TWI798098B - Method for detecting three-dimensional target object, electronic device and storage medium - Google Patents

Abstract

The present application provides a method for detecting a three-dimensional (3D) target object, an electronic device and a storage medium. The method includes: acquiring a detection image and a depth image corresponding to the detection image; inputting the detection image into a trained target detection model, and determining a category of an object in the detection image and a two-dimensional (2D) bounding box of the object; determining an object model and a 3D bounding box from a 3D object model library according to the object type; calculating point cloud data and a distance from a depth camera to the object model according to the depth image and the 2D bounding box; determining a rotation angle of the object model according to the object model and point cloud data; determining a position of the object model in the 3D space according to the distance from the depth camera to the object model, the rotation angle and the 3D bounding box,. The present application can quickly determine a position of an object in 3D space from an image.

Description

Three-dimensional object detection method, electronic device and computer-readable storage medium

The present application relates to computer vision and deep learning technology, in particular to a three-dimensional object detection method, electronic equipment and computer-readable storage media.

In the field of automatic driving, the automatic driving system will use different types of sensors to detect objects in front of or near the vehicle and make corresponding decisions. Therefore, the automatic driving system needs to quickly and accurately detect the type and location of objects to ensure driving safety. At present, most 3D object detection algorithms require a large number of labeled samples to detect the category of objects. It is difficult to label the rotation angle of objects and requires the use of regression operations, which takes a long time to predict. In addition, most of the current 3D target detection algorithms also need to accurately detect the distance between the vehicle and the object in front. At present, most methods use lidar or radar to obtain depth information, but the cost of using lidar or radar is high and the field of view is relatively small.

In view of the above, it is necessary to provide a three-dimensional object detection method, an electronic device and a computer-readable storage medium, which can solve the problems of difficult labeling of rotation angles and high detection costs.

An embodiment of the present application provides a three-dimensional target detection method, the three-dimensional target detection method includes: acquiring a detection image and a depth image corresponding to the detection image, wherein the depth image is acquired by a depth camera; The detection image is input to the trained target detection model, and the target detection model is used to determine the object category of the object in the detection image and the two-dimensional boundary frame of the object; According to the object category, determine the object model corresponding to the object and the three-dimensional boundary frame corresponding to the object model from the three-dimensional object model library; according to the depth image and the two-dimensional boundary frame, calculate the The point cloud data of the object framed by the two-dimensional border frame and the distance from the depth camera to the object model; according to the object model and the point cloud data, determine the rotation angle of the object model; according to the The distance from the depth camera to the object model, the rotation angle, and the three-dimensional boundary frame determine the position of the object model in three-dimensional space.

In an optional implementation manner, according to the depth image and the two-dimensional border frame, the point cloud data of the object framed by the two-dimensional border frame and the depth camera to the object are calculated The distance of the model includes: obtaining the depth value and coordinates of the object framed by the two-dimensional border frame according to the depth image, and determining the distance from the depth camera to the object model of the object according to the depth value; The coordinates and the internal and external parameter matrix transformation formula of the depth camera are used to obtain the point cloud data.

In an optional implementation manner, the determining the rotation angle of the object model according to the object model and the point cloud data includes: obtaining a first point cloud of the object outline according to the point cloud data data; converting the object model into second point cloud data; performing point cloud matching on the first point cloud data and the second point cloud data to obtain the rotation angle of the object model.

In an optional embodiment, the method further includes: obtaining training images; constructing a target detection model based on You Only Look Once (YOLO) network; inputting the training images into the target detection model for training, The eigenvalue data of the training image is obtained after alternate processing of convolution and mean pooling by the target detection model; the eigenvalue data is processed by the fully connected layer of the target detection model to obtain the training image The two-dimensional boundary frame and object category of the object in the image are obtained by adjusting the parameters of the target detection model to minimize the loss function to obtain the trained target detection model.

In an optional implementation manner, the feature value data is processed by using the fully connected layer of the target detection model to obtain the two-dimensional bounding box and object category package of the object in the training image Including: using the fully connected layer of the target detection model to process the feature value data to obtain a plurality of candidate two-dimensional border frames of the object in the training image, and demaximizing the plurality of candidate two-dimensional border frames value suppression operation to obtain the two-dimensional boundary frame and object category of the object in the training image.

