Target detection method and device, electronic equipment and storage medium

By performing secondary detection after a single-stage model, using a small network to improve the accuracy of object detection, the problems of insufficient accuracy of single-stage model and excessive computing power of multi-stage model are solved, and deployment and computing are realized under different algorithm platforms Strive to save.

CN119006793AInactive Publication Date: 2024-11-22大卓智能科技有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411146195.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-20
Publication Date
2024-11-22
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The detection accuracy of single-stage models is insufficient, and the computing power requirements of multi-stage models are too high, making it difficult to deploy under different algorithm platforms.

Method used

After a single-stage model, a small network is provided for secondary detection. By obtaining the point cloud data to be detected, input it to the first preset neural network model to obtain the preliminary detection results, and intercepting the point cloud data of the target object according to the results for coordinate conversion Then, input it to the second preset neural network model to improve detection accuracy and reduce the computing power expenditure required for training.

Benefits of technology

Without the need for secondary training of the first-stage model, it is economical and practical to improve the detection accuracy of the model, simplify model deployment, and reduce computing power requirements.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006793A_ABST
    Figure CN119006793A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of vehicles, in particular to a target detection method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining the cloud data of a to-be-detected point; inputting the point cloud data to be detected into a first preset neural network model to obtain a first detection result, intercepting point cloud data including a target object according to the detection result, and performing coordinate conversion on the point cloud data including the target object to obtain point cloud data after coordinate conversion; and inputting the point cloud data after coordinate conversion into a second preset neural network model to obtain a target detection result. Therefore, the problems that a single-stage model is insufficient in accuracy and a multi-stage model is too high in computing power requirement are solved, a small network is provided for secondary detection after the single-stage model, the detection accuracy of the model is improved on the premise that secondary training of the first-stage model is not needed, deployment of the model under different algorithm platforms is facilitated, and the method is suitable for popularization and application. And the calculation power cost generated by model training is reduced, and the method has economical efficiency and practicability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of vehicle technology, and in particular to a target detection method, device, electronic device and storage medium. Background Art

[0002] At present, in the field of target detection, single-stage neural network models are often used in consideration of running speed and efficiency, but the accuracy of single-stage network models is often not as good as that of multi-stage models.

[0003] Related technologies have proposed cascade networks such as Cascade RCNN, but the integrated multi-stage network requires high computing power and high platform requirements, making it difficult to deploy on different algorithm platforms, and its actual application is relatively limited. Summary of the invention

[0004] The present application provides a target detection method, device, electronic device and storage medium to solve the problems of insufficient accuracy of single-stage models and excessive computing power requirements of multi-stage models. A small network is provided after the single-stage model for secondary detection. Without the need for secondary training of the one-stage model, the detection accuracy of the model is improved, the model is convenient to deploy on different algorithm platforms, and the computing power cost generated by model training is reduced. The application is economical and practical.

[0005] The first aspect of the present application provides a target detection method, comprising the following steps:

[0006] Obtain the point cloud data to be detected;

[0007] Inputting the point cloud data to be detected into a first preset neural network model to obtain a first detection result, and based on the first detection result, intercepting the point cloud data containing the target object, and performing coordinate transformation on the point cloud data containing the target object to obtain point cloud data after coordinate transformation;

[0008] The point cloud data after the coordinate transformation is input into a second preset neural network model to obtain a target detection result.

[0009] Optionally, in some embodiments, before the point cloud data to be detected is input into the first preset neural network model to obtain the first detection result, the method includes:

[0010] Obtain target point cloud dataset;

[0011] Divide the target point cloud dataset into a training set and a test set,

[0012] The training set is labeled to obtain a labeling result, and a regression target interval is determined according to the labeling result, and according to the regression target interval, the training set is input into an initial neural network model to obtain a trained neural network model, and the trained neural network model is tested using test set data to obtain a test result;

[0013] The trained neural network model is optimized using the test results to obtain a first preset neural network model.

