Target detection method, device and equipment and storage medium thereof

Abstract

The embodiment of the invention belongs to the technical field of image processing, is applied to an object three-dimensional modeling reconstruction scene, and relates to a target detection method, device and equipment and a storage medium thereof. Estimating boundary points and non-boundary points; carrying out regionalization segmentation to obtain segmented regionalization point clouds; identifying boundary points in the three-dimensional point clouds included in all the regionalized point clouds; an integrated learning algorithm layer is adopted to extract fusion features of all regional point clouds; carrying out object surface modeling; identifying all target classification objects included in the three-dimensional point cloud to be analyzed and appearance forms of all the target classification objects; and adding the content as description text content into the object surface modeling result to finish the target detection. When the target detection method is applied to human body medical image scanning and focus form recognition scenes, medical staff can be assisted in visually performing medical image analysis, and the working efficiency of the medical staff is improved.

Description

Technical Field

[0001] The present application relates to the field of image processing technology, and is applied to three-dimensional modeling and reconstruction of objects, and in particular to a target detection method, device, equipment and storage medium thereof. Background Art

[0002] In the field of computer vision and machine learning, current 3D point cloud processing methods mainly include deep learning methods based directly on the original points, extracting features directly on the original point cloud, or using graph neural networks (GCN) for modeling, representing the point cloud as a graph, and strengthening the learning of the relationship between neighboring points in the point cloud through the point-edge relationship in the graph.

[0003] However, the above two processing methods directly extract features from the original point cloud, which is more troublesome when there are more point clouds. The graph neural network method usually requires modeling each point in the point cloud, which has huge resource overhead. Since the geometric features between some points are mostly similar, the efficiency of mapping each point is low and there are a lot of redundant calculations. Therefore, the above two processing methods are used to process the point cloud corresponding to the medical image. Due to the large computational overhead, the feature analysis of the point cloud is too time-consuming, resulting in a less than ideal improvement in the work efficiency of medical staff. Summary of the invention

[0004] The purpose of the embodiments of the present application is to propose a target detection method, device, equipment and storage medium thereof to solve the problem that when processing the point cloud corresponding to the medical image, the computational overhead is large and the feature analysis of the point cloud is too time-consuming, resulting in an unsatisfactory improvement in the work efficiency of medical personnel.

[0005] In order to solve the above technical problems, the embodiment of the present application provides a target detection method, which adopts the following technical solution:

[0006] A target detection method comprises the following steps:

[0007] Obtain the three-dimensional point cloud to be analyzed;

[0008] Using a three-dimensional space coordinate system to estimate the type of points in the three-dimensional point cloud, and estimating boundary points and non-boundary points as estimation results;

[0009] Using a three-dimensional point cloud segmentation technology to perform regional segmentation on the three-dimensional point cloud to obtain a segmented regional point cloud;

[0010] Identifying boundary points in the three-dimensional point cloud contained in all regionalized point clouds according to the estimation result;

[0011] Based on the boundary points in the three-dimensional point cloud contained in all the regionalized point clouds, an integrated learning algorithm layer is used to extract fusion features of all the regionalized point clouds;

[0012] The object surface is modeled based on the fusion features of all regionalized point clouds to obtain the object surface modeling result;

[0013] According to the object surface modeling result and the preset object classification model, identifying all target classification objects contained in the three-dimensional point cloud to be analyzed and the appearance of all target classification objects;

[0014] The name information of all target classified objects and the appearance of all target classified objects are added as descriptive text content to the object surface modeling result to complete this target detection.

[0015] Furthermore, the step of obtaining the three-dimensional point cloud to be analyzed specifically includes:

[0016] Use a preset 3D scanner to perform 3D scanning on the target object;

[0017] Based on a preset three-dimensional space coordinate system, obtaining a three-dimensional scanning cloud point set output by the three-dimensional scanner;

[0018] The three-dimensional scanning cloud point set is used as the three-dimensional point cloud to be analyzed.

[0019] Furthermore, the step of estimating the type of points in the three-dimensional point cloud by using a three-dimensional space coordinate system and estimating boundary points and non-boundary points as estimation results specifically includes:

[0020] Selecting any point from the three-dimensional point cloud as a target point;

[0021] According to the spatial coordinate information of all points in the three-dimensional point cloud in the three-dimensional spatial coordinate system, the spatial vector values ​​of other points relative to the target point are calculated one by one to obtain a set of spatial vector values;

[0022] By comparing the elements in the space vector value set in pairs, boundary points and non-boundary points in the three-dimensional point cloud are screened out.

[0023] Furthermore, the step of screening out boundary points and non-boundary points in the three-dimensional point cloud by comparing the elements in the space vector value set in pairs specifically includes:

[0024] Step 501, determining whether two elements in the space vector value set currently being compared are coplanar and co-directional vectors, wherein the coplanar and co-directional vectors indicate that the two space vector values ​​are coplanar and have the same direction in the three-dimensional space coordinate system;

[0025] Step 502: If two elements in the space vector value set currently being compared are coplanar and co-directional vectors, then mark the point corresponding to the element with the smaller space vector value as a non-boundary point, and continue to execute step 503;

[0026] Step 503, reselect two elements from the spatial vector value set, and continue to perform steps 501 to 502 until all elements in the spatial vector value set are compared two by two, and all marked points, i.e., non-boundary points in the three-dimensional point cloud, are obtained;

[0027] Step 504: Obtain all unmarked points, ie, boundary points in the three-dimensional point cloud.

[0028] Furthermore, before executing the step of regionalizing the three-dimensional point cloud using the three-dimensional point cloud segmentation technology to obtain the segmented regionalized point cloud, the method further includes:

[0029] Using a K-means algorithm to perform clustering processing on non-boundary points in the three-dimensional point cloud to obtain a clustering processing result, wherein the K value in the K-means algorithm represents the center point data in the three-dimensional point cloud and is set to a fixed value of 1;

[0030] According to the clustering processing result, the center point of all non-boundary points is identified as the center point of the three-dimensional point cloud.