In an optional implementation manner, the method further includes: establishing the three-dimensional object model library, wherein the three-dimensional object model library includes a plurality of object models corresponding to different object categories and a corresponding object model A three-dimensional border frame, where the three-dimensional border frame includes the length, width, and height corresponding to each object category.

In an optional implementation manner, the determining the position of the object model in the three-dimensional space according to the distance from the depth camera to the object model, the rotation angle, and the three-dimensional border box includes: according to the The rotation angle determines the direction of the object model in the three-dimensional space; according to the direction of the object model in the three-dimensional space, the distance from the depth camera to the object model, and the three-dimensional edge of the object model frame to determine the position of the object model in the three-dimensional space.

In an optional implementation manner, the method further includes: using the position of the object model in the three-dimensional space as the position of the object in the three-dimensional space, and outputting the object category and the object's position in the three-dimensional space s position.

The embodiment of the present application also provides an electronic device, the electronic device includes a processor and a memory, and the processor is configured to execute a computer program stored in the memory to implement the three-dimensional object detection method.

The embodiment of the present application also provides a computer-readable storage medium, the computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by a processor, the above three-dimensional object detection method is realized.

By using the technical solutions provided by the embodiments of the present application, there is no need to perform complex calculations and mark the rotation angle of the object, which reduces labor costs and can quickly obtain the three-dimensional position of the object.

4: Electronic equipment

401: memory

402: Processor

403:Computer program

404: communication bus

101-106: Steps

21-24: Steps

FIG. 1 is a flowchart of a three-dimensional object detection method provided by an embodiment of the present application.

FIG. 2 is a flow chart of the non-maximum value suppression method provided by the embodiment of the present application.

FIG. 3 is a schematic diagram of determining a three-dimensional boundary frame provided by an embodiment of the present application.

FIG. 4 is a schematic structural diagram of an electronic device provided by an embodiment of the present application.

In order to more clearly understand the above objects, features and advantages of the present application, the present application will be described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the specific embodiments described here are only used to explain the present application, and are not used to limit the present application.

A lot of specific details are set forth in the following description to facilitate a full understanding of the application, and the described embodiments are only a part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in this application, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the scope of protection of this application.

Hereinafter, the terms "first" and "second" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of indicated technical features. Thus, a feature defined as "first" and "second" may explicitly or implicitly include one or more of these features. In the description of some embodiments of the present application, words such as "exemplary" or "for example" are used as examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in some embodiments of the present application should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "such as" is intended to present related concepts in a concrete manner.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field of the application. The terms used herein in the specification of the application are only for the purpose of describing specific embodiments, and are not intended to limit the application.

Referring to FIG. 1 , FIG. 1 is a flowchart of a three-dimensional object detection method provided by an embodiment of the present application. The method is applied to an electronic device (for example, the electronic device 4 shown in FIG. 4 ), and the electronic device may be any electronic product capable of human-computer interaction with the user, such as a personal computer, a tablet computer, a smart phone , Personal Digital Assistant (PDA), game consoles, Internet Protocol Television (IPTV), smart wearable devices, etc.

The electronic device is a device that can automatically perform numerical calculations and/or information processing according to preset or stored instructions, and its hardware includes, but is not limited to: microprocessors, Application Specific Integrated Circuits (ASICs) ), Programmable Gate Array (Field-Programmable Gate Array, FPGA), Digital Signal Processor (Digital Signal Processor, DSP), embedded devices, etc.

The network where the electronic device is located includes, but is not limited to: the Internet, a wide area network, a metropolitan area network, a local area network, a virtual private network (Virtual Private Network, VPN) and the like.

The method specifically includes the following.

101. Acquire a detection image and a depth image corresponding to the detection image.

In at least one embodiment of the present application, a camera installed inside or outside the vehicle is used for shooting, and the captured image in front of the vehicle is used as the detection image.