[0014] Optionally, in some embodiments, inputting the training set into an initial neural network model to obtain a trained neural network model further includes:

[0015] Determine an actual bounding box of the training set according to the labeling result;

[0016] Determining a predicted bounding box of the training set according to an output result of the neural network model;

[0017] A target loss function is determined according to the actual bounding box and the predicted bounding box, and the target loss function is used to adjust the parameters of the initial neural network model to train the initial neural network model.

[0018] Optionally, in some embodiments, adjusting the parameters of the initial neural network model using the target loss function includes:

[0019] Determining the number of training times of the initial neural network model;

[0020] Determine whether the training times reaches a preset threshold, and if so, output the trained neural network model.

[0021] Optionally, in some embodiments, the objective loss function is wings loss.

[0022] A second aspect of the present application provides a target detection device, including:

[0023] An acquisition module is used to acquire point cloud data to be detected;

[0024] A first detection module, used for inputting the point cloud data to be detected into a first preset neural network model to obtain a first detection result, and according to the first detection result, intercepting the point cloud data containing the target object, and performing coordinate transformation on the point cloud data containing the target object to obtain point cloud data after coordinate transformation;

[0025] The second detection module is used to input the point cloud data after the coordinate transformation into a second preset neural network model to obtain a target detection result.

[0026] Optionally, in some embodiments, before the point cloud data to be detected is input into the first preset neural network model to obtain the first detection result, the first detection module includes:

[0027] An acquisition unit, used for acquiring a target point cloud data set;

[0028] A division unit is used to divide the target point cloud data set into a training set and a test set,

[0029] A training unit, used to label the training set to obtain a labeling result, and determine a regression target interval according to the labeling result, input the training set into an initial neural network model to obtain a trained neural network model according to the regression target interval, and test the trained neural network model using test set data to obtain a test result;

[0030] An optimization unit is used to optimize the trained neural network model using the test results to obtain a first preset neural network model.

[0031] Optionally, in some embodiments, the training unit further includes:

[0032] a labeling subunit, configured to determine an actual bounding box of the training set according to the labeling result;

[0033] A prediction subunit, configured to determine a prediction bounding box of the training set according to an output result of the neural network model;

[0034] A training subunit is used to determine a target loss function according to the actual bounding box and the predicted bounding box, and use the target loss function to adjust the parameters of the initial neural network model to train the initial neural network model.

[0035] Optionally, in some embodiments, the training subunit includes:

[0036] A determination component, used to determine the number of training times of the initial neural network model;

[0037] The judging component is used to judge whether the number of training times reaches a preset threshold, and if the number of training times reaches the preset threshold, output the trained neural network model.

[0038] Optionally, in some embodiments, the objective loss function is wings loss.

[0039] The third aspect of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the target detection method as described in the above embodiment.

[0040] The fourth aspect of the present application provides a computer-readable storage medium on which a computer program is stored. The program is executed by a processor to implement the target detection method as described in the above embodiment.

[0041] Thus, the first detection result is obtained by acquiring the point cloud data to be detected and inputting the point cloud data to be detected into the first preset neural network model, and according to the detection result, the point cloud data containing the target object is intercepted, and the point cloud data containing the target object is coordinate-converted to obtain the point cloud data after coordinate conversion, and the point cloud data after coordinate conversion is input into the second preset neural network model to obtain the target detection result. Thus, the problem of insufficient accuracy of the single-stage model and excessively high computing power requirements of the multi-stage model is solved, and a small network is provided after the single-stage model for secondary detection. Under the premise of not requiring secondary training of the first-stage model, the detection accuracy of the model is improved, the model is convenient to be deployed on different algorithm platforms, and the computing power cost generated by model training is reduced, which is economical and practical.

