Efficient multi-view three-dimensional reconstruction method and system for outdoor scene

Abstract

The invention discloses an outdoor scene-oriented efficient multi-view three-dimensional reconstruction method and system, and belongs to the field of multi-view three-dimensional reconstruction, and the method specifically comprises the steps: preparing a multi-view three-dimensional reconstruction data set for model training; constructing a multi-view three-dimensional reconstruction network comprising an adaptive feature extraction converter module based on big kernel attention, a cost regularization module and a depth estimation module; performing training and fine tuning on the network model to obtain a final model; building a multi-view three-dimensional reconstruction system and equipment; according to the method, the problems of poor accuracy and low integrity of a reconstruction result when an existing multi-view three-dimensional reconstruction method faces an outdoor large-scale scene can be solved, and the training reasoning time and the calculation cost of multi-view three-dimensional reconstruction can be reduced; and an important foundation is laid for application of outdoor large-scale scene image data in the field of multi-view three-dimensional reconstruction and development of a three-dimensional reconstruction technology.

Description

Technical Field

[0001] The present invention relates to multi-view Figure 3 The present invention relates to the fields of computer graphics and computer vision, and specifically to an efficient multi-view stereo reconstruction method and system for outdoor scenes. Background Art

[0002] In recent years, multi-view stereo reconstruction based on deep learning has attracted more and more interest due to its wide application in unmanned driving, virtual reality, augmented reality and other fields. Although existing multi-view stereo reconstruction methods have made great progress in indoor controllable environments, these methods are difficult to deal with problems such as occlusion between objects, non-Lambertian surfaces, uneven lighting and weak texture areas in complex outdoor scenes. Although recent studies have introduced visual converter modules to enable the network to pay attention to the global context and enhance the quality of the 3D model, due to the quadratic complexity of self-attention, these multi-view stereo reconstruction methods that introduce visual converters require a lot of running time and high video memory capacity, which seriously affects the development and application of multi-view stereo 3D reconstruction technology. Therefore, it is of great research significance to design an efficient multi-view stereo reconstruction method for outdoor scenes.

[0003] The Chinese patent publication number is "CN115908723A", and its name is "Limit-guided multi-view stereo reconstruction method based on interval perception". This method first captures the global contextual information inside and across images through the epipolar line guided aggregation submodule, then generates the cost volume using the visible fusion submodule, and finally uses the depth estimation module to calculate the loss. Although this method has addressed the problem of weak cross-information feature extraction capabilities of the multi-view stereo reconstruction network to a certain extent, it can neither solve problems such as occlusion between objects, non-Lambertian surfaces, uneven lighting, and weak texture areas, nor can it effectively deal with problems such as low time efficiency, large model parameters, and large model calculations in the multi-view stereo reconstruction process based on outdoor scenes. Therefore, designing a method that can efficiently complete multi-view stereo reconstruction tasks in real outdoor scenes is a key problem that the present invention focuses on solving. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides an efficient multi-view stereo reconstruction method and system for outdoor scenes, which can quickly calculate a dense point cloud model of a high-quality reconstruction scene from a large-scale image in a complex outdoor scene, and has a faster running time, less computing cost and smaller storage cost than the existing multi-view stereo method, promoting the advancement of multi-view stereo reconstruction technology for outdoor scenes. At the same time, the present invention can improve the stability and performance of training and has high generalization.

[0005] In order to achieve the above-mentioned purpose, the present invention specifically adopts the following technical solutions:

[0006] An efficient multi-view stereo reconstruction method for outdoor scenes comprises the following steps:

[0007] Step 1, prepare a data set: prepare a first multi-view stereo reconstruction data set and a second multi-view stereo reconstruction data set, and divide the two data sets into a training data set and a test data set respectively; the training data set is used to train the network model, and the test data set is used to evaluate the model training effect;

[0008] Step 2, build the network model: the network consists of a backbone network, an adaptive feature extraction converter module based on large core attention, a cost regularization module, and a depth estimation module.

