
在一个公开实施例中，上述步骤S132提及的“将父节点对应的特征点划分到至少一个新的节点”之后，本公开实施例的图像特征点选取方法还可以包括步骤：将属于同一父节点的节点之间进行链路式指向，且使位于链路的链尾的节点指向于父节点指向的节点，使指向父节点的节点更新指向于位于链路的链头的节点；其中，节点的指向用于确定节点的执行顺序。
在一个实施方式中，可以通过在每个节点的第一信息中存储指向的下一节点在节点存放空间中的存放位置信息，作为节点之间进行链路式指向。
在一个例子中，节点A指向节点B，节点C指向节点A。将节点A划分为节点A1、节点A2、节点A3和节点A4以后。将属于同一父节点的节点之间进行链路式指向，且使位于链路的链尾的节点指向于父节点指向的节点，使指向父节点的节点更新指向于位于链路的链头的节点，可以是节点C指向节点A1，节点A1指向节点A2，节点A2指向节点A3，节点A3指向节点A4，节点A4指向节点B。
因此，通过链路指向方式可实现后续节点划分的顺序。
上述方案，通过申请固定大小的节点存放空间，并将节点的第一信息存储至节点存放空间中，使得在采用树形划分方式对若干特征点划分进行划分时，无需重复的向内存申请内存空间来存放节点的第一信息，减少了针对内存的操作，以此可以提高特征点划分方法的执行效率。
请参阅图8，图8是本公开实施例图像特征点选取方法第二实施例的流程示意图。本实 施例包括以下步骤：
步骤S21：获取图像帧的多个特征图，其中，每个特征图的分辨率不同。
多个不同分辨率的特征图可以是基于图像帧的图像金字塔进行特征提取得到的，也可以是针对一帧图像帧，进行不同的特征提取，得到多个不同分辨率的特征图。
步骤S22：对每个特征图并行执行以下特征点选取：从特征图获取若干特征点，并获取若干特征点的选取结果。
在本实施例中，获取每个特征图的若干特征点的选取结果是根据上述图像特征点选取方法第一实施例描述的技术方案进行特征点选取得到的。
由此，通过对不同分辨率的特征图并行执行以下特征点选取，使得执行本公开实施例图像特征点选取方法的设备可以同步对多个不同分辨率的特征图进行特征点选取，并获得特征点的选取结果，提高了针对多个不同分辨率的特征图进行特征点选取的整体执行效率。
在一个实施方式中，在针对多个不同分辨率的特征图并行执行图像特征点选取方法时，特征图包含的特征点总数越多，用于对特征图执行特征点选取的处理资源越多。可以理解的，特征图包含的特征点总数越多，需要划分的特征点的数量就越多，所需要的算力就越多，因此通过确定当特征图包含的特征点总数越多，用于对特征图执行特征点选取的处理资源越多，以此提高特征点选取的整体执行效率。
上述方案，通过对多个不同分辨率的特征图执行图像特征点选取方法，提高了整体执行效率。
本领域技术人员可以理解，在具体实施方式的上述方法中，各步骤的撰写顺序并不意味着严格的执行顺序而对实施过程构成任何限定，各步骤的具体执行顺序应当以其功能和可能的内在逻辑确定。
请参阅图9，图9是本公开图像特征点选取装置一实施例的框架示意图。图像特征点选取装置90包括获取部分91、内存申请部分92和特征点划分部分93以及特征点选取部分94。获取部分91配置为获取图像帧中的若干特征点；内存申请部分92配置为基于若干特征点，预配置固定大小的节点存放空间；特征点划分部分93配置为采用树形划分方式将若干特征点划分到至少一个节点，并将节点的第一信息存储至节点存放空间中，其中，节点的第一信息包括节点对应的特征点的特征相关信息；特征点选取部分94配置为从最终划分得到的每个节点中选取特征点，以得到若干特征点的选取结果。
其中，上述的内存申请部分92配置为基于若干特征点，预配置固定大小的节点存放空间，包括：基于目标选取特征点数，确定目标节点划分数量，其中，目标选取特征点数为需从若干特征点中选取的特征点数量，目标节点划分数量为最终划分得到的节点的数量；为若干特征点申请大小与目标节点划分数量匹配的节点存放空间。
