Monocular 6D pose estimation method and system based on enhanced feature projection decoupling

Abstract

The disclosure provides a monocular 6D pose estimation method and system based on enhanced feature projection decoupling, relating to the technical field of robot visual perception, comprising: acquiring a monocular RGB image of a target object scene; inputting the monocular RGB image into a trained pose estimation network to obtain the 6D pose of the object; the pose estimation network extracts and fuses local features and multi-scale global features to obtain high-dimensional object enhanced feature representation; the high-dimensional object enhanced feature representation is subjected to coordinate projection decoupling to obtain the dense corresponding mapping between the three-dimensional model and the two-dimensional image of the target object and the object mask; the dense corresponding mapping and the object mask of the target object are subjected to pose regression to obtain the 6D pose of the object; the present application does not need to rely on depth information, multi-view images and multiple intermediate representations, but only needs to use monocular RGB images to accurately estimate the object pose, reduces the network design and data labeling cost and improves the robustness to complex scenes.

Description

Technical Field

[0001] This disclosure relates to the field of robot vision perception technology, specifically to a monocular 6D pose estimation method and system based on enhanced feature projection decoupling. Background Technology

[0002] 6D object pose estimation is a measurement task that calculates the rigid transformation from the target object's coordinate system to the camera coordinate system based on an input image, including 3D rotation and 3D translation. This technology is currently widely used in vision-guided robotic systems such as grasping, automated assembly, and human-computer interaction.

[0003] Most deep learning-based object pose estimation methods rely on depth information. However, depth sensors are susceptible to interference from multipath and ambient light, and have inherent limitations in measurement range, field of view, and frame rate. This makes these methods less robust to environmental noise and sparsely textured objects, and also results in low operating efficiency. Recently, some RGB modal-based methods have mainly fallen into two paradigms: multi-view image-based methods and multi-intermediate representation-based methods. Both methods essentially utilize geometric consistency to introduce multiple constraints to alleviate the feature sparsity challenges caused by complex scenes such as occlusion, truncation, clutter, and textureless objects, thereby improving the accuracy of object pose estimation.

[0004] The current RGB-modal-based technology paradigm mainly suffers from the following problems:

[0005] Multi-view-based methods involve complex camera calibration processes, which are not conducive to practical deployment. Furthermore, they require significant computing resources during training and inference, limiting their application in resource-constrained scenarios.

[0006] Methods based on multiple intermediate representations require labeling various types of intermediate representations, such as keypoints, contours, dense correspondences, and surface normals. Furthermore, different estimation networks need to be designed for different types of intermediate representations, resulting in high costs for model design and labeling.

[0007] Therefore, existing pose estimation methods based on RGB modalities rely on multiple viewpoints or intermediate representations, resulting in high annotation costs and low operating efficiency. Summary of the Invention

[0008] To address the aforementioned issues, this disclosure proposes a monocular 6D pose estimation method and system based on enhanced feature projection decoupling. This method does not rely on depth information, multi-view images, or various intermediate representations. It can accurately estimate object pose using only a monocular RGB image and dense correspondence mapping as an intermediate representation. Furthermore, it exhibits strong robustness to complex scenes such as occluded, truncated, cluttered, and sparsely textured objects, thereby reducing network design and data annotation costs and improving robustness to complex scenes.

[0009] According to some embodiments, the present disclosure adopts the following technical solutions:

[0010] A monocular 6D pose estimation method based on enhanced feature projection decoupling includes:

[0011] Acquire a monocular RGB image of the target object scene;

[0012] The monocular RGB image is input into the trained pose estimation network to obtain the 6D pose of the object.

[0013] The pose estimation network extracts and fuses local features and multi-scale global features to obtain a high-dimensional object enhancement feature representation; it then decouples the high-dimensional object enhancement feature representation by coordinate projection to obtain a dense correspondence mapping between the 3D model and the 2D image of the target object, as well as an object mask; based on the dense correspondence mapping and the object mask of the target object, it performs pose regression to obtain the 6D pose of the object.

