Monocular three-dimensional target detection method and system based on nested reasoning chain
By constructing nested inference chains and combining dynamic aggregation and error correction mechanisms, the problems of inconsistent inference results and error accumulation in monocular 3D target detection are solved, achieving higher geometric consistency and detection accuracy, which is suitable for monocular vision applications.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANKAI UNIV
- Filing Date
- 2026-01-16
- Publication Date
- 2026-04-28
AI Technical Summary
In existing monocular 3D target detection methods, parallel prediction of 3D geometric attributes leads to inconsistent inference results. Errors are prone to accumulate and propagate during multi-stage inference, and multi-stage feature information is not fully utilized, lacking an effective correction mechanism.
A nested inference chain-based approach is adopted, which constructs a multi-stage nested inference chain to predict step by step according to the dependencies between three-dimensional geometric attributes. During step-by-step prediction, the result of the previous stage is used as the condition input for the next stage. At the same time, dynamic aggregation and error correction mechanisms are introduced, and correction is performed in combination with the original context information.
It improves the geometric consistency and accuracy of 3D target detection results, enhances the robustness of the detection process, and requires no additional sensors, making it suitable for a variety of monocular vision application scenarios.
Smart Images
Figure CN121937987A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of target detection technology, specifically relating to a monocular three-dimensional target detection method and system based on nested inference chains. Background Technology
[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.
[0003] With the development of technologies such as autonomous driving, intelligent transportation, and robotic perception, 3D target detection, as one of the important foundational technologies for environmental perception, has received widespread attention. Compared with 3D target detection schemes that rely on LiDAR or multi-sensor fusion, 3D target detection methods based on monocular images have advantages such as low cost and flexible deployment, and therefore have significant research value and application prospects.
[0004] Existing monocular 3D target detection methods typically require the simultaneous prediction of multiple 3D geometric attributes of a target from a 2D image, such as the target's 3D dimensions, spatial orientation, and spatial depth. Since monocular images themselves lack direct depth information, the prediction of these geometric attributes often involves significant uncertainty. To improve detection accuracy, existing technologies typically employ parallel prediction, i.e., simultaneously regressing the aforementioned multiple geometric parameters based on the same feature representation.
[0005] However, the aforementioned parallel prediction methods fail to fully consider the inherent dependencies between various 3D geometric attributes. For example, the target's 3D dimensions, spatial orientation, and spatial depth are geometrically interrelated; independent prediction can easily lead to inconsistencies between inference results, thus affecting overall detection accuracy. Furthermore, in multi-stage prediction or cascaded structures, existing methods often directly pass the prediction results from the previous stage to the next. If errors exist in intermediate predictions, these errors will be continuously amplified and propagated, reducing the system's robustness.
[0006] On the other hand, existing methods, when utilizing multi-stage feature information, typically employ simple concatenation or fixed-weight fusion, making it difficult to adaptively adjust the contribution of different inference stages to the final detection result. Furthermore, most methods fail to effectively backtrack to the original image context information during subsequent inference processes to correct intermediate prediction results, further limiting the improvement of detection performance. Summary of the Invention
[0007] This invention addresses the common problems in existing monocular 3D target detection methods, such as inconsistent inference results due to parallel prediction of 3D geometric attributes, easy accumulation and propagation of errors during multi-stage inference, insufficient utilization of multi-stage feature information, and lack of effective correction mechanisms. It proposes a monocular 3D target detection method and system based on nested inference chains. This invention can improve the geometric consistency, accuracy, and robustness of monocular 3D target detection results without introducing additional sensors.
[0008] According to some embodiments, the present invention adopts the following technical solution: A monocular 3D target detection method based on nested inference chains includes the following steps: Acquire a monocular image and extract feature representations for target detection based on the monocular image; Based on the feature representation, a multi-stage nested inference chain is constructed. According to the preset geometric dependency relationship, the three-dimensional size parameters, spatial orientation parameters and spatial depth parameters of the target are predicted step by step. In the step-by-step prediction, the prediction result of the previous stage is used as the condition input for the prediction of the next stage. In the multi-stage nested inference chain, the intermediate features generated in each inference stage are dynamically aggregated to obtain the fused feature representation; By using an error correction mechanism, the fused feature representation is combined with the original context information of the monocular image to correct the intermediate prediction results and suppress error propagation. Based on the corrected prediction results, the three-dimensional detection results of the target are obtained.