In at least one embodiment of the present application, obtaining the depth image corresponding to the detection image includes: using a depth camera to obtain a depth image, and using the depth camera installed on the vehicle to take an image in front of the vehicle as the depth image . It should be noted that when a camera installed inside or outside the vehicle is used to capture an image in front of the vehicle as a detection image, the depth camera simultaneously captures the image in front of the vehicle. A square image is used as a depth image, and the depth image corresponds to the detection image. For example, different types of cameras are used to take pictures of the same object in front of the vehicle to obtain detection images and depth images.

In the embodiment of the present application, the application scenarios of the three-dimensional object detection method include the field of vehicle automatic driving. During the running of the vehicle, the three-dimensional object detection method is applied to realize three-dimensional detection of objects in front of the vehicle.

102. Input the detection image into the trained target detection model, and use the target detection model to determine the object category of the object in the detection image and the two-dimensional boundary frame of the object.

In at least one embodiment of the present application, the trained target detection model includes: constructing a target detection model based on a You Only Look Once (YOLO) network, and the YOLO network includes a YOLOv3 network or a YOLOv5 network.

In at least one embodiment of the present application, the method for training a target detection model to obtain the trained target detection model includes: acquiring a training image; inputting the training image into the target detection model for training, by means of the The target detection model is alternately processed by convolution and mean value pooling to obtain the eigenvalue data of the training image; using the fully connected layer of the target detection model to process the eigenvalue data to obtain the object in the training image The two-dimensional bounding box and object category, by adjusting the parameters of the target detection model to minimize the loss function, to obtain the trained target detection model. In this embodiment, the parameters of the target detection model include, but are not limited to, the learning rate and the number of iterations of the target detection model. In this embodiment, the loss function of the target detection model includes a mean square error loss function.

In at least one embodiment of the present application, the acquiring training images further includes: performing data enhancement operations on the training images to acquire more different training samples, the data enhancement Operations include, but are not limited to, flipping images, rotating images, scaling images, and cropping images. Through the data enhancement operation, the sample data can be effectively expanded, and the target detection model can be trained and optimized using more training images (for example, images in front of the vehicle) in different scenarios, so that the target detection model can be more robust sex.

In at least one embodiment of the present application, the process of using the fully connected layer of the target detection model to process the feature value data to obtain the two-dimensional bounding box and object category of the object in the training image includes: using the The fully connected layer of the target detection model processes the feature value data to obtain a plurality of candidate two-dimensional border boxes and scores of the plurality of candidate two-dimensional border boxes. In this embodiment, the score of the candidate two-dimensional bounding box includes the score of the fully connected layer after performing category prediction on the object category in the candidate two-dimensional bounding box, that is, the object category is included in the candidate two-dimensional bounding box. Scores for probabilities inside the bounding box. In this embodiment, non-maximum suppression operation (Non-Maximum Suppression, NMS) is performed on multiple candidate two-dimensional bounding boxes to obtain the two-dimensional bounding box and object category of the object.

In this embodiment, the non-maximum suppression operation (Non-Maximum Suppression, NMS) refers to the flow chart described in FIG. 2, specifically including:

21. According to the scores of the candidate 2D border frames, sort the candidate 2D border frames, and select the candidate 2D border frame with the highest score. The "candidate 2D bounding box" is a candidate 2D bounding box of the object in the training image.

22. Traverse other candidate two-dimensional border frames, calculate the intersection over union (IOU) between other candidate two-dimensional border frames and the selected candidate two-dimensional border frame, and delete the intersection over union ratio corresponding to the preset threshold value Candidate 2D bounding boxes. In this embodiment, the intersection ratio is the degree of overlap between the selected candidate 2D border frame (ie, the one with the highest score) and other candidate 2D border frames.

23. Determine whether other candidate two-dimensional border boxes exist besides the selected candidate two-dimensional border boxes. If there are other candidate two-dimensional border boxes, the process returns to step 21 . if not exist For other candidate 2D bounding boxes, perform step 24, and output the selected candidate 2D bounding boxes as the 2D bounding boxes of the objects in the training image.

In at least one embodiment of the present application, by using the above method, the training of the target detection model can be completed, and the trained target detection model can be obtained. Further, the detection image is input into the trained target detection model, and the object category and the two-dimensional boundary frame in the detection image are output.

103. Determine an object model corresponding to the object and a three-dimensional border box corresponding to the object model from a three-dimensional object model library according to the object category.