[0014] According to some embodiments, the present disclosure adopts the following technical solutions:

[0015] A monocular 6D pose estimation system based on enhanced feature projection decoupling includes:

[0016] The image acquisition module is configured to acquire a monocular RGB image of the target object scene;

[0017] The pose estimation module is configured to input a monocular RGB image into the trained pose estimation network to obtain the 6D pose of the object.

[0018] The pose estimation network extracts and fuses local features and multi-scale global features to obtain a high-dimensional object enhancement feature representation; it then decouples the high-dimensional object enhancement feature representation by coordinate projection to obtain a dense correspondence mapping between the 3D model and the 2D image of the target object, as well as an object mask; based on the dense correspondence mapping and the object mask of the target object, it performs pose regression to obtain the 6D pose of the object.

[0019] According to some embodiments, the present disclosure adopts the following technical solutions:

[0020] A computer program product includes a computer program that, when executed by a processor, implements the monocular 6D pose estimation method based on enhanced feature projection decoupling.

[0021] According to some embodiments, the present disclosure adopts the following technical solutions:

[0022] A non-transitory computer-readable storage medium is provided for storing computer instructions, which, when executed by a processor, implement the monocular 6D pose estimation method based on enhanced feature projection decoupling.

[0023] According to some embodiments, the present disclosure adopts the following technical solutions:

[0024] An electronic device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the monocular 6D pose estimation method based on enhanced feature projection decoupling.

[0025] Compared with the prior art, the beneficial effects of this disclosure are as follows:

[0026] This invention provides a monocular 6D pose estimation method and system based on enhanced feature projection decoupling, aiming to overcome the problems of high annotation cost and low running efficiency caused by existing RGB modality-based pose estimation methods that rely on multiple views or intermediate representations. The proposed method alleviates the feature sparsity problem caused by complex environments by extracting complementary high-dimensional object enhancement features using dual-stream branches. At the same time, it projects the high-dimensional object enhancement features onto three two-dimensional orthogonal planes and introduces multiple geometric constraints by utilizing the coordinate consistency between the planes to estimate the dense correspondence mapping between the object's 3D model and 2D image. This allows the network to effectively improve the running efficiency and accuracy of pose estimation in various complex scenes such as occlusion, truncation, and textureless objects, relying only on a single-view RGB image and a single intermediate representation, and has high engineering practical value.

[0027] This invention does not rely on depth information, multi-view images, or multiple intermediate representations. It can accurately estimate object pose using only a single intermediate representation: a monocular RGB image and dense correspondence mapping. It also exhibits strong robustness to complex scenes such as occluded, truncated, cluttered, and sparsely textured objects, significantly reducing the cost of network design and manual annotation. Furthermore, it can run in real time on devices with limited computing power, such as laptops, expanding its application in resource-constrained scenarios. Attached Figure Description

[0028] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0029] Figure 1 This is a flowchart illustrating the overall method of an embodiment of this disclosure.

[0030] Figure 2 This is a diagram of the pose estimation network structure according to an embodiment of the present disclosure.

[0031] Figure 3 This is a diagram of the context clustering network structure in an embodiment of this disclosure.

[0032] Figure 4 This diagram illustrates context clustering network grouping, dynamic feature aggregation, and feature updating in an embodiment of this disclosure.

[0033] Figure 5 This is a schematic diagram of coordinate projection decoupling in an embodiment of this disclosure. Detailed Implementation

[0034] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0035] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of this disclosure. 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 disclosure pertains.

[0036] 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 exemplary embodiments according to this disclosure. 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.

[0037] Example 1

[0038] One embodiment of this disclosure provides a monocular 6D pose estimation method based on enhanced feature projection decoupling, including:

[0039] Acquire a monocular RGB image of the target object scene;

[0040] The monocular RGB image is input into the trained pose estimation network to obtain the 6D pose of the object.