其中，上述的特征点划分部分93配置为采用树形划分方式将若干特征点划分到至少一个节点，并将节点的第一信息存储至节点存放空间中，包括：将若干特征点的部分或全部特征点划分至一节点，并将节点的第一信息存放在节点存放空间中；依序对节点存放空间的每个节点执行划分步骤，直至节点存放空间不再满足第一要求；其中，划分步骤包括：将节点确定为父节点，并将父节点对应的特征点划分到至少一个新的节点，将新的节点的第一信息存放在节点存放空间中。
其中，上述方案中与当前执行的节点属于同一父节点的其他节点的执行顺序先于由当前执行的节点划分得到的节点的执行顺序；和/或，上述的依序对节点存放空间的每个节点执行划分步骤，包括：对于属于同一父节点的子节点，根据子节点对应的特征点数量确定子节点的子节点执行顺序，根据子节点执行顺序对子节点执行划分步骤，其中，子节点为父节点划分得到的节点；和/或，在特征点划分部分93将父节点对应的特征点划分到至少一个新的节点之后，特征点划分部分93还可以配置为将属于同一父节点的节点之间进行链路式指向，且使位于链路的链尾的节点指向于父节点指向的节点，使指向父节点的节点更新指向于位于链路的链头的节点；其中，节点的指向用于确定节点的执行顺序。
其中，上述的特征点划分部分93配置为将新的节点的第一信息存放在节点存放空间中， 包括：将其中一新的节点的第一信息覆盖父节点的第一信息进行存放，剩余新的节点的第一信息存放在节点存放空间中的剩余存放空间中。
其中，上述的特征点划分部分93配置为将节点确定为父节点，包括：在节点对应的特征点数量满足第二要求的情况下，将节点作为父节点；和/或，特征点划分部分93配置为将父节点对应的特征点划分到至少一个新的节点，包括：将父节点对应的父区域划分为第一数量个子区域，其中，父区域为父节点对应的特征点在图像帧中的区域；在子区域中的特征点数量满足第三要求的情况下，为子区域生成对应的子节点，其中子节点对应的特征点为子区域中的特征点；在所子区域的特征点数量不满足第三要求的情况下，不为子区域生成对应的子节点。
其中，上述方案中第一要求为节点存放空间中的当前存放节点数少于目标节点划分数量；第二要求为节点对应的特征点数量大于第一阈值；第三要求为子区域中的特征点数量大于第二阈值；第一数量为4或8。
其中，上述的特征点划分部分93配置为将若干特征点的部分或全部特征点划分至一节点，包括：将图像帧中除预设边界区域以外的区域作为根区域；将根区域中的特征点划分至一节点；和/或，依序对节点存放空间的每个节点执行划分步骤，直至节点存放空间不再满足第一要求的步骤，是在检测到当前划分的节点所对应的特征数大于目标选取特征点数的情况下执行的。
其中，装置90还包括特征点空间申请部分。在获取部分91配置为获取图像帧中的若干特征点之后，特征点空间申请部分配置为为若干特征点申请特征存放空间，并将若干特征点的第二信息存储至特征存放空间中；其中，节点对应的特征信息包括节点对应的特征点在特征存放空间的存放位置。
其中，上述方案中，节点对应的特征点的特征相关信息包括节点对应的特征点在特征存放空间中的起始存放位置和结束存放位置。上述的特征点划分部分93每次将特征点划分到节点之后，特征点划分部分93还配置为调整特征存放空间中第二信息的存放位置，以使同一节点对应的特征点的第二信息存放在相邻位置上。
其中，上述方案中，特征存放空间为第一数组，第一数组中的每个元素配置为存放一个特征点的第二信息；和/或，特征点的第二信息包括以下至少一者：特征点在图像帧中的位置信息、特征点的响应度、特征点所属的特征图的标识，其中，特征点所属的特征图为图像帧对应的多个不同分辨率的特征图中的其中一个。
其中，上述方案中，节点存放空间为第二数组，第二数组中的每个元素配置为存放一个节点的第一信息；或者，节点存放空间为单向链表，单向链表中的每个结点配置为存放一个节点的第一信息；和/或，节点的第一信息还包括以下至少一者：节点在图像帧中对应的区域的位置信息、节点是否为预设节点、节点所指向的下一节点在节点存放空间中的存放位置信息、节点在树形划分方式得到的树形结构中所处的层标识。
其中，上述方案中，树形划分方式包括四叉树划分方式和八叉树划分方式中的至少一种；和/或，特征点选取部分94配置为从最终划分得到的每个节点中选取特征点，以得到若干特征点的选取结果，包括：分别从每个最终划分节点中选择第二数量个特征点，以得到若干特征点的选取结果。