[0031] Furthermore, the step of using the three-dimensional point cloud segmentation technology to perform regional segmentation on the three-dimensional point cloud to obtain the segmented regional point cloud includes:

[0032] The center point is set as the spatial coordinate origin in the three-dimensional spatial coordinate system, and a spatial reference coordinate system having an X-axis, a Y-axis and a Z-axis is divided by the spatial coordinate origin;

[0033] According to the spatial reference coordinate system and the preset segmentation plane, the three-dimensional point cloud is segmented to obtain a segmentation result;

[0034] A local map is generated for each local area according to the segmentation processing result to obtain local maps corresponding to all boundary points.

[0035] Furthermore, the step of regionalizing the three-dimensional point cloud using the three-dimensional point cloud segmentation technology to obtain the segmented regionalized point cloud further includes:

[0036] Using the KNN algorithm, all boundary points are divided into local areas to obtain local area division results, wherein the K value in the KNN algorithm represents the number of boundary points in each local area, and K is a positive integer greater than 1;

[0037] A local map is generated for each local area according to the local area division result and the center point of the three-dimensional point cloud, so as to obtain local maps corresponding to all boundary points.

[0038] Furthermore, the integrated learning algorithm layer includes a first feature extraction sublayer, a second feature extraction sublayer and a third feature extraction sublayer, and the step of extracting fusion features of all regionalized point clouds using the integrated learning algorithm layer based on the boundary points in the three-dimensional point clouds contained in all regionalized point clouds specifically includes:

[0039] According to the first feature extraction sublayer, position features are extracted for edge points in all local images respectively, and edge features corresponding to all local images are obtained as first feature extraction results;

[0040] According to the second feature extraction sublayer, position feature extraction is performed on non-edge points in all local images respectively, and non-edge features corresponding to all local images are obtained as second feature extraction results;

[0041] According to the third feature extraction sublayer, respectively extracting position features of points in all local images relative to the center point of the three-dimensional point cloud as third feature extraction results;

[0042] Performing feature fusion on the first feature extraction result, the second feature extraction result and the third feature extraction result corresponding to the same local image, and obtaining fused feature information as the fused feature corresponding to the corresponding local image;

[0043] According to the correspondence between the local graph and the regionalized point cloud, and the fusion features corresponding to each local graph, the fusion features corresponding to all the regionalized point clouds are obtained.

[0044] In order to solve the above technical problems, the embodiment of the present application also provides a target detection device, which adopts the following technical solution:

[0045] A target detection device, comprising:

[0046] A three-dimensional point cloud acquisition module is used to acquire the three-dimensional point cloud to be analyzed;

[0047] A point type estimation module, used to estimate the type of points in the three-dimensional point cloud by using a three-dimensional space coordinate system, and estimate boundary points and non-boundary points as estimation results;

[0048] A point cloud regional segmentation module is used to perform regional segmentation on the three-dimensional point cloud using a three-dimensional point cloud segmentation technology to obtain a segmented regional point cloud;

[0049] A boundary point recognition module, used to recognize boundary points in the three-dimensional point cloud contained in all regionalized point clouds according to the estimation result;

[0050] A fusion feature extraction module, used to extract fusion features of all regionalized point clouds using an integrated learning algorithm layer based on boundary points in the three-dimensional point clouds contained in all regionalized point clouds;

[0051] The object surface modeling module is used to perform object surface modeling based on the fusion features of all regionalized point clouds to obtain the object surface modeling results;

[0052] An object classification and recognition module, used to recognize all target classification objects and the appearance of all target classification objects contained in the three-dimensional point cloud to be analyzed according to the object surface modeling result and a preset object classification model;

[0053] The description text content adding module is used to add the name information of all target classification objects and the appearance of all target classification objects as description text content to the object surface modeling result to complete this target detection.

[0054] In order to solve the above technical problems, the embodiment of the present application further provides a computer device, which adopts the following technical solution:

[0055] A computer device includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the target detection method described above when executing the computer-readable instructions.

[0056] In order to solve the above technical problems, the embodiment of the present application further provides a computer-readable storage medium, which adopts the following technical solution:

[0057] A computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the target detection method as described above.

[0058] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0059] The target detection method described in the embodiment of the present application is to obtain the three-dimensional point cloud to be analyzed; estimate the boundary points and non-boundary points; regionalize and segment to obtain the segmented regionalized point cloud; identify the boundary points in the three-dimensional point cloud contained in all regionalized point clouds; use the integrated learning algorithm layer to extract the fusion features of all regionalized point clouds; perform object surface modeling; identify all target classification objects contained in the three-dimensional point cloud to be analyzed and the appearance of all target classification objects; add the descriptive text content to the object surface modeling result to complete this target detection. Applying the target detection method to the human medical image scanning and lesion morphology recognition scenarios can assist medical personnel in intuitively performing medical image analysis and improve the work efficiency of medical personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the scheme in the present application, a brief introduction is given below to the drawings required for use in the description of the embodiments of the present application. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0061] Figure 1 is an exemplary system architecture diagram to which the present application may be applied;

[0062] Figure 2 is a flow chart of an embodiment of a target detection method according to the present application;

[0063] Figure 3 yes Figure 2 A flowchart of a specific embodiment of step 201 is shown;

[0064] Figure 4 yes Figure 2 A flowchart of a specific embodiment of step 202 is shown;

[0065] Figure 5 yes Figure 4 A flowchart of a specific embodiment of step 403 is shown;

[0066] Figure 6 yes Figure 2 A flowchart of a specific embodiment of identifying the center point in the three-dimensional point cloud before step 203 is shown;

[0067] Figure 7 yes Figure 2 A flowchart of a specific embodiment of step 203 is shown;

[0068] Figure 8 yes Figure 2 A flowchart of another specific embodiment of step 203 is shown;

[0069] Fig. 9 yes Figure 2 A flowchart of a specific embodiment of step 205 is shown;

[0070] Fig.10 is a schematic structural diagram of an embodiment of a target detection device according to the present application;

[0071] Fig.11 It is a structural diagram of an embodiment of a computer device according to the present application. DETAILED DESCRIPTION

[0072] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application or the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0073] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0074] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the accompanying drawings.