[0009] Step 3, training the network model: input the data set prepared in step 1 into the network model built in step 2 for training; by minimizing the loss between the network output and the label, better depth prediction results can be achieved;

[0010] Step 4: Screen the optimal network model; use the test data set as input to evaluate and screen the optimal network model parameters;

[0011] Step 5, fine-tune the network model; use the second multi-view stereo reconstruction dataset to fine-tune the network model so that the model can be generalized from indoor scenes with controllable lighting to large-scale outdoor scenes.

[0012] Step 6, save the network model; after the network training is completed, save the optimal network parameters, and then input the multi-view scene image to be three-dimensionally reconstructed into the network to obtain the predicted depth map, and obtain the reconstructed three-dimensional point cloud through depth filtering and fusion operations.

[0013] Furthermore, the first multi-view stereo reconstruction dataset is a DTU dataset, which obtains multi-view images, depth maps and corresponding camera internal and external parameters of each scene.

[0014] Furthermore, the backbone network is a feature pyramid feature extraction network, which is used to extract multi-scale features of the input reference image and the source image;

[0015] Further, the large core attention based adaptive feature extraction converter module includes an adaptive large core attention block, a hybrid feedforward neural network block, a linear cross attention block and an upsampling module;

[0016] The adaptive large-core attention block uses multi-scale convolution kernels with different expansion coefficients to form a large-size convolution kernel, which effectively reduces the computational cost and improves the computational efficiency of the model without sacrificing performance. On this basis, by introducing deformable convolution, the large-size convolution kernel can adaptively learn multi-view targets of different scales and obtain a larger receptive field, so that the model can better understand the target context information in outdoor scenes, thereby improving the ability to extract complex features.

[0017] The hybrid feedforward neural network block better aggregates global and multi-scale local information and extends the design of layer normalization to a general form, which promotes the optimization of skip connections and improves the network feature extraction capability.

[0018] The linear cross attention block enables the network to learn the feature association between images from different perspectives, helping the model to better understand the spatial structure of the object and improve the accuracy and stability of stereo reconstruction.

[0019] Furthermore, the cost regularization module, by replacing the three-dimensional convolution kernel with a three-dimensional adaptive large kernel attention, can expand the receptive field without increasing model parameters and computational complexity, so that the model can better capture the global information in the scene, thereby improving the accuracy of stereo reconstruction.

[0020] Furthermore, the depth estimation module uses maximum probability depth estimation to select, for each pixel position, a depth value with the highest probability value from the probability body, and uses it as the depth value of the pixel position in the depth estimation map.

[0021] Furthermore, the preset thresholds in the training network model include a preset value of the loss function, a preset value of the number of iterations, and a preset value of the optimized learning rate.

[0022] Furthermore, the loss function is a focal loss to strengthen the single hotspot supervision in the fuzzy area.

[0023] Furthermore, the process of training the network model also includes evaluating the accuracy and completeness of the algorithm's reconstruction results through evaluation indicators.

[0024] Furthermore, the second multi-view stereo reconstruction dataset is a BlendedMVS dataset.

[0025] A multi-view stereo reconstruction system for outdoor scenes, the system comprising:

[0026] The image acquisition module is used to acquire training data and test data; the training data and test data include multi-view images and corresponding depth images.

[0027] A model building module is used to build the network model proposed by this method and initialize the model weights;

[0028] A data set and model loading module, used to load input images and network models onto a hardware device for model training; the hardware device refers to an image processor;

[0029] A model training module is used to perform supervised model training on the network model constructed by this method for the multi-view images of each scene in the training data to obtain the optimal model weight;

[0030] A depth estimation module is used to perform depth estimation on the multi-view images of each scene in the test data through a network model equipped with optimal model weights to obtain a multi-view depth map of each scene;

[0031] The depth filtering and fusion module is used to obtain a 3D point cloud file for each scene after performing depth filtering and fusion operations on the multi-view depth map predicted by the network;

[0032] Point cloud visualization module, used to parse and process 3D point cloud files through the MeshLab software platform to achieve 3D point cloud visualization;

[0033] A storage medium is used to store a multi-view stereo reconstruction system.