[0041] The pose estimation network extracts and fuses local features and multi-scale global features to obtain a high-dimensional object enhancement feature representation; it then decouples the high-dimensional object enhancement feature representation by coordinate projection to obtain a dense correspondence mapping between the 3D model and the 2D image of the target object, as well as an object mask; based on the dense correspondence mapping and the object mask of the target object, it performs pose regression to obtain the 6D pose of the object.

[0042] As one embodiment, the monocular 6D pose estimation method based on enhanced feature projection decoupling disclosed herein utilizes, for example... Figure 2The pose estimation network shown does not rely on depth information and multi-view images. It can accurately estimate the object pose using only a single intermediate representation, a monocular RGB image and a dense correspondence mapping. It also has strong robustness to complex scenes such as occlusion, truncation, clutter, and objects with sparse textures. The specific implementation process is as follows:

[0043] Step S1: Obtain a monocular RGB image of the target object scene.

[0044] The monocular RGB image of the target object scene is obtained by cropping the target object region from the original scene image and using the target object region as the monocular RGB image of the target object scene.

[0045] Specifically, in step S11, using object detection algorithms, such as the YOLO series algorithms and the Mask RCNN series algorithms, the two-dimensional bounding box of the region where the target object is located in the original scene image is estimated.

[0046] Step S12: Crop the target object region based on the two-dimensional bounding box and scale it to a predefined size, such as 256×256. This image is the monocular RGB image of the target object scene.

[0047] Step S2: Input the RGB image into the feature enhancement network to extract the local features and multi-scale global features of the object. Then, the global features are successively fused into the local features and upsampled to obtain a high-dimensional object enhancement feature representation that is 1 / 4 the size of the input image.

[0048] The feature enhancement network mainly includes local feature extraction, global feature extraction, and collaborative feature fusion upsampling, specifically:

[0049] Step S21, Local Feature Extraction: Local features are extracted from the input image using ConvNext-base to obtain high-dimensional local features with a resolution of 1 / 32 of the input image size.

[0050] Step S22: Utilize each pixel I in the input image i,j Location information To enhance the image and convert the enhanced image into a two-dimensional point set. Where n = H × W represents the number of points, and each pixel after enhancement contains both color and position information.

[0051] Step S23, Multi-stage global feature extraction: Utilizing, for example... Figure 3 The context clustering network shown performs multi-stage downsampling, grouping, dynamic feature aggregation, and feature updating on images in the form of two-dimensional point sets, resulting in multi-scale features with resolutions of 1 / 4, 1 / 8, 1 / 16, and 1 / 32 of the input image size.

[0052] Specifically, it includes the following steps:

[0053] Step S231: Use strided convolution to downsample the input features by a factor of two.

[0054] Step S232, as Figure 4 As shown by the orange path, two fully connected layers are used to project the feature P, which is in the form of a two-dimensional point set, onto a high-dimensional space of the same dimension, to obtain P. q and P v This is used for subsequent similarity calculations, utilizing average pooling to calculate the similarity from P. q and P v A predetermined number of center points P are selected uniformly in the middle. q (c) and P v (c) A multi-head similarity calculation is performed on all center points and other points. Finally, a fully connected layer (FC) is used to concatenate and fuse the multi-head calculation results to obtain the similarity matrix Sim:

[0055] Sim=FC(Cos(P q (c),P q )

[0056] Where Cos represents the cosine similarity calculated using the multi-head pattern;

[0057] Step S233: Divide all feature points into several groups based on similarity. Each point is assigned to the group containing the center point with the highest similarity. All feature points within the same group are dynamically aggregated to the center point. Taking the i-th group as an example, after feature aggregation, the center point feature P... v (c i The formula for calculating )′ is:

[0058]

[0059] Where sig represents the sigmoid function, used to normalize weights, s ij Let μ represent the feature similarity score between the center point of group i and the j-th point in the group. μ and v are trainable parameters used to dynamically weight each feature point.