[0075] like Figure 1 As shown, the system architecture 100 may include a terminal device 101, a network 102 and a server 103. The terminal device 101 may be a laptop 1011, a tablet computer 1012 or a mobile phone 1013. The network 102 is used to provide a medium for a communication link between the terminal device 101 and the server 103. The network 102 may include various connection types, such as wired, wireless communication links or optical fiber cables.

[0076] The user can use the terminal device 101 to interact with the server 103 through the network 102 to receive or send messages, etc. Various communication client applications can be installed on the terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

[0077] The terminal device 101 can be various electronic devices with a display screen and supporting web browsing. In addition to the laptop computer 1011, the tablet computer 1012 or the mobile phone 1013, the terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 (Moving Picture Experts Group Audio Layer IV) player, a laptop computer and a desktop computer, etc.

[0078] The server 103 may be a server that provides various services, such as a background server that provides support for a web page displayed on the terminal device 101 .

[0079] It should be noted that the target detection method provided in the embodiments of the present application is generally executed by a terminal device, and accordingly, the target detection device is generally arranged in the terminal device.

[0080] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0081] Continue to refer Figure 2 , shows a flow chart of an embodiment of a target detection method according to the present application. The target detection method comprises the following steps:

[0082] Step 201, obtaining a three-dimensional point cloud to be analyzed;

[0083] In this embodiment, the three-dimensional point cloud to be analyzed is obtained by scanning and obtaining the three-dimensional point cloud to be analyzed by medical imaging scanning. The target detection method is applied to human medical imaging scanning and lesion morphology recognition scenarios, which can assist medical personnel to intuitively perform medical imaging analysis and intuitively represent it in a graphical manner, thereby improving the work efficiency of medical personnel.

[0084] In this embodiment, after executing the step of obtaining the three-dimensional point cloud to be analyzed, the method further includes: inputting the three-dimensional point cloud to be analyzed into a preset attention pooling layer to obtain the global position features of the three-dimensional point cloud to be analyzed.

[0085] By obtaining the global position features of the three-dimensional point cloud to be analyzed, it is possible to have a holistic understanding and description of the position information of all points in the three-dimensional point cloud to be analyzed, which can play a certain role as auxiliary information reference when modeling the surface of an object or modeling the interior of an object.

[0086] Step 202, using a three-dimensional space coordinate system to estimate the type of points in the three-dimensional point cloud, and estimating boundary points and non-boundary points as estimation results;

[0087] By estimating the types of points in the three-dimensional point cloud, boundary points and non-boundary points are estimated, so that the boundary points can be directly selected for object surface modeling. Of course, non-boundary points can also be used to model the internal conditions of the scanned object.

[0088] In this embodiment, after executing the step of estimating the type of points in the three-dimensional point cloud using the three-dimensional spatial coordinate system and estimating boundary points and non-boundary points as estimation results, the method further includes: using a distinguishing marking method to perform distinguishing marking processing on the estimated boundary points and non-boundary points, so as to facilitate subsequent rapid analysis of the three-dimensional point cloud. Specifically, the distinguishing marking processing includes distinguishing marking processing based on visual effects, for example: using different marking colors for the boundary points and the non-boundary points, so as to facilitate subsequent direct identification of boundary points and non-boundary points in the three-dimensional point cloud by marking colors; of course, the distinguishing marking processing also includes distinguishing marking processing based on distinguishing format numbers, for example: using the "A-value" format for numbering boundary points and the "B-value" format for numbering non-boundary points, and subsequently directly identifying boundary points and non-boundary points in the three-dimensional point cloud by the corresponding numbering format.

[0089] In this embodiment, after executing the step of using a three-dimensional space coordinate system to estimate the type of points in the three-dimensional point cloud and estimating boundary points and non-boundary points as estimation results, the method also includes: using a point-by-point MLP (Multi-Layer Perceptron) method to extract position features of all non-boundary points to obtain feature extraction results corresponding to all non-boundary points.

[0090] By obtaining the feature extraction results corresponding to all the non-boundary points, it is possible to model the internal situation of the scanned object in combination with the feature extraction results corresponding to all the non-boundary points. At the same time, by using a multi-layer perceptron through the superposition of multiple hidden layers, it is possible to learn more complex and abstract feature representations. MLP is a fully connected neural network, and the neurons in each layer are connected to each neuron in the next layer. This structure also helps to fully transmit feature information point by point.

[0091] Step 203, using a 3D point cloud segmentation technology to perform regional segmentation on the 3D point cloud to obtain a segmented regional point cloud;

[0092] In this embodiment, the three-dimensional point cloud is regionalized by using three-dimensional point cloud segmentation technology to obtain the segmented regionalized point cloud. The purpose is to convert the overall external area analysis composed of all boundary points into analysis of each regionalized boundary area, thereby realizing regionalized analysis when there are too many boundary points, reducing the difficulty of analysis, and improving the analysis and processing efficiency.

[0093] In this embodiment, according to different selection methods, two three-dimensional point cloud segmentation technologies are provided. One is to perform three-dimensional point cloud segmentation based on the spatial reference coordinate system and the preset segmentation surface, and the other is to use the KNN algorithm to divide all boundary points into local areas and combine them with the center point in the spatial reference coordinate system, thereby realizing three-dimensional point cloud segmentation.

[0094] Of course, in order to subsequently identify the segmented regionalized point cloud, each segmented regionalized point cloud can be distinguished and marked.

[0095] Step 204, identifying boundary points in the three-dimensional point cloud contained in all regionalized point clouds according to the estimation result;

[0096] In this embodiment, the boundary points in the three-dimensional point cloud contained in all regionalized point clouds are identified based on the estimation results. The identification steps are first identifying each regionalized point cloud based on the distinguishing mark processing results of each regionalized point cloud, and then identifying the boundary points in each regionalized point cloud based on the distinguishing marks of the boundary points.