[0009] As an alternative implementation, the process of extracting feature representations for target detection based on the monocular image includes extracting multi-level feature information from the monocular image using a pre-trained neural network model and generating original contextual features for subsequent 3D target detection. The original contextual features serve as shared contextual information for each inference stage in the nested inference chain.
[0010] As an optional implementation, the process of constructing a multi-stage nested inference chain includes: the nested inference chain includes an initial query vector and multiple sequentially connected inference stages. The initial query vector is used to initialize the model of the target and to predict the two-dimensional bounding box. In the first inference stage, the initial query vector is passed through the identity module and then sent to the correction module for processing to obtain the query vector output by the first stage, which is used to predict the three-dimensional size information of the target. In subsequent inference stages, each inference stage is based on the output result of the previous stage.
[0011] As a further defined implementation method, the first Query vectors for each inference stage Determined by the query vector from the previous stage and the original context features, the update relationship is expressed as follows:
[0012] in, Indicates aggregation operation, Indicates a calibration operation. This represents the set of query vectors output during the preliminary inference stage. The original context features are used; each inference stage completes the inference step by step according to the dependency relationship between the three-dimensional geometric attributes, forming a nested inference chain structure, which enhances the consistency between the prediction results of different geometric attributes.
[0013] As an alternative implementation, the process of dynamically aggregating the intermediate features generated at each inference stage to obtain the fused feature representation includes: obtaining the query vector of the current stage and multiple query vectors output by the preceding inference stage, and generating dynamic weights through a linear layer to characterize the contribution of different inference stages. The dynamic weights are used to reflect the relative contribution of different inference stages to the inference task of the current stage; and fusing the weighted multiple query vectors to obtain the aggregated query vector.
[0014] As an alternative implementation, the process of correcting intermediate prediction results to suppress error propagation includes: using aggregated features from the current inference stage. As input, and incorporate the extracted original contextual features. The following steps are performed sequentially: normalization, multi-head attention module, and feedforward neural network, as shown below:
[0015] in, This indicates normalization processing. This represents the attention calculation process. This represents a feedforward neural network.
[0016] As an alternative implementation, the three-dimensional detection results include the target's three-dimensional dimensions, spatial orientation, and spatial depth information.
[0017] A monocular 3D target detection system based on nested inference chains includes: The feature extraction module is configured to acquire a monocular image and extract feature representations for target detection based on the monocular image; The nested inference module is configured to construct a multi-stage nested inference chain based on the feature representation, and to predict the three-dimensional size parameters, spatial orientation parameters and spatial depth parameters of the target step by step according to the preset geometric dependency relationship. In the step-by-step prediction, the prediction result of the previous stage is used as the condition input for the prediction of the next stage. The dynamic aggregation module is configured to dynamically aggregate intermediate features generated at each inference stage in the multi-stage nested inference chain to obtain a fused feature representation. The correction module is configured to use an error correction mechanism to combine the fused feature representation with the original context information of the monocular image to correct the intermediate prediction results in order to suppress error propagation. The output module is configured to obtain the three-dimensional detection results of the target based on the corrected prediction results.
[0018] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the steps in the above method.
[0019] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the steps in the method described above.
[0020] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention constructs a multi-stage nested inference chain and predicts parameters such as the three-dimensional size, spatial orientation, and spatial depth of the target step by step according to the dependency relationship between three-dimensional geometric attributes. This avoids the problem of inconsistent inference results caused by independent prediction of multiple geometric parameters, thereby improving the geometric consistency of three-dimensional target detection results.
[0021] This invention introduces a dynamic aggregation mechanism in the multi-stage inference process to adaptively fuse feature information from different inference stages, making full use of the multi-stage inference results and further improving detection accuracy. In addition, by combining the contextual information of the original monocular image to review and correct intermediate prediction results, the accumulation and propagation of errors in the multi-stage inference process are effectively suppressed, enhancing the robustness of the detection process.
[0022] Without relying on additional sensors or complex calibration processes, this invention is applicable to a variety of monocular vision applications and has good practicality and promotional value.
[0023] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description
[0024] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.
[0025] Figure 1This is a schematic diagram of the overall structure of a monocular 3D target detection model based on nested inference chains in one embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of the correction module in one embodiment of the present invention; Figure 3 This is a schematic diagram of the aggregation module in one embodiment of the present invention. Detailed Implementation
[0026] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0027] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.
[0028] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.
[0029] Where there is no conflict, the embodiments and features described in this application may be combined with each other.