[0060] Step S234, as Figure 4 As shown by the green path, the aggregated features of the center point are assigned to other points within the group based on the similarity matrix to update their features, ensuring that points within the same group can exchange information with each other. Taking the j-th point P(i,J) in the i-th group as an example, the updated feature P(i,j)′ is:

[0061] P(i,j)′=P(i,j)+FC(sig(μs ij +v)*P v (c i )′)

[0062] Step S235: Input the updated feature point set into a fully connected layer with residual connections to obtain the final output features.

[0063] Step S24, Collaborative Feature Fusion Upsampling: The local features output in step S21 are fused with the global features output in step S23 that match the resolution of the local features, and upsampling is performed. The upsampled features are then further fused with the global features at the corresponding resolution. The final output is a feature-enhanced representation with a size of 1 / 4 of the input image size.

[0064] Furthermore, the specific steps include:

[0065] Step S241: The local features output in step S21 are additively fused with the global features matching the resolution, and then upsampled by a factor of two. The fused features are then sequentially input into a 3×3 convolution and a dilated convolution with residual connections for spatial interaction, outputting fused features with a resolution of 1 / 16 of the input image size. This step is repeated twice to finally obtain fused features with a resolution of 1 / 4 of the input image size.

[0066] The techniques used in the three 2x upsampling operations are bilinear sampling, bilinear sampling, and transposed convolution, respectively.

[0067] Step S242: Continue to fuse the output of step S241 with the global features corresponding to the resolution, and perform spatial interaction to obtain enhanced features with a resolution of 1 / 4 of the input image size.

[0068] Step S3: Input the high-dimensional object enhancement feature representation into the coordinate projection decoupling module to obtain the dense correspondence mapping between the three-dimensional model and the two-dimensional image of the target object, as well as the object mask.

[0069] Please refer to Figure 5 Specifically, it includes the following steps:

[0070] Step S31: Input the enhanced features output from step S242 into three sub-networks and one fully connected layer, and regress the projections C of the dense correspondence mapping onto three two-dimensional orthogonal planes (i.e., the xy, yz, and xz planes), respectively. xy C yz C xz and the mask C of the target object m :

[0071]

[0072] Where, N xy N yz N xz N mThese are the sub-network for predicting projected coordinates and the fully connected layer for predicting the object mask, respectively, with parameters ∈, ζ, θ, v.

[0073] Step S32: Calculate the mean coordinates between the two-dimensional projection planes using average pooling (Avg) to obtain the dense correspondence mapping, and then multiply it by the target object mask to remove the background, thus obtaining the dense correspondence mapping of the target object.

[0074]

[0075] Where, p x p y and p z Let C represent the sets of three coordinate components of the dense projection mapping. xy [0] represents the x-coordinate component of the dense correspondence mapping projected onto the xy plane, and other symbols follow the same pattern.

[0076] The three subnetworks have the same structure. Each subnetwork consists of three convolutional layers with a kernel size of 3×3 and a stride of 1, using grouped normalization and Gaussian error linear unit activation functions.

[0077] Step S4: Input the background-removed target object dense correspondence mapping into the pose regression network to calculate the object's 6D pose.

[0078] The pose regression network consists of three strided downsampling convolutional layers and five fully connected layers. The first three fully connected layers are used to reduce the dimensionality of the flattened features, and the last two fully connected layers are used to calculate the 3D rotation R and 3D translation T of the object, respectively, to obtain the 6D pose of the object.

[0079] Step S5: Based on the difference between the network's forward output and the true value, perform iterative training, update the parameters of all modules, and use the converged pose estimation network to process the input target object image to obtain the object's 6D pose.

[0080] Specifically, during iterative training in step S5, a multi-task learning strategy is adopted. Multiple loss functions are defined and summed according to different task attributes to guide the model's parameter updates. This includes:

[0081] S51. Define masking loss:

[0082]

[0083] in, Let be the actual value of the mask for the i-th pixel. Let be a conditional function, when the mask prediction value of the i-th pixel is... Greater than or equal to The function is 1 if the condition is true and 0 otherwise. In this embodiment,

[0084] S52. Define the pose loss function:

[0085]

[0086] Where, x i This represents the i-th point sampled from the 3D model of the object. and These are the actual and predicted values ​​for 3D rotation, respectively. and These represent the actual and predicted values ​​of the 3D translation in the SITE format, respectively. In this embodiment, k = 1024.