请参阅图10，图10是本公开图像特征点选取装置另一实施例的框架示意图。图像特征点选取装置10包括获取部分100和特征点选取部分110。获取部分100配置为获取图像帧的多个特征图，其中，每个特征图的分辨率不同；特征点选取部分110配置为对每个特征图并行执行以下特征点选取：从特征图获取若干特征点，并若干特征点的选取结果，其中，若干特征点的选取结果是利用上述的图像特征点选取方法实施例得到的。
其中，上述方案中的特征图包含的特征点总数越多，用于对特征图执行特征点选取的处理资源越多。
请参阅图11，图11是本公开电子设备一实施例的框架示意图。电子设备11包括相互耦接的存储器111和处理器112，处理器112配置为执行存储器111中存储的程序指令，以实 现上述任一图像特征点选取方法实施例的步骤。在一个实施场景中，电子设备11可以包括但不限于：微型计算机、服务器，此外，电子设备11还可以包括笔记本电脑、平板电脑等移动设备，在此不作限定。
具体而言，处理器112配置为控制其自身以及存储器111以实现上述任一图像特征点选取方法实施例的步骤。处理器112还可以称为CPU(Central Processing Unit，中央处理单元)。处理器112可能是一种集成电路芯片，具有信号的处理能力。处理器112还可以是通用处理器、数字信号处理器(Digital Signal Processor,DSP)、专用集成电路(Application Specific Integrated Circuit,ASIC)、现场可编程门阵列(Field-Programmable Gate Array,FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。另外，处理器112可以由集成电路芯片共同实现。
请参阅图12，图12为本公开计算机可读存储介质一实施例的框架示意图。计算机可读存储介质120存储有能够被处理器运行的程序指令121，程序指令121用于实现上述任一图像特征点选取方法实施例的步骤。
上述方案，能够提高特征点划分方法的执行效率。
在一些实施例中，本公开实施例实施例提供的装置具有的功能或包含的部分可以配置为执行上文方法实施例描述的方法，其具体实现可以参照上文方法实施例的描述。
计算机可读存储介质可以是保持和存储由指令执行设备使用的指令的有形设备，可以是易失性存储介质或非易失性存储介质。
上文对各个实施例的描述倾向于强调各个实施例之间的不同之处，其相同或相似之处可以互相参考。
在本公开所提供的几个实施例中，应该理解到，所揭露的方法和装置，可以通过其它的方式实现。例如，以上所描述的装置实施方式仅仅是示意性的，例如，部分或单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，装置或单元的间接耦合或通信连接，可以是电性、机械或其它的形式。
作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施方式方案的目的。
另外，在本公开各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。
集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本公开实施例的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)或处理器(processor)执行本公开各个实施方式方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
工业实用性
本公开实施例公开了一种图像特征点选取方法及装置、设备、存储介质和程序产品，其中图像特征点选取方法包括：获取图像帧中的若干特征点；基于若干特征点，预配置固定大小的节点存放空间；采用树形划分方式将若干特征点划分到至少一个节点，并将节点的第一信息存储至节点存放空间中，其中，节点的第一信息包括节点对应的特征点的特征相关信息； 从最终划分得到的每个节点中选取特征点，以得到若干特征点的选取结果。通过该方法，可以提高特征点选择的执行效率。