[0097] Step 205, based on the boundary points in the three-dimensional point cloud contained in all the regionalized point clouds, an integrated learning algorithm layer is used to extract fusion features of all the regionalized point clouds;

[0098] In this embodiment, the integrated learning algorithm layer includes a first feature extraction sublayer, a second feature extraction sublayer and a third feature extraction sublayer, wherein the first feature extraction sublayer is used to perform position feature extraction on edge points in the local image to obtain edge features corresponding to all local images respectively; the second feature extraction sublayer is used to perform position feature extraction on non-edge points in the local image to obtain non-edge features corresponding to all local images respectively, for example: a point-by-point MLP method is used to perform position feature extraction on non-edge points in the local image to obtain feature extraction results corresponding to all non-edge points; the third feature extraction sublayer is used to extract position features of points in the local image relative to the center point of the three-dimensional point cloud; then, feature fusion is performed on the first feature extraction result, the second feature extraction result and the third feature extraction result corresponding to the same local image to obtain the fused feature information as the fused feature corresponding to the corresponding local image; according to the correspondence between the local image and the regionalized point cloud and the fused feature corresponding to each local image, the fused features corresponding to all regionalized point clouds are obtained.

[0099] In this embodiment, the feature fusion, for example: taking the current local image as an example, the edge features corresponding to the current local image, the non-edge features corresponding to the current local image, and the position features of the points in the local image relative to the center point of the three-dimensional point cloud can be weightedly fused according to the differences between the points in the current local image, thereby obtaining the fusion features of the regionalized point cloud corresponding to the current local image.

[0100] The above steps 203 to 205 are essentially secondary processing of all boundary points in the three-dimensional point cloud. The outer surface of the object composed of all boundary points is segmented by the three-dimensional point cloud segmentation technology. Then, the edge features, non-edge features and position features relative to the center point of the three-dimensional point cloud are extracted from the segmented area. After that, feature fusion is performed to obtain the fusion features of all regionalized point clouds. Especially when the surface of the object is more complex, this processing method can greatly reduce the difficulty of feature extraction and improve the efficiency of feature extraction.

[0101] Step 206, performing object surface modeling based on the fusion features of all regionalized point clouds to obtain an object surface modeling result;

[0102] Specifically, the object surface is modeled based on the fusion features of all regionalized point clouds to obtain the object surface modeling result. It can be understood that after obtaining the fusion features corresponding to all regionalized point clouds, the fusion features corresponding to all regionalized point clouds are fused again to obtain the final fusion features. The object surface is modeled according to the final fusion features to obtain the object surface modeling result. In essence, the final fusion features are the fusion features of the outer surface of the object formed by all boundary points. By modeling the object surface, the object corresponding to the three-dimensional point cloud to be analyzed can be identified.

[0103] Step 207, identifying all target classified objects and the appearance of all target classified objects contained in the three-dimensional point cloud to be analyzed according to the object surface modeling result and the preset object classification model;

[0104] Specifically, the preset object classification model contains medical lesion or disease body classification knowledge set in advance as predictive experience. According to the object surface modeling result and the preset object classification model, all target classification objects contained in the three-dimensional point cloud to be analyzed and the appearance of all target classification objects can be identified. The object surface modeling result can be compared with the medical lesion or disease body classification knowledge set in advance as predictive experience contained in the preset object classification model, so as to identify all target classification objects contained in the three-dimensional point cloud to be analyzed and the appearance of all target classification objects.

[0105] Step 208: Add the name information of all the target classified objects and the appearance of all the target classified objects as descriptive text content to the object surface modeling result to complete the target detection.

[0106] By adding the name information of all target classified objects and the appearance of all target classified objects as descriptive text content to the corresponding medical images, medical personnel can be assisted to intuitively perform medical image analysis and improve their work efficiency.

[0107] In this embodiment, by obtaining a three-dimensional point cloud to be analyzed; estimating boundary points and non-boundary points; regional segmentation, obtaining a segmented regional point cloud; identifying boundary points in the three-dimensional point cloud contained in all regionalized point clouds; using an integrated learning algorithm layer to extract fusion features of all regionalized point clouds; performing object surface modeling; identifying all target classification objects contained in the three-dimensional point cloud to be analyzed and the appearance of all target classification objects; adding the descriptive text content to the object surface modeling result, the target detection is completed. The target detection method is applied to human medical image scanning and lesion morphology recognition scenarios, which can assist medical personnel in intuitively performing medical image analysis and improve the work efficiency of medical personnel.

[0108] Continue to refer Figure 3 , Figure 3 yes Figure 2 The flowchart of a specific embodiment of step 201 shown includes the following steps:

[0109] Step 301, using a preset three-dimensional scanner to perform a three-dimensional scan on a target object;

[0110] Step 302, based on a preset three-dimensional space coordinate system, obtaining a three-dimensional scanning cloud point set output by the three-dimensional scanner;

[0111] Step 303: Use the three-dimensional scanning cloud point set as the three-dimensional point cloud to be analyzed.

[0112] Specifically, the preset three-dimensional scanner includes a three-dimensional scanner for medical scanning scenarios, which can scan internal organs or tissue structures of the human body through X-ray and other technologies, and then obtain the three-dimensional scanning cloud point set output by the three-dimensional scanner to analyze and detect it.

[0113] By applying the target detection method to human medical image scanning and lesion morphology recognition scenarios, it can assist medical personnel in intuitively performing medical image analysis and improve their work efficiency.