[0042] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0044] Figure 1 A flowchart of a target detection method provided according to an embodiment of the present application;

[0045] Figure 2 A schematic diagram of the principle of a target detection method provided according to an embodiment of the present application;

[0046] Figure 3 A block diagram of a target detection device provided according to an embodiment of the present application;

[0047] Figure 4 It is a schematic diagram of the structure of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0048] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0049] The target detection method, device, electronic device and storage medium of the embodiment of the present application are described below with reference to the accompanying drawings. In view of the problem that the accuracy of the single-stage model is insufficient and the computing power requirement of the multi-stage model is too high mentioned in the above background technology, the present application provides a target detection method, in which the point cloud data to be detected is obtained, and the point cloud data to be detected is input into a first preset neural network model to obtain a first detection result, and according to the detection result, the point cloud data containing the target object is intercepted, and the point cloud data containing the target object is coordinate-converted to obtain the point cloud data after the coordinate conversion, and the point cloud data after the coordinate conversion is input into a second preset neural network model to obtain the target detection result. In this way, the problem that the accuracy of the single-stage model is insufficient and the computing power requirement of the multi-stage model is too high is solved, and a small network is provided after the single-stage model for secondary detection. Under the premise that the secondary training of the one-stage model is not required, the detection accuracy of the model is improved, the model is convenient to be deployed under different algorithm platforms, and the computing power cost generated by model training is reduced, which is economical and practical.

[0050] Specifically, Figure 1 A flowchart of a target detection method provided in an embodiment of the present application.

[0051] like Figure 1 As shown, the target detection method includes the following steps:

[0052] In step S101, the point cloud data to be detected is obtained.

[0053] Specifically, the embodiments of the present application can collect point cloud data to be detected through sensors such as laser radar or cameras to perform 3D target detection on the point cloud data to be detected. The point cloud data to be detected can be point cloud data collected in real time or point cloud data stored in any terminal.

[0054] In step S102, the point cloud data to be detected is input into a first preset neural network model to obtain a first detection result, and based on the first detection result, the point cloud data containing the target object is intercepted, and the point cloud data containing the target object is coordinate transformed to obtain the point cloud data after coordinate transformation.

[0055] Among them, the first preset neural network model can be any 3D neural network model.

[0056] The embodiment of the present application can perform target detection through a two-stage neural network. Preliminary detection is performed through the neural network model of the first stage, and the detection results are input into the second stage neural network for optimization. Point cloud features are extracted by directly sampling point cloud data, thereby improving detection accuracy and reducing computing power requirements.

[0057] In this step, the embodiment of the present application performs the first stage of target detection. It can be understood that converting the point cloud data into a coordinate system centered on the detection frame can more accurately analyze the size and shape of the target object and more easily identify the key features of the target object. Therefore, after obtaining the point cloud data containing the target object, the embodiment of the present application can convert the coordinate system of the point cloud data containing the target object into a detection coordinate system.

[0058] Specifically, the embodiment of the present application can input the point cloud data to be detected into a pre-trained first preset neural network model to obtain a first detection result, the first detection result includes the target object, and the point cloud of the target area including the target object is intercepted, and the coordinate system of the point cloud data including the target object is converted into a detection coordinate system, that is, a coordinate system with the detection result as the center point and the heading direction of the detection frame as the positive X direction, and then obtain the point cloud data after the coordinate conversion. It should be noted that for the first detection result, the point cloud within a certain range of the target frame is intercepted, and the target frame can be dynamically adjusted according to the size of the target object.

[0059] Optionally, in some embodiments, before the point cloud data to be detected is input into the first preset neural network model to obtain the first detection result, it includes: obtaining a target point cloud data set; dividing the target point cloud data set into a training set and a test set, annotating the training set to obtain an annotation result, and determining a regression target interval based on the annotation result, and according to the regression target interval, inputting the training set into the initial neural network model to obtain a trained neural network model, and using the test set data to test the trained neural network model to obtain a test result; and using the test result to optimize the trained neural network model to obtain the first preset neural network model.

[0060] It is understandable that before performing target detection through the first preset neural network model, any initial neural network model suitable for target detection can be trained. There are many neural network models suitable for target detection in the relevant technology, which are not specifically limited here.