Claims (33)

  1. 一种图像特征点选取方法，包括：
    获取图像帧中的若干特征点；
    基于所述若干特征点，预配置固定大小的节点存放空间；
    采用树形划分方式将所述若干特征点划分到至少一个节点，并将所述节点的第一信息存储至所述节点存放空间中，其中，所述节点的第一信息包括所述节点对应的特征点的特征相关信息；
    从最终划分得到的每个节点中选取所述特征点，以得到所述若干特征点的选取结果。
  2. 根据权利要求1所述的方法，其中，所述基于所述若干特征点，预配置固定大小的节点存放空间，包括：
    基于目标选取特征点数，确定目标节点划分数量，其中，所述目标选取特征点数为需从所述若干特征点中选取的特征点数量，所述目标节点划分数量为最终划分得到的节点的数量；
    为所述若干特征点申请大小与所述目标节点划分数量匹配的节点存放空间。
  3. 根据权利要求1或2所述的方法，其中，所述采用树形划分方式将所述若干特征点划分到至少一个节点，并将所述节点的第一信息存储至所述节点存放空间中，包括：
    将所述若干特征点的部分或全部特征点划分至一个节点，并将所述节点的第一信息存放在节点存放空间中；
    依序对所述节点存放空间的每个所述节点执行划分步骤，直至所述节点存放空间不再满足第一要求；其中，所述划分步骤包括：将所述节点确定为父节点，并将所述父节点对应的特征点划分到至少一个新的节点，将所述新的节点的第一信息存放在所述节点存放空间中。
  4. 根据权利要求3所述的方法，其中，与当前执行的所述节点属于同一父节点的其他节点的执行顺序先于由当前执行的节点划分得到的节点的执行顺序；
    和/或，所述依序对所述节点存放空间的每个所述节点执行划分步骤，包括：对于属于同一所述父节点的子节点，根据所述子节点对应的特征点数量确定所述子节点的子节点执行顺序，根据所述子节点执行顺序对所述子节点执行划分步骤，其中，所述子节点为所述父节点划分得到的节点；
    和/或，在所述将所述父节点对应的特征点划分到至少一个新的节点之后，所述方法还包括：
    将属于同一所述父节点的节点之间进行链路式指向，且使位于所述链路的链尾的节点指向于所述父节点指向的节点，使指向所述父节点的节点更新指向于位于所述链路的链头的节点；其中，所述节点的指向用于确定所述节点的执行顺序。
  5. 根据权利要求3或4所述的方法，其中，所述将所述新的节点的第一信息存放在节点存放空间中，包括：
    将其中一所述新的节点的所述第一信息覆盖所述父节点的第一信息进行存放，剩余所述新的节点的所述第一信息存放在所述节点存放空间中的剩余存放空间中。
  6. 根据权利要求3至5任一项所述的方法，其中，所述将所述节点确定为父节点，包括：
    在所述节点对应的特征点数量满足第二要求的情况下，将所述节点作为父节点；
    和/或，所述将所述父节点对应的特征点划分到至少一个新的节点，包括：
    将所述父节点对应的父区域划分为第一数量个子区域，其中，所述父区域为所述父节点对应的特征点在图像帧中的区域；
    在所述子区域中的特征点数量满足第三要求的情况下，为所述子区域生成对应的子节点，其中所述子节点对应的特征点为所述子区域中的特征点；
    在所子区域的特征点数量不满足第三要求的情况下，不为所述子区域生成对应的子节点。
  7. 根据权利要求6所述的方法，其中，所述第一要求为所述节点存放空间中的当前存放节点数少于所述目标节点划分数量；
    所述第二要求为所述节点对应的特征点数量大于第一阈值；
    所述第三要求为所述子区域中的特征点数量大于第二阈值。