[0114] Continue to refer Figure 4 , Figure 4 yes Figure 2 The flowchart of a specific embodiment of step 202 shown includes the following steps:

[0115] Step 401, randomly selecting a point from the three-dimensional point cloud as a target point;

[0116] Step 402, based on the spatial coordinate information of all points in the three-dimensional point cloud in the three-dimensional spatial coordinate system, calculate the spatial vector values ​​of other points relative to the target point one by one to obtain a set of spatial vector values;

[0117] In this embodiment, according to the spatial coordinate information of all points in the three-dimensional point cloud in the three-dimensional space coordinate system, wherein the spatial coordinate information is the coordinate information in the (X, Y, Z) format, the spatial vector values ​​of other points relative to the target point are calculated one by one to obtain a set of spatial vector values. The specific calculation method includes: obtaining the spatial coordinate information of the target point; obtaining the spatial coordinate information of other points; according to the preset calculation method:

[0118]

[0119] Calculate the spatial vector values ​​of other points relative to the target point, where i is the number information of other points. is the space vector value, x i is the X-axis coordinate value of the point numbered i, y i is the Y-axis coordinate value of the point numbered i, z i is the Z-axis coordinate value of the point numbered i, x0 is the X-axis coordinate value of the target point, y0 is the Y-axis coordinate value of the target point, and z0 is the Z-axis coordinate value of the target point.

[0120] Step 403 : Filter out boundary points and non-boundary points in the three-dimensional point cloud by comparing elements in the space vector value set in pairs.

[0121] By comparing the elements in the spatial vector value set in pairs, the boundary points and non-boundary points in the three-dimensional point cloud are screened out, and all boundary points and non-boundary points contained in the three-dimensional point cloud can be identified, which facilitates subsequent object surface modeling processing.

[0122] Continue to refer Figure 5 , Figure 5 yes Figure 4 The flowchart of a specific embodiment of step 403 shown includes the following steps:

[0123] Step 501, determining whether two elements in the space vector value set currently being compared are coplanar and co-directional vectors, wherein the coplanar and co-directional vectors indicate that the two space vector values ​​are coplanar and have the same direction in the three-dimensional space coordinate system;

[0124] Step 502: If two elements in the space vector value set currently being compared are coplanar and co-directional vectors, then mark the point corresponding to the element with the smaller space vector value as a non-boundary point, and continue to execute step 503;

[0125] Specifically, for example: the two elements in the space vector value set are That is, the spatial coordinate information of the target point is (1,1,1). At this time, the spatial coordinate information of the point numbered 1 is (2,2,2), and the spatial coordinate information of the point numbered 2 is (3,3,3). Obviously, and They are coplanar and codirectional vectors. As for the point numbered 1 and the point numbered 2, the point numbered 1 is a non-boundary point, and the type of the point numbered 2 needs to be further compared with other points.

[0126] If the two elements in the space vector value set currently being compared are not coplanar and codirectional vectors, step 503 is directly executed.

[0127] Step 503, reselect two elements from the spatial vector value set, and continue to perform steps 501 to 502 until all elements in the spatial vector value set are compared two by two, and all marked points, i.e., non-boundary points in the three-dimensional point cloud, are obtained;

[0128] Step 504: Obtain all unmarked points, ie, boundary points in the three-dimensional point cloud.

[0129] By comparing the elements in the spatial vector value set in pairs, the boundary points and non-boundary points in the three-dimensional point cloud are screened out, and all boundary points and non-boundary points contained in the three-dimensional point cloud can be identified, which facilitates subsequent object surface modeling processing.

[0130] Continue to refer Figure 6 In some optional implementations, before step 203, a step of identifying a center point in the three-dimensional point cloud is also included. Figure 6 yes Figure 2 The flowchart of a specific embodiment of identifying the center point in the three-dimensional point cloud before step 203 includes the following steps:

[0131] Step 601, clustering the non-boundary points in the three-dimensional point cloud using a K-means algorithm to obtain a clustering result, wherein the K value in the K-means algorithm represents the center point data in the three-dimensional point cloud and is set to a fixed value of 1;

[0132] Step 602: According to the clustering processing result, identify the center point of all non-boundary points as the center point of the three-dimensional point cloud.

[0133] By adopting the K-means algorithm and setting the number of clusters to 1, the center point in the cluster can be identified as the center point of the three-dimensional point cloud, so as to extract the position features of all the above localized point clouds in combination with the center point.

[0134] Continue to refer Figure 7 , Figure 7 yes Figure 2 The flowchart of a specific embodiment of step 203 shown includes the following steps:

[0135] Step 701, setting the center point as the spatial coordinate origin in the three-dimensional spatial coordinate system, and dividing the spatial reference coordinate system having an X-axis, a Y-axis and a Z-axis by the spatial coordinate origin;

[0136] Specifically, the spatial reference coordinate system divided by the spatial coordinate origin into an X-axis, a Y-axis, and a Z-axis can be understood as a spatial reference coordinate system after updating the preset three-dimensional spatial coordinate system. It can be understood that the spatial coordinate origin of the preset three-dimensional spatial coordinate system is a preset value, which is not necessarily the center point, and the X-axis, Y-axis, and Z-axis are also preset, and the spatial coordinate origin of the updated spatial reference coordinate system, that is, the intersection of the X-axis, Y-axis, and Z-axis is the center point.

[0137] Step 702, segmenting the three-dimensional point cloud according to the spatial reference coordinate system and the preset segmentation plane to obtain a segmentation result;

[0138] In this embodiment, in order to reduce the number of computer calculation steps, the preset dividing plane is preferably the plane where the center point is located. Furthermore, the dividing plane is the plane formed by the X-axis and the Y-axis in the three-dimensional space coordinate system, the plane formed by the X-axis and the Z-axis in the three-dimensional space coordinate system, and the plane formed by the Z-axis and the Y-axis in the three-dimensional space coordinate system.

[0139] Step 703: Generate a local map for each local area according to the segmentation result to obtain local maps corresponding to all boundary points.