[0061] Specifically, the embodiments of the present application can collect 3D point cloud data for training and testing a neural network model to obtain a target point cloud data set; divide the target point cloud data set into two parts, a training set for model learning, and a test set for evaluating model performance to ensure that the model has good generalization capabilities; annotate the point cloud data in the training set, typically including a bounding box of the marked object, a category label, etc., and annotation is a key step in supervised learning, and the initial neural network model requires this information to learn to identify and classify objects; determine the target object range predicted by the model, for example, the position and size of the target object bounding box, based on the annotation results; then use the annotated training set data to train the initial neural network model, and use the test set to evaluate the performance of the trained model to obtain test results, and based on the test results, adjust and optimize the model to obtain a first preset neural network model.

[0062] Optionally, in some embodiments, the training set is input into the initial neural network model to obtain a trained neural network model, which also includes: determining an actual bounding box of the training set based on the labeling results; determining a predicted bounding box of the training set based on the output results of the neural network model; determining a target loss function based on the actual bounding box and the predicted bounding box, and using the target loss function to adjust the parameters of the initial neural network model to train the initial neural network model.

[0063] Among them, the target loss function is wings loss.

[0064] Specifically, based on the annotation results, the actual bounding box of each target in the point cloud data is determined, and the neural network model is used to predict the training set to obtain the predicted bounding box of each target. According to the difference between the actual bounding box and the predicted bounding box, the target loss function is defined. The gap between the predicted result and the actual result is measured by the target loss function. During the training process, the gradient of the target loss function is used to update the parameters of the model through the back propagation algorithm to make the predicted bounding box as close to the actual bounding box as possible.

[0065] Optionally, in some embodiments, the parameters of the initial neural network model are adjusted using the target loss function, including: determining the number of training times of the initial neural network model; determining whether the number of training times reaches a preset threshold, and outputting the trained neural network model if the number of training times reaches the preset threshold.

[0066] The preset threshold may be pre-set by a user, obtained through a limited number of experiments, or obtained through a limited number of computer simulations, and is not specifically limited here.

[0067] It is understandable that the initial neural network model can be continuously iteratively optimized on the training set to improve the generalization ability. The embodiment of the present application can set a preset threshold and determine the number of training times to ensure that the model iteratively learns the characteristics of the training data, while preventing excessive model training and thus avoiding overfitting.

[0068] The embodiment of the present application can monitor the changes in the loss function during the training process. If the loss no longer decreases significantly, it may mean that the model has converged and continued training may not bring performance improvement. The model with the best performance under different training times can be selected, the training times of the initial neural network model can be determined, and when the training times reach a preset threshold, the trained model can be output.

[0069] In step S103, the point cloud data after coordinate conversion is input into a second preset neural network model to obtain a target detection result.

[0070] Specifically, combined Figure 2 As shown, in the embodiment of the present application, for the point cloud data to be detected, the output is first obtained through a first preset neural network, and then the point cloud in the area is intercepted and converted into a coordinate system centered on the detection frame. After that, the point cloud is input into a second preset neural network. The network can output the result of the secondary refinement in the detection coordinate system, and perform inverse operations to obtain the detection result of the detection frame in the lidar coordinate system.

[0071] According to the target detection method proposed in the embodiment of the present application, the point cloud data to be detected is obtained, and the point cloud data to be detected is input into the first preset neural network model to obtain the first detection result, and according to the detection result, the point cloud data containing the target object is intercepted, and the point cloud data containing the target object is coordinate-converted to obtain the point cloud data after coordinate conversion, and the point cloud data after coordinate conversion is input into the second preset neural network model to obtain the target detection result. In this way, the problem of insufficient accuracy of the single-stage model and excessively high computing power requirements of the multi-stage model is solved, and a small network is provided after the single-stage model for secondary detection. Under the premise of not requiring secondary training of the one-stage model, the detection accuracy of the model is improved, the model is convenient to be deployed on different algorithm platforms, and the computing power cost generated by model training is reduced, which is economical and practical.

[0072] Next, the target detection device proposed according to the embodiment of the present application is described with reference to the accompanying drawings.