  8. 根据权利要求3至7任一项所述的方法，其中，所述将所述若干特征点的部分或全部特征点划分至一个节点，包括：
    将所述图像帧中除预设边界区域以外的区域作为根区域；
    将所述根区域中的所述特征点划分至所述一个节点；
    和/或，所述依序对所述节点存放空间的每个所述节点执行划分步骤，直至所述节点存放空间不再满足第一要求的步骤，是在检测到当前划分的节点所对应的特征数大于所述目标选取特征点数的情况下执行的。
  9. 根据权利要求1至8任一项所述的方法，其中，在所述获取图像帧中的若干特征点之后，所述方法还包括：
    为所述若干特征点申请特征存放空间，并将所述若干特征点的第二信息存储至特征存放空间中；其中，所述节点对应的特征信息包括所述节点对应的特征点在所述特征存放空间的存放位置。
  10. 根据权利要求9所述的方法，其中，所述节点对应的特征点的特征相关信息包括所述节点对应的特征点在所述特征存放空间中的起始存放位置和结束存放位置，所述方法还包括：
    每次将所述特征点划分到所述节点之后，调整所述特征存放空间中所述第二信息的存放位置，以使同一所述节点对应的特征点的第二信息存放在相邻位置上。
  11. 根据权利要求9或10所述的方法，其中，所述特征存放空间为第一数组，所述第一数组中的每个元素用于存放一个特征点的第二信息；
    和/或，所述特征点的第二信息包括以下至少一者：所述特征点在所述图像帧中的位置信息、所述特征点的响应度、所述特征点所属的特征图的标识，其中，所述特征点所属的特征图为所述图像帧对应的多个不同分辨率的特征图中的其中一个。
  12. 根据权利要求1至11任一项所述的方法，其中，所述节点存放空间为第二数组，所述第二数组中的每个元素用于存放一个节点的第一信息；或者，所述节点存放空间为单向链表，所述单向链表中的每个结点用于存放一个节点的第一信息；
    和/或，所述节点的第一信息还包括以下至少一者：所述节点在所述图像帧中对应的区域的位置信息、所述节点是否为预设节点、所述节点所指向的下一节点在所述节点存放空间中的存放位置信息、所述节点在所述树形划分方式得到的树形结构中所处的层标识。
  13. 根据权利要求1至12任一项所述的方法，其中，所述树形划分方式包括四叉树划分方式和八叉树划分方式中的至少一种；
    和/或，所述从最终划分得到的每个节点中选取所述特征点，以得到所述若干特征点的选取结果，包括：
    分别从每个所述最终划分节点中选择第二数量个所述特征点，以得到所述若干特征点的选取结果。
  14. 一种图像特征点选取方法，包括：
    获取图像帧的多个特征图，其中，每个所述特征图的分辨率不同；
    对每个所述特征图并行执行以下特征点选取：从所述特征图获取若干特征点，通过权利要求1至13中任一项所述的方法获取所述若干特征点的选取结果。
  15. 根据权利要求14所述的方法，其中，所述特征图包含的特征点总数越多，用于对所述特征图执行所述特征点选取的处理资源越多。
  16. 一种图像特征点选取装置，包括：
    获取部分，配置为获取图像帧中的若干特征点；
    内存申请部分，配置为基于所述若干特征点，预配置固定大小的节点存放空间；
    特征点划分部分，配置为采用树形划分方式将所述若干特征点划分到至少一个节点，并将所述节点的第一信息存储至所述节点存放空间中，其中，所述节点的第一信息包括所述节点对应的特征点的特征相关信息；
    特征点选取部分，配置为从最终划分得到的每个节点中选取所述特征点，以得到所述若干特征点的选取结果。