[0140] Continue to refer Figure 8 , Figure 8 yes Figure 2 The flowchart of another specific embodiment of step 203 shown includes the following steps:

[0141] Step 801, using the KNN algorithm to perform local area division on all boundary points to obtain a local area division result, wherein the K value in the KNN algorithm represents the number of boundary points in each local area, and K is a positive integer greater than 1;

[0142] Specifically, assuming that the K value in the KNN algorithm is 20, that is, all the boundary points contained in the three-dimensional point cloud are clustered so that one clustering result contains 20 boundary points. For example, if the three-dimensional point cloud contains 120 boundary points, the KNN algorithm is used to divide all the boundary points into A regionalized point cloud. Of course, the number of boundary points contained in the three-dimensional point cloud here is exactly divisible by the K value, that is, 120 and 20; if the number of boundary points contained in the three-dimensional point cloud is not divisible by the K value in the KNN algorithm, for example: the number of boundary points is 130 and the K value is 20, then the method of rounding up is adopted to obtain the number of the regionalized point cloud as 7.

[0143] Step 802: Generate a local map for each local area according to the local area division result and the center point of the three-dimensional point cloud, and obtain local maps corresponding to all boundary points.

[0144] Through the above two different local area division methods, that is, two different three-dimensional point cloud segmentation methods, different segmentation methods can be provided to analysts, which are more scientific and humane and combined with the actual division situation.

[0145] Continue to refer Fig. 9 , Fig. 9 yes Figure 2 The flowchart of a specific embodiment of step 205 shown includes the following steps:

[0146] Step 901: extract position features of edge points in all local images respectively according to the first feature extraction sublayer, and obtain edge features corresponding to all local images respectively as first feature extraction results;

[0147] Step 902: performing position feature extraction on non-edge points in all local images respectively according to the second feature extraction sublayer, and obtaining non-edge features corresponding to all local images respectively as second feature extraction results;

[0148] Step 903, according to the third feature extraction sublayer, respectively extracting position features of points in all local images relative to the center point of the three-dimensional point cloud as third feature extraction results;

[0149] Step 904, performing feature fusion on the first feature extraction result, the second feature extraction result and the third feature extraction result corresponding to the same local image, and obtaining fused feature information as a fused feature corresponding to the corresponding local image;

[0150] Step 905 , according to the correspondence between the local graph and the regionalized point cloud, and the fusion features corresponding to each local graph, the fusion features corresponding to all the regionalized point clouds are obtained.

[0151] In essence, the first feature extraction sublayer is used to extract position features of edge points in all local images, the second feature extraction sublayer is used to extract position features of non-edge points in all local images, and the third feature extraction sublayer is used to extract position features of points in all local images relative to the center point of the three-dimensional point cloud. After that, the features of the same point extracted by the first feature extraction sublayer and the third feature extraction sublayer are fused, and then the features of the same point extracted by the second feature extraction sublayer and the third feature extraction sublayer are fused, and then the fused features corresponding to each point are fused again to obtain feature fusion information of the local image. The local image processing method avoids the object surface corresponding to the three-dimensional point cloud from being too complex, and reduces the difficulty of feature extraction and feature fusion.

[0152] This application completes the target detection by acquiring the three-dimensional point cloud to be analyzed; estimating the boundary points and non-boundary points; regional segmentation, obtaining the segmented regional point cloud; identifying the boundary points in the three-dimensional point cloud contained in all regional point clouds; using the integrated learning algorithm layer to extract the fusion features of all regional point clouds; performing object surface modeling; identifying all target classification objects contained in the three-dimensional point cloud to be analyzed and the appearance of all target classification objects; adding the descriptive text content to the object surface modeling result. Applying the target detection method to the human medical image scanning and lesion morphology recognition scenarios can assist medical personnel in intuitively analyzing medical images and improve their work efficiency.

[0153] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0154] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0155] In the embodiment of the present application, by obtaining a three-dimensional point cloud to be analyzed; estimating boundary points and non-boundary points; regional segmentation, obtaining a segmented regional point cloud; identifying boundary points in the three-dimensional point cloud contained in all regionalized point clouds; using an integrated learning algorithm layer to extract fusion features of all regionalized point clouds; performing object surface modeling; identifying all target classification objects contained in the three-dimensional point cloud to be analyzed and the appearance of all target classification objects; adding the descriptive text content to the object surface modeling result, the target detection is completed. The target detection method is applied to human medical image scanning and lesion morphology recognition scenarios, which can assist medical personnel in intuitively performing medical image analysis and improve the work efficiency of medical personnel.

[0156] Further references Fig.10 , as a response to the above Figure 2 In order to realize the method shown in the figure, the present application provides an embodiment of a target detection device, and the device embodiment is Figure 2 Corresponding to the method embodiment shown, the device can be specifically applied to various electronic devices.

[0157] like Fig.10 As shown, the target detection device 10 described in this embodiment includes: a three-dimensional point cloud acquisition module 10a, a point type estimation module 10b, a point cloud regional segmentation module 10c, a boundary point recognition module 10d, a fusion feature extraction module 10e, an object surface modeling module 10f, an object classification and recognition module 10g and a description text content adding module 10h. Among them:

[0158] A three-dimensional point cloud acquisition module 10a is used to acquire a three-dimensional point cloud to be analyzed;

[0159] A point type estimation module 10b is used to estimate the type of points in the three-dimensional point cloud by using a three-dimensional space coordinate system, and estimate boundary points and non-boundary points as estimation results;

[0160] The point cloud regionalization segmentation module 10c is used to perform regional segmentation on the three-dimensional point cloud using a three-dimensional point cloud segmentation technology to obtain a segmented regionalized point cloud;

[0161] A boundary point identification module 10d, configured to identify boundary points in the three-dimensional point cloud contained in all regionalized point clouds according to the estimation result;

[0162] A fusion feature extraction module 10e is used to extract fusion features of all regionalized point clouds using an integrated learning algorithm layer based on the boundary points in the three-dimensional point clouds contained in all regionalized point clouds;

[0163] The object surface modeling module 10f is used to perform object surface modeling based on the fusion features of all regionalized point clouds to obtain the object surface modeling result;

[0164] The object classification and recognition module 10g is used to recognize all target classification objects and the appearance of all target classification objects contained in the three-dimensional point cloud to be analyzed according to the object surface modeling result and the preset object classification model;

[0165] The description text content adding module 10h is used to add the name information of all target classification objects and the appearance of all target classification objects as description text content to the object surface modeling result to complete this target detection.