In at least one embodiment of the present application, the three-dimensional object detection method further includes: pre-establishing a three-dimensional object model library, wherein the three-dimensional object model library includes multiple object models corresponding to different object categories and corresponding to each object model The three-dimensional boundary frame, each three-dimensional boundary frame includes the length, width and height corresponding to the object category.

In this embodiment, the 3D object model library is searched according to the object category to determine the object model, and the 3D boundary frame of the object model is determined according to the object model. For example, as shown in FIG. 3 , FIG. 3 is a schematic diagram of determining a three-dimensional boundary frame provided by an embodiment of the present application. When the object category is a car, search for the object model of the car based on the three-dimensional object model library, and search for the three-dimensional border frame of the car according to the object model of the car; when the object category is a small truck, based on the three-dimensional The object model library searches for the object model of the small truck, and searches for the three-dimensional border frame of the small truck according to the object model of the small truck; when the object category is an electric vehicle, based on the three-dimensional object model library, searches for the object model of the electric vehicle, according to The object model of the electric vehicle is searched for the three-dimensional boundary frame of the electric vehicle; when the object type is a large bus, based on the three-dimensional object model library, the object model of the large bus is searched, and the three-dimensional edge of the large bus is searched according to the object model of the large bus frame. In this embodiment, the object model includes, but not limited to, a three-dimensional model.

104. According to the depth image and the two-dimensional border frame, calculate the point cloud data of the object framed by the two-dimensional border frame and the distance from the depth camera to the object model.

In at least one embodiment of the present application, the method for determining the distance from the depth camera to the object model includes: acquiring the depth value of the object framed by the two-dimensional border frame according to the depth image; The depth value determines the distance from the depth camera to the object model of the object. In this embodiment, the depth value is obtained by a depth camera. When a depth image is captured by a depth camera, the depth camera displays a depth value, and the depth value is the distance from the depth camera to the object. In an embodiment, the distance from the depth camera to the object is taken as the distance from the depth camera to the object model of the object.

In at least one embodiment of the present application, the method for obtaining the point cloud data includes: obtaining the coordinate set of the object framed by the two-dimensional border frame according to the depth image; according to the coordinate set and the The transformation formula of the internal and external parameter matrix of the depth camera is used to obtain the point cloud data. In this embodiment, the coordinate set of the object framed by the two-dimensional border frame is the pixel coordinate set of the object; the point cloud data corresponds to the coordinates in the coordinate set of the object framed by the two-dimensional border frame The world coordinates of . The point cloud data is data used to characterize the outline of an object. Converting the coordinates in the coordinate set of the object framed by the two-dimensional border frame into the corresponding world coordinates needs to be obtained through the transformation formula of the internal and external parameters matrix, and the transformation formula of the internal and external parameters matrix is:

Wherein (x, y, z) is the world coordinate, used to represent the point cloud of a pixel coordinate, f is the focal length, D is the depth value, (x ₁ , y ₁ ) is the object framed by the two-dimensional border The pixel coordinates of any pixel in the coordinate set of . Using the above formula to convert all the coordinates in the coordinate set into world coordinates one by one to obtain the point cloud data.

105. Determine a rotation angle of the object model of the object according to the object model of the object and the point cloud data.

Obtain the first point cloud data of the object outline according to the point cloud data; convert the object model of the object into the second point cloud data; combine the first point cloud data with the second point cloud data Perform point cloud matching to obtain the rotation angle of the object model of the object.

In at least one embodiment of the present application, the converting the object model of the object into the second point cloud data includes, using multiple functions in the point cloud library (Point Cloud Library, PCL) to read the data of the object. The object model and generate the point cloud data of the object model of the object as the second point cloud data.

In at least one embodiment of the present application, performing point cloud matching on the first point cloud data and the second point cloud data to obtain the rotation angle of the object model of the object includes: combining the first point cloud The points of the object outline in the data are fitted into a first plane and the curvature of the first plane is calculated, and the points of the second point cloud data are fitted into a second plane and the curvature of the second plane is calculated; the second plane is calculated; The difference between the curvature of the first plane and the curvature of the second plane is used to obtain a curvature deviation value, and the rotation angle of the object model of the object is determined according to the curvature deviation value.