[0073] Figure 3 It is a block diagram of a target detection device according to an embodiment of the present application.

[0074] like Figure 3 As shown, the target detection device 10 includes: an acquisition module 100 , a first detection module 200 and a second detection module 300 .

[0075] The acquisition module 100 is used to acquire the point cloud data to be detected.

[0076] The first detection module 200 is used to input the point cloud data to be detected into the first preset neural network model to obtain a first detection result, and based on the first detection result, intercept the point cloud data containing the target object, and perform coordinate transformation on the point cloud data containing the target object to obtain the point cloud data after coordinate transformation.

[0077] The second detection module is used to input the point cloud data after coordinate conversion into a second preset neural network model to obtain a target detection result.

[0078] Optionally, in some embodiments, before the point cloud data to be detected is input into the first preset neural network model to obtain the first detection result, the first detection module 200 includes: an acquisition unit, a division unit, a training unit and an optimization unit.

[0079] The acquisition unit is used to acquire the target point cloud data set.

[0080] The partitioning unit is used to divide the target point cloud dataset into a training set and a test set.

[0081] The training unit is used to label the training set to obtain the labeling results, and determine the regression target interval according to the labeling results. According to the regression target interval, the training set is input into the initial neural network model to obtain the trained neural network model, and the trained neural network model is tested with the test set data to obtain the test results.

[0082] The optimization unit is used to optimize the trained neural network model using the test results to obtain a first preset neural network model.

[0083] Optionally, in some embodiments, the training unit further includes: a labeling subunit, a prediction subunit and a training subunit.

[0084] Among them, the labeling subunit is used to determine the actual bounding box of the training set according to the labeling results.

[0085] The prediction subunit is used to determine the prediction bounding box of the training set based on the output results of the neural network model.

[0086] The training subunit is used to determine the target loss function according to the actual bounding box and the predicted bounding box, and use the target loss function to adjust the parameters of the initial neural network model to train the initial neural network model.

[0087] Optionally, in some embodiments, the training subunit includes: determining a subcomponent and judging a subcomponent.

[0088] Among them, the determination component is used to determine the number of training times of the initial neural network model.

[0089] The judgment component is used to judge whether the number of training times reaches a preset threshold. If the number of training times reaches the preset threshold, the trained neural network model is output.

[0090] Optionally, in some embodiments, the objective loss function is wings loss.

[0091] It should be noted that the aforementioned explanation of the target detection method embodiment is also applicable to the target detection device of this embodiment, and will not be repeated here.

[0092] According to the target detection device proposed in the embodiment of the present application, the first detection result is obtained by acquiring the point cloud data to be detected and inputting the point cloud data to be detected into the first preset neural network model, and according to the detection result, the point cloud data containing the target object is intercepted, and the point cloud data containing the target object is coordinate-converted to obtain the point cloud data after coordinate conversion, and the point cloud data after coordinate conversion is input into the second preset neural network model to obtain the target detection result. In this way, the problem of insufficient accuracy of the single-stage model and excessively high computing power requirements of the multi-stage model is solved, and a small network is provided after the single-stage model for secondary detection. Under the premise of not requiring secondary training of the one-stage model, the detection accuracy of the model is improved, the model is convenient to be deployed on different algorithm platforms, and the computing power cost generated by model training is reduced, which is economical and practical.

[0093] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include:

[0094] Memory 401 , processor 402 , and a computer program stored in the memory 401 and executable on the processor 402 .

[0095] When the processor 402 executes the program, the target detection method provided in the above embodiment is implemented.

[0096] Furthermore, the electronic device further comprises:

[0097] The communication interface 403 is used for communication between the memory 401 and the processor 402 .

[0098] The memory 401 is used to store computer programs that can be executed on the processor 402 .

[0099] The memory 401 may include a high-speed RAM (Random Access Memory) memory, and may also include a non-volatile memory, such as at least one disk memory.