[0030] Example 1 As described in the background section, how to fully model the dependencies between 3D geometric attributes, make reasonable use of multi-stage inference information, and effectively suppress error propagation to improve the consistency and accuracy of detection results during monocular 3D target detection remains a technical problem that urgently needs to be solved in this field.
[0031] To address the aforementioned problems, this embodiment provides a monocular 3D target detection method based on nested inference chains, comprising the following steps: Acquire a monocular image and extract feature representations for target detection based on the monocular image; Based on the feature representation, a multi-stage nested inference chain is constructed. According to the preset geometric dependency relationship, the three-dimensional size parameters, spatial orientation parameters and spatial depth parameters of the target are predicted step by step. In the step-by-step prediction, the prediction result of the previous stage is used as the condition input for the prediction of the next stage. In the multi-stage nested inference chain, the intermediate features generated in each inference stage are dynamically aggregated to obtain the fused feature representation; By using an error correction mechanism, the fused feature representation is combined with the original context information of the monocular image to correct the intermediate prediction results and suppress error propagation. Based on the corrected prediction results, the three-dimensional detection results of the target are obtained.
[0032] In this embodiment, the process of extracting feature representations for target detection based on the monocular image includes extracting multi-level feature information from the monocular image using a pre-trained neural network model and generating original context features for subsequent 3D target detection. The original context features serve as shared context information for each inference stage in the nested inference chain.
[0033] In this embodiment, the process of constructing a multi-stage nested inference chain includes: the nested inference chain includes an initial query vector and multiple sequentially connected inference stages. The initial query vector is used to initialize the model of the target and to predict the two-dimensional bounding box. In the first inference stage, the initial query vector is passed through the identity module and then sent to the correction module for processing to obtain the query vector output by the first stage, which is used to predict the three-dimensional size information of the target. In subsequent inference stages, each inference stage is based on the output result of the previous stage.
[0034] No. Query vectors for each inference stage Determined by the query vector from the previous stage and the original context features, the update relationship is expressed as follows:
[0035] in, Indicates aggregation operation, Indicates a calibration operation. This represents the set of query vectors output during the preliminary inference stage. The original context features are used; each inference stage completes the inference step by step according to the dependency relationship between the three-dimensional geometric attributes, forming a nested inference chain structure, which enhances the consistency between the prediction results of different geometric attributes.
[0036] In this embodiment, the process of dynamically aggregating the intermediate features generated in each inference stage to obtain the fused feature representation includes: obtaining the query vector of the current stage and multiple query vectors output by the preceding inference stage, and generating dynamic weights through a linear layer to characterize the contribution of different inference stages. The dynamic weights are used to reflect the relative contribution of different inference stages to the inference task of the current stage; and fusing the weighted multiple query vectors to obtain the aggregated query vector.
[0037] In this embodiment, the process of correcting intermediate prediction results to suppress error propagation includes: using aggregated features from the current inference stage. As input, and incorporate the extracted original contextual features. The following steps are performed sequentially: normalization, multi-head attention module, and feedforward neural network, as shown below:
[0038] in, This indicates normalization processing. This represents the attention calculation process. This represents a feedforward neural network.
[0039] In this embodiment, the three-dimensional detection results include the target's three-dimensional dimensions, spatial orientation, and spatial depth information.
[0040] Example 2 A monocular 3D target detection system based on nested inference chains, such as Figure 1 As shown, it includes: The feature extraction module is configured to acquire a monocular image and extract feature representations for target detection based on the monocular image; The nested inference module is configured to construct a multi-stage nested inference chain based on the feature representation, and to predict the three-dimensional size parameters, spatial orientation parameters and spatial depth parameters of the target step by step according to the preset geometric dependency relationship. In the step-by-step prediction, the prediction result of the previous stage is used as the condition input for the prediction of the next stage. The dynamic aggregation module is configured to dynamically aggregate intermediate features generated at each inference stage in the multi-stage nested inference chain to obtain a fused feature representation. The correction module is configured to use an error correction mechanism to combine the fused feature representation with the original context information of the monocular image to correct the intermediate prediction results in order to suppress error propagation. The output module is configured to obtain the three-dimensional detection results of the target based on the corrected prediction results.