106. Determine the position of the object in the three-dimensional space.

In at least one embodiment of the present application, the direction of the object model in the three-dimensional space is determined according to the rotation angle, and according to the direction of the object model in the three-dimensional space, the depth camera to the The distance of the object model and the 3D boundary frame of the object model determine the position of the object model in the 3D space. Specifically, the position of the object model in the three-dimensional space is used as the position of the object in the three-dimensional space, and the object category and the three-dimensional space of the object are output position in . For example, the object category and the position of the object in the three-dimensional space are displayed on a display screen in the form of a three-dimensional border frame.

The above is only a specific embodiment of the application, but the scope of protection of the application is not limited thereto. For those of ordinary skill in the art, without departing from the inventive concept of the application, they can also make Improvements, but these all belong to the protection scope of the present application.

As shown in FIG. 4 , FIG. 4 is a schematic structural diagram of an electronic device provided in an embodiment of the present application. The electronic device 4 includes a memory 401 , at least one processor 402 , a computer program 403 stored in the memory 401 and executable on the at least one processor 402 , and at least one communication bus 404 .

Those skilled in the art can understand that the schematic diagram shown in FIG. 4 is only an example of the electronic device 4, and does not constitute a limitation to the electronic device 4, and may include more or less components than those shown in the illustration, or a combination Certain components, or different components, for example, the electronic device 4 may also include input and output devices, network access devices, and the like.

The at least one processor 402 can be a central processing unit (Central Processing Unit, CPU), and can also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC ), Field-Programmable Gate Array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The at least one processor 402 can be a microprocessor or the at least one processor 402 can also be any conventional processor, etc., the at least one processor 402 is the control center of the electronic device 4, using various interfaces and lines Connect the various parts of the whole electronic equipment 4.

The memory 401 can be used to store the computer program 403, and the at least one processor 402 runs or executes the computer program 403 stored in the memory 401 and calls the data stored in the memory 401, Various functions of the electronic device 4 are realized. The memory 401 It can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function (such as a sound playback function, an image playback function, etc.), etc.; The use of the device 4 creates data such as audio data, and the like. In addition, the memory 401 may include non-volatile memory, such as hard disk, internal memory, plug-in hard disk, smart memory card (Smart Media Card, SMC), secure digital (Secure Digital, SD) card, flash memory card ( Flash Card), at least one disk memory device, flash memory device, or other non-volatile solid-state memory device.

If the integrated modules/units of the electronic device 4 are realized in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, all or part of the processes in the methods of the above embodiments of the present application can also be completed by instructing related hardware through computer programs, and the computer programs can be stored in a computer-readable storage medium. When the computer program is executed by the processor, it can realize the steps of the above-mentioned various method embodiments. Wherein, the computer program includes computer program code, and the computer program code may be in the form of source code, object code, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, flash drive, removable hard disk, magnetic disk, optical disk, computer memory, and read-only memory (Read-Only Memory) , ROM).

It will be apparent to those skilled in the art that the present application is not limited to the details of the exemplary embodiments described above, but that the present application can be implemented in other specific forms without departing from the spirit or essential characteristics of the present application. Therefore, no matter from any point of view, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the application is defined by the appended claims rather than the above description, so it is intended to All changes within the meaning and range of equivalents of the elements are embraced in this application. Any attached reference mark in a claim shall not be deemed to limit the claim to which it relates.

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application without limitation. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be Make modifications or equivalent replacements without departing from the spirit and scope of the technical solutions of the present application.