  17. 根据权利要求16所述的装置，其中，所述内存申请部分配置为基于所述若干特征点，预配置固定大小的节点存放空间，包括：基于目标选取特征点数，确定目标节点划分数量，其中，所述目标选取特征点数为需从所述若干特征点中选取的特征点数量，所述目标节点划分数量为最终划分得到的节点的数量；为所述若干特征点申请大小与所述目标节点划分数量匹配的节点存放空间。
  18. 根据权利要求16或17所述的装置，其中，所述特征点划分部分配置为所述采用树形划分方式将所述若干特征点划分到至少一个节点，并将所述节点的第一信息存储至所述节点存放空间中，包括：将所述若干特征点的部分或全部特征点划分至一个节点，并将所述节点的第一信息存放在节点存放空间中；依序对所述节点存放空间的每个所述节点执行划分步骤，直至所述节点存放空间不再满足第一要求；其中，所述划分步骤包括：将所述节点确定为父节点，并将所述父节点对应的特征点划分到至少一个新的节点，将所述新的节点的第一信息存放在节点存放空间中。
  19. 根据权利要求18所述的装置，其中，与当前执行的所述节点属于同一父节点的其他节点的执行顺序先于由当前执行的节点划分得到的节点的执行顺序；和/或，所述依序对所述节点存放空间的每个所述节点执行划分步骤，包括：对于属于同一所述父节点的子节点，根据所述子节点对应的特征点数量确定所述子节点的子节点执行顺序，根据所述子节点执行顺序对所述子节点执行划分步骤，其中，所述子节点为所述父节点划分得到的节点；和/或，在所述将所述父节点对应的特征点划分到至少一个新的节点之后，所述特征点划分部分还配置为将属于同一所述父节点的节点之间进行链路式指向，且使位于所述链路的链尾的节点指向于所述父节点指向的节点，使指向所述父节点的节点更新指向于位于所述链路的链头的节点；其中，所述节点的指向用于确定所述节点的执行顺序。
  20. 根据权利要求18或19所述的装置，其中，所述特征点划分部分配置为将新的节点的第一信息存放在节点存放空间中，包括：将其中一所述新的节点的所述第一信息覆盖所述父节点的第一信息进行存放，剩余所述新的节点的所述第一信息存放在所述节点存放空间中的剩余存放空间中。
  21. 根据权利要求18至20任一项所述的装置，其中，所述特征点划分部分配置为将所述节点确定为父节点，包括：在所述节点对应的特征点数量满足第二要求的情况下，将所述节点作为父节点；和/或，所述将所述父节点对应的特征点划分到至少一个新的节点，包括：将所述父节点对应的父区域划分为第一数量个子区域，其中，所述父区域为所述父节点对应的特征点在图像帧中的区域；在所述子区域中的特征点数量满足第三要求的情况下，为所述子区域生成对应的子节点，其中所述子节点对应的特征点为所述子区域中的特征点；在所子区域的特征点数量不满足第三要求的情况下，不为所述子区域生成对应的子节点。
  22. 根据权利要求21所述的装置，其中所述第一要求为所述节点存放空间中的当前存放节点数少于所述目标节点划分数量；所述第二要求为所述节点对应的特征点数量大于第一阈值；所述第三要求为所述子区域中的特征点数量大于第二阈值。
  23. 根据权利要求18至22任一项所述的装置，其中，所述特征点划分部分配置为将所述若干特征点的部分或全部特征点划分至一个节点，包括：将所述图像帧中除预设边界区域以外的区域作为根区域；将所述根区域中的所述特征点划分至所述一个节点；和/或，所述依序对所述节点存放空间的每个所述节点执行划分步骤，直至所述节点存放空间不再满足第一要求的步骤，是在检测到当前划分的节点所对应的特征数大于所述目标选取特征点数的情况 下执行的。
  24. 根据权利要求16至23任一项所述的装置，其中，所述装置还包括特征点空间申请部分，在所述获取部分配置为获取图像帧中的若干特征点之后，所述特征点空间申请部分配置为为所述若干特征点申请特征存放空间，并将所述若干特征点的第二信息存储至特征存放空间中；其中，所述节点对应的特征信息包括所述节点对应的特征点在所述特征存放空间的存放位置。