[0166] This application completes the target detection by acquiring the three-dimensional point cloud to be analyzed; estimating the boundary points and non-boundary points; regional segmentation, obtaining the segmented regional point cloud; identifying the boundary points in the three-dimensional point cloud contained in all regional point clouds; using the integrated learning algorithm layer to extract the fusion features of all regional point clouds; performing object surface modeling; identifying all target classification objects contained in the three-dimensional point cloud to be analyzed and the appearance of all target classification objects; adding the descriptive text content to the object surface modeling result. Applying the target detection method to the human medical image scanning and lesion morphology recognition scenarios can assist medical personnel in intuitively analyzing medical images and improve their work efficiency.

[0167] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through computer-readable instructions, and the computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, the aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0168] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0169] To solve the above technical problems, the present application also provides a computer device. Fig.11 , Fig.11 This is a basic structural block diagram of the computer device in this embodiment.

[0170] The computer device 11 includes a memory 11a, a processor 11b, and a network interface 11c which are interconnected through a system bus. It should be noted that: Fig.11 Only a computer device 11 having components memory 11a, processor 11b, and network interface 11c is shown, but it should be understood that it is not required to implement all the components shown, and more or fewer components may be implemented instead. Among them, those skilled in the art can understand that the computer device here is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital processors (DSPs), embedded devices, etc.

[0171] The computer device may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, etc. The computer device may interact with a user through a keyboard, a mouse, a remote controller, a touch pad, or a voice control device.

[0172] The memory 11a includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 11a can be an internal storage unit of the computer device 11, such as a hard disk or memory of the computer device 11. In other embodiments, the memory 11a can also be an external storage device of the computer device 11, such as a plug-in hard disk equipped on the computer device 11, a smart memory card (Smart Med ia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. Of course, the memory 11a can also include both the internal storage unit of the computer device 11 and its external storage device. In this embodiment, the memory 11a is generally used to store an operating system and various application software installed on the computer device 11, such as a computer-readable instruction of a target detection method, etc. In addition, the memory 11a can also be used to temporarily store various data that have been output or will be output.

[0173] The processor 11b may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor, or other data processing chip in some embodiments. The processor 11b is generally used to control the overall operation of the computer device 11. In this embodiment, the processor 11b is used to run the computer-readable instructions stored in the memory 11a or process data, such as computer-readable instructions for running the target detection method.

[0174] The network interface 11c may include a wireless network interface or a wired network interface. The network interface 11c is generally used to establish a communication connection between the computer device 11 and other electronic devices.

[0175] The computer device proposed in this embodiment belongs to the field of image processing technology and is applied to the scene of three-dimensional modeling and reconstruction of objects. This application obtains the three-dimensional point cloud to be analyzed; estimates the boundary points and non-boundary points; regionalizes the segmentation to obtain the regionalized point cloud; identifies the boundary points in the three-dimensional point cloud contained in all regionalized point clouds; uses the integrated learning algorithm layer to extract the fusion features of all regionalized point clouds; performs object surface modeling; identifies all target classification objects contained in the three-dimensional point cloud to be analyzed and the appearance of all target classification objects; adds the descriptive text content to the object surface modeling result to complete this target detection. The target detection method is applied to the human medical image scanning and lesion morphology recognition scenes, which can assist medical personnel in intuitively performing medical image analysis and improve the work efficiency of medical personnel.

[0176] The present application also provides another implementation, namely, providing a computer-readable storage medium, wherein the computer-readable storage medium stores computer-readable instructions, and the computer-readable instructions can be executed by a processor to enable the processor to perform the steps of the target detection method as described above.

[0177] The computer-readable storage medium proposed in this embodiment belongs to the field of image processing technology and is applied to the scene of three-dimensional modeling and reconstruction of objects. This application obtains the three-dimensional point cloud to be analyzed; estimates the boundary points and non-boundary points; regionalizes the segmentation to obtain the regionalized point cloud; identifies the boundary points in the three-dimensional point cloud contained in all regionalized point clouds; uses the integrated learning algorithm layer to extract the fusion features of all regionalized point clouds; performs object surface modeling; identifies all target classification objects contained in the three-dimensional point cloud to be analyzed and the appearance of all target classification objects; adds the descriptive text content to the object surface modeling result to complete this target detection. The target detection method is applied to the human medical image scanning and lesion morphology recognition scenes, which can assist medical personnel in intuitively performing medical image analysis and improve the work efficiency of medical personnel.

[0178] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0179] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods, or to perform equivalent replacement of some of the technical features therein. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A target detection method, characterized in that: The steps include: Obtain the three-dimensional point cloud to be analyzed; Using a three-dimensional space coordinate system to estimate the type of points in the three-dimensional point cloud, and estimating boundary points and non-boundary points as estimation results; Using a three-dimensional point cloud segmentation technology to perform regional segmentation on the three-dimensional point cloud to obtain a segmented regional point cloud; Identifying boundary points in the three-dimensional point cloud contained in all regionalized point clouds according to the estimation result; Based on the boundary points in the three-dimensional point cloud contained in all the regionalized point clouds, an integrated learning algorithm layer is used to extract fusion features of all the regionalized point clouds; The object surface is modeled based on the fusion features of all regionalized point clouds to obtain the object surface modeling result; According to the object surface modeling result and the preset object classification model, identifying all target classification objects contained in the three-dimensional point cloud to be analyzed and the appearance of all target classification objects; The name information of all target classified objects and the appearance of all target classified objects are added as descriptive text content to the object surface modeling result to complete this target detection.

2. The target detection method according to claim 1, characterized in that: The step of obtaining the three-dimensional point cloud to be analyzed specifically includes: Use a preset 3D scanner to perform 3D scanning on the target object; Based on a preset three-dimensional space coordinate system, obtaining a three-dimensional scanning cloud point set output by the three-dimensional scanner; The three-dimensional scanning cloud point set is used as the three-dimensional point cloud to be analyzed.