101-106: Steps

Claims (9)

A three-dimensional object detection method applied to electronic equipment, wherein the three-dimensional object detection method includes: acquiring a detection image and a depth image corresponding to the detection image, wherein the depth image is obtained by a depth camera; Inputting the detection image into the trained target detection model, using the target detection model to determine the object category of the object in the detection image and the two-dimensional boundary frame of the object, including: using the full range of the target detection model The connection layer processes the eigenvalue data to obtain a plurality of candidate two-dimensional border frames of objects in the training image, and performs non-maximum value suppression operations on the plurality of candidate two-dimensional border frames to obtain the training image The two-dimensional border frame and the object category of the object in the object, the non-maximum value suppression operation includes: according to the score of the candidate two-dimensional border frame, sorting a plurality of candidate two-dimensional border frames, and selecting the candidate two-dimensional border frame with the highest score frame; traverse other candidate two-dimensional border frames, calculate the intersection and union ratio between other candidate two-dimensional border frames and the selected candidate two-dimensional border frame, and delete the candidate two-dimensional border frame corresponding to the intersection and union ratio greater than the preset threshold; judge In addition to the selected candidate two-dimensional border frame, whether there are other candidate two-dimensional border frames; according to the object category, determine the object model corresponding to the object and the object model corresponding to the object from the three-dimensional object model library The three-dimensional border frame corresponding to the model includes: pre-establishing a three-dimensional object model library, wherein the three-dimensional object model library includes a plurality of object models corresponding to different object categories and a three-dimensional border frame corresponding to each object model; according to the Depth image and the two-dimensional border frame, calculating the point cloud data of the object framed by the two-dimensional border frame and the distance from the depth camera to the object model; according to the object model and the point cloud data to determine the rotation angle of the object model; The position of the object model in the three-dimensional space is determined according to the distance from the depth camera to the object model, the rotation angle, and the three-dimensional boundary frame. The three-dimensional object detection method according to claim 1, wherein, according to the depth image and the two-dimensional border frame, the point cloud data and the depth of the object framed by the two-dimensional border frame are calculated The distance from the camera to the object model includes: obtaining the depth value and coordinates of the object framed by the two-dimensional border frame according to the depth image, and determining the object from the depth camera to the object according to the depth value The distance of the model; obtain the point cloud data according to the coordinates and the internal and external parameter matrix transformation formula of the depth camera, and the internal and external parameter matrix transformation formula is:

Wherein (x, y, z) is the world coordinate, which is used to represent the point cloud of a primitive coordinate, f is the focal length, D is the depth value, (x ₁ , y ₁ ) is the frame selected by the two-dimensional border The primitive coordinates of any primitive point in the coordinate set of the object. The three-dimensional object detection method according to claim 1, wherein the determining the rotation angle of the object model according to the object model and the point cloud data includes: obtaining the object outline according to the point cloud data the first point cloud data; converting the object model into a second point cloud data; performing point cloud matching on the first point cloud data and the second point cloud data to obtain the rotation angle of the object model. The three-dimensional target detection method as claimed in item 1, wherein the method also includes: obtaining training images; building a target detection model based on You Only Look Once (YOLO) network; The training image is input into the target detection model for training, and the feature value data of the training image is obtained after the alternate processing of convolution and mean value pooling by the target detection model; using the target detection model The fully connected layer processes the feature value data to obtain the two-dimensional border frame and object category of the object in the training image, and adjusts the parameters of the target detection model to minimize the loss function to obtain the training completed Object detection model. The 3D object detection method according to claim 1, wherein the method further includes: establishing the 3D object model library, wherein the 3D object model library includes multiple object models corresponding to different object categories and each A three-dimensional border frame corresponding to each object model, the three-dimensional border frame includes the length, width and height corresponding to each object category. The three-dimensional object detection method according to claim 1, wherein, according to the distance from the depth camera to the object model, the rotation angle, and the three-dimensional bounding box, determine the object model in three-dimensional space The position includes: determining the direction of the object model in the three-dimensional space according to the rotation angle; according to the direction of the object model in the three-dimensional space, the distance from the depth camera to the object model and the The 3D bounding box of the object model determines the position of the object model in the 3D space. The three-dimensional object detection method according to claim 6, wherein the method further includes: using the position of the object model in the three-dimensional space as the position of the object in the three-dimensional space, outputting the object category and the The position of an object in three-dimensional space. An electronic device, wherein the electronic device includes a processor and a memory, and the processor is used to execute a computer program stored in the memory to realize the three-dimensional object detection method described in any one of claims 1 to 7 . A computer-readable storage medium, wherein the computer-readable storage medium stores at least one instruction, and when the at least one instruction is executed by a processor, the three-dimensional object detection as described in any one of claims 1 to 7 is realized method.

Images