  25. 根据权利要求24所述的装置，其中，所述节点对应的特征点的特征相关信息包括所述节点对应的特征点在所述特征存放空间中的起始存放位置和结束存放位置，所述特征点划分部分还配置为每次将所述特征点划分到所述节点之后，调整所述特征存放空间中所述第二信息的存放位置，以使同一所述节点对应的特征点的第二信息存放在相邻位置上。
  26. 根据权利要求24或25所述的装置，其中，所述特征存放空间为第一数组，所述第一数组中的每个元素用于存放一个特征点的第二信息；和/或，所述特征点的第二信息包括以下至少一者：所述特征点在所述图像帧中的位置信息、所述特征点的响应度、所述特征点所属的特征图的标识，其中，所述特征点所属的特征图为所述图像帧对应的多个不同分辨率的特征图中的其中一个。
  27. 根据权利要求16至26任一项所述的装置，其中，所述节点存放空间为第二数组，所述第二数组中的每个元素用于存放一个节点的第一信息；或者，所述节点存放空间为单向链表，所述单向链表中的每个结点用于存放一个节点的第一信息；和/或，所述节点的第一信息还包括以下至少一者：所述节点在所述图像帧中对应的区域的位置信息、所述节点是否为预设节点、所述节点所指向的下一节点在所述节点存放空间中的存放位置信息、所述节点在所述树形划分方式得到的树形结构中所处的层标识。
  28. 根据权利要求16至27任一项所述的装置，其中，所述树形划分方式包括四叉树划分方式和八叉树划分方式中的至少一种；和/或，所述特征点选取部分配置为从最终划分得到的每个节点中选取所述特征点，以得到所述若干特征点的选取结果，包括：分别从每个所述最终划分节点中选择第二数量个所述特征点，以得到所述若干特征点的选取结果。
  29. 一种图像特征点选取装置，包括：
    获取部分，配置为获取图像帧的多个特征图，其中，每个所述特征图的分辨率不同；
    特征点选取部分，配置为对每个所述特征图并行执行以下特征点选取：从所述特征图获取若干特征点，通过权利要求1至13中任一项所述的方法获取所述若干特征点的选取结果。
  30. 根据权利要29所述的装置，其中，所述特征图包含的特征点总数越多，用于对所述特征图执行所述特征点选取的处理资源越多。
  31. 一种图像特征点选取设备，包括相互耦接的处理器和存储器，其中，
    所述处理器配置为执行所述存储器存储的计算机程序以执行权利要求1至13任一项所述的方法，或者是权利要求14至15任一项所述的方法。
  32. 一种计算机可读存储介质，存储有能够被处理器运行的计算机程序，所述计算机程序用于实现如执行权利要求1至13任一项所述的方法，或者是权利要求14至15任一项所述的方法。
  33. 一种计算机程序产品，所述计算机程序产品包括存储了计算机程序的非瞬时性计算机可读存储介质，所述计算机程序被计算机读取并执行时，所述计算机程序被计算机读取并执行时，实现权利要求1至13中任一项所述的方法，或者实现权利要求14至15任一项所述的方法。
PCT/CN2022/098194 2021-11-23 2022-06-10 图像特征点选取方法及装置、设备、存储介质和程序产品 WO2023092997A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202111395862.5A CN114022721A (zh) 2021-11-23 2021-11-23 图像特征点选取方法及相关装置、设备和存储介质
CN202111395862.5 2021-11-23