3. The target detection method according to claim 1 or 2, characterized in that: The step of estimating the type of points in the three-dimensional point cloud by using a three-dimensional space coordinate system and estimating boundary points and non-boundary points as estimation results specifically includes: Selecting any point from the three-dimensional point cloud as a target point; According to the spatial coordinate information of all points in the three-dimensional point cloud in the three-dimensional spatial coordinate system, the spatial vector values ​​of other points relative to the target point are calculated one by one to obtain a set of spatial vector values; By comparing the elements in the space vector value set in pairs, boundary points and non-boundary points in the three-dimensional point cloud are screened out.

4. The target detection method according to claim 3, characterized in that: The step of screening out boundary points and non-boundary points in the three-dimensional point cloud by comparing the elements in the space vector value set in pairs specifically includes: Step 501, determining whether two elements in the space vector value set currently being compared are coplanar and co-directional vectors, wherein the coplanar and co-directional vectors indicate that the two space vector values ​​are coplanar and have the same direction in the three-dimensional space coordinate system; Step 502: If two elements in the space vector value set currently being compared are coplanar and co-directional vectors, then mark the point corresponding to the element with the smaller space vector value as a non-boundary point, and continue to execute step 503; Step 503, reselect two elements from the spatial vector value set, and continue to perform steps 501 to 502 until all elements in the spatial vector value set are compared two by two, and all marked points, i.e., non-boundary points in the three-dimensional point cloud, are obtained; Step 504: Obtain all unmarked points, ie, boundary points in the three-dimensional point cloud.

5. The target detection method according to claim 1, characterized in that: Before executing the step of regionalizing the three-dimensional point cloud using the three-dimensional point cloud segmentation technology to obtain the segmented regionalized point cloud, the method further includes: Using a K-means algorithm to perform clustering processing on non-boundary points in the three-dimensional point cloud to obtain a clustering processing result, wherein the K value in the K-means algorithm represents the center point data in the three-dimensional point cloud and is set to a fixed value of 1; According to the clustering processing result, the center point of all non-boundary points is identified as the center point of the three-dimensional point cloud.

6. The target detection method according to claim 5, characterized in that: The step of using the 3D point cloud segmentation technology to perform regional segmentation on the 3D point cloud to obtain the segmented regional point cloud includes: The center point is set as the spatial coordinate origin in the three-dimensional spatial coordinate system, and a spatial reference coordinate system having an X-axis, a Y-axis and a Z-axis is divided by the spatial coordinate origin; According to the spatial reference coordinate system and the preset segmentation plane, the three-dimensional point cloud is segmented to obtain a segmentation result; A local map is generated for each local area according to the segmentation processing result to obtain local maps corresponding to all boundary points.

7. The target detection method according to claim 5, characterized in that: The step of regionalizing the three-dimensional point cloud using the three-dimensional point cloud segmentation technology to obtain the segmented regionalized point cloud also includes: Using the KNN algorithm, all boundary points are divided into local areas to obtain local area division results, wherein the K value in the KNN algorithm represents the number of boundary points in each local area, and K is a positive integer greater than 1; A local map is generated for each local area according to the local area division result and the center point of the three-dimensional point cloud, so as to obtain local maps corresponding to all boundary points.

8. The target detection method according to claim 6 or 7, characterized in that: The integrated learning algorithm layer includes a first feature extraction sublayer, a second feature extraction sublayer and a third feature extraction sublayer. The step of extracting fusion features of all regionalized point clouds using the integrated learning algorithm layer based on the boundary points in the three-dimensional point clouds contained in all regionalized point clouds specifically includes: According to the first feature extraction sublayer, position features are extracted for edge points in all local images respectively, and edge features corresponding to all local images are obtained as first feature extraction results; According to the second feature extraction sublayer, position feature extraction is performed on non-edge points in all local images respectively, and non-edge features corresponding to all local images are obtained as second feature extraction results; According to the third feature extraction sublayer, respectively extracting position features of points in all local images relative to the center point of the three-dimensional point cloud as third feature extraction results; Performing feature fusion on the first feature extraction result, the second feature extraction result and the third feature extraction result corresponding to the same local image, and obtaining fused feature information as the fused feature corresponding to the corresponding local image; According to the correspondence between the local graph and the regionalized point cloud, and the fusion features corresponding to each local graph, the fusion features corresponding to all the regionalized point clouds are obtained.

9. A target detection device, characterized in that: include: A three-dimensional point cloud acquisition module is used to acquire the three-dimensional point cloud to be analyzed; A point type estimation module, used to estimate the type of points in the three-dimensional point cloud by using a three-dimensional space coordinate system, and estimate boundary points and non-boundary points as estimation results; A point cloud regional segmentation module is used to perform regional segmentation on the three-dimensional point cloud using a three-dimensional point cloud segmentation technology to obtain a segmented regional point cloud; A boundary point recognition module, used to recognize boundary points in the three-dimensional point cloud contained in all regionalized point clouds according to the estimation result; A fusion feature extraction module, used to extract fusion features of all regionalized point clouds using an integrated learning algorithm layer based on boundary points in the three-dimensional point clouds contained in all regionalized point clouds; The object surface modeling module is used to perform object surface modeling based on the fusion features of all regionalized point clouds to obtain the object surface modeling results; An object classification and recognition module, used to recognize all target classification objects and the appearance of all target classification objects contained in the three-dimensional point cloud to be analyzed according to the object surface modeling result and a preset object classification model; The description text content adding module is used to add the name information of all target classification objects and the appearance of all target classification objects as description text content to the object surface modeling result to complete this target detection.

10. A computer device, characterized in that: The method comprises a memory and a processor, wherein the memory stores computer-readable instructions, and the processor implements the steps of the target detection method according to any one of claims 1 to 8 when executing the computer-readable instructions.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by a processor, the steps of the target detection method according to any one of claims 1 to 8 are implemented.

Images