[0087] S53. Define the dense correspondence mapping loss:

[0088]

[0089] Where, n and H c ×W c These represent the resolution and the number of foreground points in the dense correspondence mapping, respectively.

[0090] Let L1 norm be the ratio of the predicted value to the actual value of the i-th dense correspondence mapping point.

[0091] S54. Define coordinate consistency loss:

[0092] L cc =|C xy [0]-C xz [0]|1+|C xy [1]-C yz [0]|1+|C xz [1]-C xz [1]|1)

[0093] Among them, C xy [0] represents the set of x-coordinate components of the dense correspondence mapping projected onto the xy plane, and other symbols follow the same pattern.

[0094] S55. Constructing a loss function based on multi-task learning

[0095] L = L mask +L pose +L map +5L cc

[0096] Table 1 shows the comparison results of this embodiment with the state-of-the-art methods on the LM-O and T-LESS datasets. The datasets contain 61 objects with different shapes and textures, and include various challenges such as occlusion, clutter, and truncation. The methods listed in the table all use the BOP metric as the performance evaluation metric.

[0097] Table 1: Performance comparison with state-of-the-art methods on the LMO and T-LESS datasets

[0098] Method Name LM-O T-LESS DPODv2 58.6 72.0 OVE-6D 62.7 54.6 ER-Pose 61.6 65.4 MatchU 68.0 66.8 Ours 69.7 79.5

[0099] As can be seen, the method in this embodiment outperforms the state-of-the-art methods on the dataset. It does not rely on depth information and multi-view images, nor does it rely on multiple intermediate representations. It can accurately estimate the object pose using only one intermediate representation: a monocular RGB image and dense correspondence mapping. This reduces the cost of network design and data annotation, and it has strong robustness to complex scenes such as occluded, truncated, cluttered, and sparsely textured objects.

[0100] Example 2

[0101] One embodiment of this disclosure provides a monocular 6D pose estimation system based on enhanced feature projection decoupling, comprising:

[0102] The image acquisition module is configured to acquire a monocular RGB image of the target object scene;

[0103] The pose estimation module is configured to input a monocular RGB image into the trained pose estimation network to obtain the 6D pose of the object.

[0104] The pose estimation network extracts and fuses local features and multi-scale global features to obtain a high-dimensional object enhancement feature representation; it then decouples the high-dimensional object enhancement feature representation by coordinate projection to obtain a dense correspondence mapping between the 3D model and the 2D image of the target object, as well as an object mask; based on the dense correspondence mapping and the object mask of the target object, it performs pose regression to obtain the 6D pose of the object.

[0105] Example 3

[0106] One embodiment of this disclosure provides a computer program product, including a computer program that, when executed by a processor, implements the monocular 6D pose estimation method based on enhanced feature projection decoupling.

[0107] Example 4

[0108] One embodiment of this disclosure provides a non-transitory computer-readable storage medium for storing computer instructions. When these computer instructions are executed by a processor, they implement the monocular 6D pose estimation method based on enhanced feature projection decoupling.

[0109] Example 5

[0110] One embodiment of this disclosure provides an electronic device, including a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the monocular 6D pose estimation method based on enhanced feature projection decoupling.

[0111] This disclosure is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. 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, create a machine 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.

[0112] 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.