[0041] Of course, in some embodiments, the detection method in Embodiment 1 can be implemented by a model. The overall model includes a feature extraction part and a nested inference chain part. The feature extraction part includes a backbone network and connected encoding and decoding structures, used to extract multi-level feature information from the input monocular image and generate raw contextual features for subsequent 3D object detection. These raw contextual features can serve as shared contextual information for each inference stage in the nested inference chain, denoted as... .
[0042] The nested inference chain is used to perform step-by-step inference on the two-dimensional and three-dimensional geometric attributes of the target based on the original context features. The nested inference chain consists of multiple inference stages connected in sequence, and the inference stages are associated with each other through conditional information.
[0043] like Figure 1 As shown, the nested inference chain includes an initial query vector and multiple sequentially connected inference stages. Let the initial query vector be... This is used for initial modeling of the target and is initially used for predicting the two-dimensional bounding box. In the first inference stage, the initial query vector is passed through the identity module and then sent to the correction module for processing to obtain the query vector output in the first stage. This is used to predict the target's three-dimensional dimensions. In subsequent inference stages, each inference stage is based on the output of the previous stage. Specifically, the... Query vectors for each inference stage Determined by the query vector from the previous stage and the original context features, the update relationship can be expressed as: ; in, Indicates aggregation operation, Indicates a calibration operation. This represents the set of query vectors output during the preceding inference phase.
[0044] In this way, each inference stage completes the inference step by step according to the dependency relationship between the three-dimensional geometric attributes, forming a nested inference chain structure, thereby enhancing the consistency between the prediction results of different geometric attributes.
[0045] like Figure 2 As shown, the correction module is used to review and correct intermediate features in the current inference stage by combining the original context features at each inference stage. In this embodiment, the correction module receives aggregated features from the current inference stage. As input, and incorporate the original contextual features from the feature extraction section. The correction module includes, in sequence, a normalization process, a multi-head attention module, and a feedforward neural network.
[0046] The processing procedure of the calibration module can be represented as follows:
[0047] in, This indicates normalization processing. This represents the attention calculation process. This represents a feedforward neural network. After the above processing, the output is a corrected query vector, which serves as the input for the next inference stage. By introducing a correction module, the preceding prediction results can be dynamically corrected during multi-stage inference, reducing the gradual accumulation of errors.
[0048] like Figure 3 As shown, the aggregation module is used to fuse the query vectors generated in the current inference stage with those generated in previous inference stages. In this embodiment, the aggregation module receives the query vector of the current stage and multiple query vectors output by previous inference stages, and generates dynamic weights to characterize the contribution of different inference stages through a linear layer.
[0049] Let the first The aggregation result of the inference phase is Its calculation method can be expressed as:
[0050] in, The weights are dynamic and satisfy the following conditions:
[0051] The dynamic weights reflect the relative contribution of different inference stages to the current inference task. Subsequently, the weighted query vectors are fused to obtain an aggregated query vector, which is used for subsequent processing in the current inference stage or as input to the correction module. This aggregation method fully utilizes the information generated during multi-stage inference, improving the stability and accuracy of the inference results.
[0052] After completing the step-by-step inference, dynamic aggregation, and correction processes at each inference stage, the model generates a 3D detection result of the target based on the query vector output from the final inference stage. The 3D detection result may include information such as the target's 3D dimensions, spatial orientation, and spatial depth.
[0053] As shown in Table 1 of the KITTI test set results, under the medium difficulty evaluation criteria, the proposed method achieves the best performance in both 3D detection accuracy and BEV detection accuracy. Specifically, the average 3D accuracy of the proposed method reaches 19.61%, significantly improving upon MonoDGP's 18.72% and MonoDETR's 16.47%. In terms of BEV average accuracy, the proposed method achieves 25.89%, also outperforming MonoDGP's 25.23% and MonoDETR's 22.11%. These results demonstrate that the proposed monocular 3D object detection method based on nested inference chains can more effectively model the dependencies between 3D geometric attributes and improve overall inference quality through multi-stage information aggregation and correction mechanisms, thus exhibiting superior detection performance on standard public datasets.
[0054] Table 1
[0055] Example 3 A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the steps in the method provided in Embodiment 1.
[0056] Example 4 An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the steps in the method provided in Embodiment 1.
[0057] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of one or more computer-usable storage media (including, but not limited to, disk storage, etc.) containing computer-usable program code. CD - ROM It takes the form of a computer program product implemented on (such as optical memory, etc.).