Publications (1)

Publication Number Publication Date
WO2023092997A1 true WO2023092997A1 (zh) 2023-06-01

Family

ID=80065827

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/098194 WO2023092997A1 (zh) 2021-11-23 2022-06-10 图像特征点选取方法及装置、设备、存储介质和程序产品

Country Status (2)

Country Link
CN (1) CN114022721A (zh)
WO (1) WO2023092997A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114022721A (zh) * 2021-11-23 2022-02-08 浙江商汤科技开发有限公司 图像特征点选取方法及相关装置、设备和存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104794219A (zh) * 2015-04-28 2015-07-22 杭州电子科技大学 一种基于地理位置信息的场景检索方法
CN105335377A (zh) * 2014-06-25 2016-02-17 联想(北京)有限公司 信息处理方法和设备
CN109144712A (zh) * 2017-06-19 2019-01-04 北京信威通信技术股份有限公司 内存池构建、内存分配方法及装置
CN110413716A (zh) * 2019-07-17 2019-11-05 北京三快在线科技有限公司 数据存储和数据查询方法、装置及电子设备
CN112017197A (zh) * 2020-09-07 2020-12-01 山东交通学院 一种图像特征提取方法及系统
CN114022721A (zh) * 2021-11-23 2022-02-08 浙江商汤科技开发有限公司 图像特征点选取方法及相关装置、设备和存储介质

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105335377A (zh) * 2014-06-25 2016-02-17 联想(北京)有限公司 信息处理方法和设备
CN104794219A (zh) * 2015-04-28 2015-07-22 杭州电子科技大学 一种基于地理位置信息的场景检索方法
CN109144712A (zh) * 2017-06-19 2019-01-04 北京信威通信技术股份有限公司 内存池构建、内存分配方法及装置
CN110413716A (zh) * 2019-07-17 2019-11-05 北京三快在线科技有限公司 数据存储和数据查询方法、装置及电子设备
CN112017197A (zh) * 2020-09-07 2020-12-01 山东交通学院 一种图像特征提取方法及系统
CN114022721A (zh) * 2021-11-23 2022-02-08 浙江商汤科技开发有限公司 图像特征点选取方法及相关装置、设备和存储介质

Also Published As

Publication number Publication date
CN114022721A (zh) 2022-02-08

Similar Documents

Publication Publication Date Title
US10803554B2 (en) Image processing method and device
JP6902611B2 (ja) 物体検出方法、ニューラルネットワークの訓練方法、装置および電子機器
CN107633526B (zh) 一种图像跟踪点获取方法及设备、存储介质
US20180260664A1 (en) Deep-learning network architecture for object detection
WO2019042426A1 (zh) 增强现实场景的处理方法、设备及计算机存储介质
US20150029222A1 (en) Dynamically configuring an image processing function
US10282059B2 (en) Graphical object appearance-invariant signature
CN109919971B (zh) 图像处理方法、装置、电子设备及计算机可读存储介质
CN112435223B (zh) 目标检测方法、装置及存储介质
WO2023092997A1 (zh) 图像特征点选取方法及装置、设备、存储介质和程序产品
JP5430636B2 (ja) データ取得装置、方法及びプログラム
CN109871249B (zh) 一种远程桌面操作方法、装置、可读存储介质及终端设备
KR20230049969A (ko) 글로벌 측위 장치 및 방법
CN114689036A (zh) 地图更新方法、自动驾驶方法、电子设备及存储介质
CN110633733A (zh) 图像智能匹配方法、装置及计算机可读存储介质
EP3410389A1 (en) Image processing method and device
CN116051736A (zh) 一种三维重建方法、装置、边缘设备和存储介质
Li et al. Multitarget tracking of pedestrians in video sequences based on particle filters
KR102224218B1 (ko) 비디오 시간 정보를 활용하는 딥러닝 기반 물체 검출 방법 및 장치
KR102382883B1 (ko) 3차원 손 자세 인식 장치 및 방법
CN112750164A (zh) 轻量化定位模型的构建方法、定位方法、电子设备
Lin et al. 6D object pose estimation with pairwise compatible geometric features
CN112767484B (zh) 定位模型的融合方法、定位方法、电子装置
US20230071291A1 (en) System and method for a precise semantic segmentation
US20240070979A1 (en) Method and apparatus for generating 3d spatial information

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22897095

Country of ref document: EP

Kind code of ref document: A1