[0113] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A monocular 6D pose estimation method based on enhanced feature projection decoupling, characterized in that, include: Acquire a monocular RGB image of the target object scene; The monocular RGB image is input into the trained pose estimation network to obtain the 6D pose of the object. The pose estimation network extracts and fuses local features and multi-scale global features to obtain a high-dimensional object enhancement feature representation; it then decouples the high-dimensional object enhancement feature representation by coordinate projection to obtain a dense correspondence mapping between the 3D model and the 2D image of the target object, as well as an object mask; based on the dense correspondence mapping and object mask of the target object, it performs pose regression to obtain the 6D pose of the object. The feature enhancement network consists of local feature extraction, multi-stage global feature extraction, and collaborative feature fusion upsampling. The local feature extraction uses ConvNext-base to extract local features from the input image; The multi-stage global feature extraction is specifically as follows: The image is enhanced by using the positional information of each pixel in the image, and the enhanced image is then converted into a two-dimensional point set. A preset number of center points are selected uniformly using average pooling, and the similarity between all center points and other points is calculated to obtain a similarity matrix. Based on the similarity matrix, all feature points are divided into several groups. Each point is assigned to the group containing the center point with the highest similarity. All feature points in the same group are dynamically aggregated to the center point. Based on the similarity matrix, the aggregated features of the center point are assigned to other points within the group to update their features, ensuring that points within the same group can exchange information with each other. The updated set of feature points is input into a fully connected layer with residual connections to obtain multi-scale global features. The collaborative feature fusion upsampling fuses local features with multi-scale global features and performs upsampling. The upsampled features are then fused with global features of the corresponding resolution to finally obtain a high-dimensional object enhanced feature representation. The coordinate projection decoupling is specifically as follows: From the enhanced feature representation of high-dimensional objects, three sub-networks and a fully connected layer branch are used to regress the projection of dense correspondence mapping onto three two-dimensional orthogonal planes and the mask of the target object, respectively. The mean coordinates between two-dimensional projection planes are calculated using average pooling to obtain a dense correspondence mapping. This dense correspondence mapping is then obtained by multiplying the mean coordinates with the target object mask.

2. The monocular 6D pose estimation method based on enhanced feature projection decoupling as described in claim 1, characterized in that, The monocular RGB image of the target object scene is obtained by cropping the target object region from the original scene image and using the target object region as the monocular RGB image of the target object scene.

3. The monocular 6D pose estimation method based on enhanced feature projection decoupling as described in claim 1, characterized in that, The pose regression is a process that uses a pose regression network to calculate the 6D pose of an object. The pose regression network consists of three strided downsampling convolutional layers and five fully connected layers. The first three fully connected layers are used to reduce the dimensionality of the flattened features, and the last two fully connected layers are used to calculate the 3D rotation of the object. and three-dimensional translation Finally, the 6D pose of the object is obtained.

4. The monocular 6D pose estimation method based on enhanced feature projection decoupling as described in claim 1, characterized in that, The loss function of the pose estimation network includes mask loss, pose loss, dense correspondence mapping loss, and coordinate consistency loss.

5. A monocular 6D pose estimation system based on enhanced feature projection decoupling, employing the monocular 6D pose estimation method based on enhanced feature projection decoupling as described in any one of claims 1-4, characterized in that, include: The image acquisition module is configured to acquire a monocular RGB image of the target object scene; The pose estimation module is configured to input a monocular RGB image into the trained pose estimation network to obtain the 6D pose of the object. The pose estimation network extracts and fuses local features and multi-scale global features to obtain a high-dimensional object enhancement feature representation; it then decouples the high-dimensional object enhancement feature representation by coordinate projection to obtain a dense correspondence mapping between the 3D model and the 2D image of the target object, as well as an object mask; based on the dense correspondence mapping and the object mask of the target object, it performs pose regression to obtain the 6D pose of the object.

6. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the monocular 6D pose estimation method based on enhanced feature projection decoupling as described in any one of claims 1-4.

7. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium is used to store computer instructions, which, when executed by a processor, implement the monocular 6D pose estimation method based on enhanced feature projection decoupling as described in any one of claims 1-4.

8. An electronic device, characterized in that, include: The device includes a processor, a memory, and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to perform the monocular 6D pose estimation method based on enhanced feature projection decoupling as described in any one of claims 1-4.

Images