[0100] If the memory 401, the processor 402 and the communication interface 403 are implemented independently, the communication interface 403, the memory 401 and the processor 402 can be connected to each other through a bus and communicate with each other. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 4 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0101] Optionally, in a specific implementation, if the memory 401, the processor 402 and the communication interface 403 are integrated on a chip, the memory 401, the processor 402 and the communication interface 403 can communicate with each other through an internal interface.

[0102] The processor 402 may be a CPU (Central Processing Unit), or an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present application.

[0103] An embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the above target detection method is implemented.

[0104] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms are not necessarily directed to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0105] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0106] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or more executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0107] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiment, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array, a field programmable gate array, etc.

[0108] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0109] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A target detection method, It is characterized in that The following steps are involved: Obtain the point cloud data to be detected; Inputting the point cloud data to be detected into a first preset neural network model to obtain a first detection result, and based on the first detection result, intercepting the point cloud data containing the target object, and performing coordinate transformation on the point cloud data containing the target object to obtain point cloud data after coordinate transformation; The point cloud data after the coordinate transformation is input into a second preset neural network model to obtain a target detection result.

2. The method according to claim 1, It is characterized in that Before the point cloud data to be detected is input into the first preset neural network model to obtain the first detection result, it includes: Obtain target point cloud dataset; Divide the target point cloud dataset into a training set and a test set, The training set is labeled to obtain a labeling result, and a regression target interval is determined according to the labeling result, and according to the regression target interval, the training set is input into an initial neural network model to obtain a trained neural network model, and the trained neural network model is tested using test set data to obtain a test result; The trained neural network model is optimized using the test results to obtain a first preset neural network model.

3. The method according to claim 2, It is characterized in that The step of inputting the training set into the initial neural network model to obtain a trained neural network model further includes: Determine an actual bounding box of the training set according to the labeling result; Determining a predicted bounding box of the training set according to an output result of the neural network model; A target loss function is determined according to the actual bounding box and the predicted bounding box, and the target loss function is used to adjust the parameters of the initial neural network model to train the initial neural network model.

4. The method according to claim 3, It is characterized in that Using the target loss function to adjust the parameters of the initial neural network model includes: Determining the number of training times of the initial neural network model; Determine whether the training times reaches a preset threshold, and if so, output the trained neural network model.

5. The method according to claim 3, It is characterized in that The target loss function is wings loss.

6. A target detection device, It is characterized in that include: An acquisition module is used to acquire point cloud data to be detected; A first detection module, used for inputting the point cloud data to be detected into a first preset neural network model to obtain a first detection result, and according to the detection result, intercepting the point cloud data containing the target object, and performing coordinate transformation on the point cloud data containing the target object to obtain point cloud data after coordinate transformation; The second detection module is used to input the point cloud data after the coordinate transformation into a second preset neural network model to obtain a target detection result.

7. The device according to claim 6, It is characterized in that Before the point cloud data to be detected is input into the first preset neural network model to obtain the first detection result, the first detection module includes: An acquisition unit, used for acquiring a target point cloud data set; A division unit is used to divide the target point cloud data set into a training set and a test set, A training unit, used to label the training set to obtain a labeling result, and determine a regression target interval according to the labeling result, input the training set into an initial neural network model to obtain a trained neural network model according to the regression target interval, and test the trained neural network model using test set data to obtain a test result; An optimization unit is used to optimize the trained neural network model using the test results to obtain a first preset neural network model.

8. The device according to claim 7, It is characterized in that The training unit further includes: a labeling subunit, configured to determine an actual bounding box of the training set according to the labeling result; A prediction subunit, configured to determine a prediction bounding box of the training set according to an output result of the neural network model; A training subunit is used to determine a target loss function according to the actual bounding box and the predicted bounding box, and use the target loss function to adjust the parameters of the initial neural network model to train the initial neural network model.

9. An electronic device, It is characterized in that include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the target detection method according to any one of claims 1 to 5.

10. A computer-readable storage medium having a computer program stored thereon, It is characterized in that The program is executed by a processor to implement the target detection method as described in any one of claims 1 to 5.