[0058] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0059] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0060] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0061] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made by those skilled in the art without creative effort within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A monocular 3D target detection method based on nested inference chains, characterized in that, Includes the following steps: Acquire a monocular image and extract feature representations for target detection based on the monocular image; Based on the feature representation, a multi-stage nested inference chain is constructed. According to the preset geometric dependency relationship, the three-dimensional size parameters, spatial orientation parameters and spatial depth parameters of the target are predicted step by step. In the step-by-step prediction, the prediction result of the previous stage is used as the condition input for the prediction of the next stage. In the multi-stage nested inference chain, the intermediate features generated in each inference stage are dynamically aggregated to obtain the fused feature representation; By using an error correction mechanism, the fused feature representation is combined with the original context information of the monocular image to correct the intermediate prediction results and suppress error propagation. Based on the corrected prediction results, the three-dimensional detection results of the target are obtained.
2. The monocular 3D target detection method based on nested inference chains as described in claim 1, characterized in that, The process of extracting feature representations for target detection based on the monocular image includes extracting multi-level feature information from the monocular image using a pre-trained neural network model and generating original contextual features for subsequent 3D target detection. The original contextual features serve as shared contextual information for each inference stage in the nested inference chain.
3. The monocular 3D target detection method based on nested inference chains as described in claim 1, characterized in that, The process of constructing a multi-stage nested inference chain includes: the nested inference chain includes an initial query vector and multiple sequentially connected inference stages. The initial query vector is used to initialize the model of the target and to predict the two-dimensional bounding box. In the first inference stage, the initial query vector is passed through the identity module and then sent to the correction module for processing to obtain the query vector output by the first stage, which is used to predict the three-dimensional size information of the target. In subsequent inference stages, each inference stage is based on the output result of the previous stage.
4. The monocular 3D target detection method based on nested inference chains as described in claim 3, characterized in that, No. Query vectors for each inference stage Determined by the query vector from the previous stage and the original context features, the update relationship is expressed as follows: in, Indicates aggregation operation, Indicates a calibration operation. This represents the set of query vectors output during the preliminary inference stage. The original context features are used; each inference stage completes the inference step by step according to the dependency relationship between the three-dimensional geometric attributes, forming a nested inference chain structure, which enhances the consistency between the prediction results of different geometric attributes.
5. The monocular 3D target detection method based on nested inference chains as described in claim 1, characterized in that, The process of dynamically aggregating intermediate features generated at each inference stage to obtain a fused feature representation includes: obtaining the query vector of the current stage and multiple query vectors output by the preceding inference stages, and generating dynamic weights through a linear layer to characterize the contribution of different inference stages. The dynamic weights are used to reflect the relative contribution of different inference stages to the inference task of the current stage; and fusing the weighted multiple query vectors to obtain the aggregated query vector.
6. The monocular 3D target detection method based on nested inference chains as described in claim 1, characterized in that, The process of correcting intermediate predictions to suppress error propagation includes: using aggregated features from the current inference stage. As input, and incorporate the extracted original contextual features. The following steps are performed sequentially: normalization, multi-head attention module, and feedforward neural network, as shown below: in, This indicates normalization processing. This represents the attention calculation process. This represents a feedforward neural network.
7. The monocular 3D target detection method based on nested inference chains as described in claim 1, characterized in that, The three-dimensional detection results include the target's three-dimensional dimensions, spatial orientation, and spatial depth information.
8. A monocular 3D target detection system based on nested inference chains, characterized in that, include: The feature extraction module is configured to acquire a monocular image and extract feature representations for target detection based on the monocular image; The nested inference module is configured to construct a multi-stage nested inference chain based on the feature representation, and to predict the three-dimensional size parameters, spatial orientation parameters and spatial depth parameters of the target step by step according to the preset geometric dependency relationship. In the step-by-step prediction, the prediction result of the previous stage is used as the condition input for the prediction of the next stage. The dynamic aggregation module is configured to dynamically aggregate intermediate features generated at each inference stage in the multi-stage nested inference chain to obtain a fused feature representation. The correction module is configured to use an error correction mechanism to combine the fused feature representation with the original contextual information of the monocular image to correct the intermediate prediction results in order to suppress error propagation. The output module is configured to obtain the three-dimensional detection results of the target based on the corrected prediction results.
9. A computer-readable storage medium, characterized in that, Used to store computer instructions, which, when executed by a processor, complete the steps of the method according to any one of claims 1-7.
10. An electronic device, characterized in that, It includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, which, when executed by the processor, perform the steps of the method according to any one of claims 1-7.