[0034] Beneficial Effects

[0035] Compared with the prior art, the present invention provides an efficient multi-view stereoscopic reconstruction method and system for outdoor scenes, which has the following beneficial effects:

[0036] 1. This paper designs a new multi-view stereo reconstruction network architecture, constructs a multi-view feature extraction and probabilistic volume prediction network based on adaptive large-core attention, solves the problems of long training and reasoning time and high computational cost of existing multi-view stereo reconstruction methods when facing large-scale outdoor scenes, and provides a new method for large-scale outdoor scene image data in multi-view stereo reconstruction. Figure 3 It lays an important foundation for the application of 3D reconstruction in the field and the development of 3D reconstruction technology.

[0037] 2. The adaptive large-core attention module proposed in this invention not only takes into account the global and local information capture capabilities of the convolution and self-attention mechanisms, but also avoids their limitations of high computational quadratic complexity, redundant model parameters, and fixed receptive field size. Without reducing the accuracy and completeness of stereo reconstruction, it can accelerate the training and inference process of the model and improve the computational efficiency of the model, especially when processing large-scale outdoor scene data.

[0038] 3. The hybrid feedforward neural network module proposed in the present invention utilizes multi-scale convolution kernels and layer normalization extensions to capture the relationship between different scale features of multi-view targets in the same scene, further enhances the information interaction between different channels, aggregates the correspondence between feature maps of different depths and targets, and thus obtains richer feature information.

[0039] 4. The cost volume transformer based on three-dimensional adaptive large kernel attention proposed in the present invention can improve the feature expression ability, information interaction ability and receptive field coverage of the model, thereby showing stronger stability and generalization when facing complex scenes and input changes.

[0040] 5. The multi-view stereo reconstruction network proposed in the present invention has shown good results in both the DTU dataset and the BlendedMVS dataset. While ensuring the reconstruction quality, the memory usage, computational complexity and inference time are reduced by 33%, 32% and 41% respectively, effectively reducing the training and use costs of the model and improving the reconstruction efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flow chart of an efficient multi-view stereo reconstruction method for outdoor scenes;

[0042] Figure 2 This is the overall architecture diagram of an efficient multi-view stereo reconstruction network model for outdoor scenes;

[0043] Figure 3 This is a module structure diagram of the adaptive feature extraction converter based on large core attention according to the present invention;

[0044] Figure 4 A schematic diagram of the specific composition of all the adaptive large-core attention modules described in the present invention;

[0045] Figure 5 It is a schematic diagram of the specific composition of all the hybrid feedforward neural network modules described in the present invention;

[0046] Figure 6 A schematic diagram of the specific composition of all cost volume converters based on three-dimensional adaptive large core attention described in the present invention;

[0047] Figure 7 This is a schematic diagram for comparing relevant indicators of the method proposed in the present invention;

[0048] Figure 8 A schematic diagram of the main modules of a multi-view stereo reconstruction system for outdoor scenes;

[0049] Fig. 9 A dense point cloud model diagram provided by an embodiment of the present application;

[0050] Fig.10 Schematic diagram of the internal structure of an electronic device for multi-view stereo reconstruction of outdoor scenes. DETAILED DESCRIPTION

[0051] The technical solution of the present invention is described in detail below, but the protection scope of the present invention is not limited to the embodiments.

[0052] Example

[0053] like Figure 1 As shown, an efficient multi-view stereoscopic reconstruction method for outdoor scenes in this embodiment includes the following steps:

[0054] Step 1, prepare the data set; prepare the first multi-view stereo reconstruction data set and the second multi-view stereo reconstruction data set, obtain the multi-view images, depth maps and corresponding internal and external parameters of the camera for each scene; divide the two data sets into training data sets and test data sets respectively; the training data set is used to train the network model, and the test data set is used to evaluate the model training effect; the first multi-view stereo reconstruction data set uses the DTU data set; the DTU data set is a large-scale multi-view stereo reconstruction data set with precise camera trajectories collected in a controlled laboratory environment, with an image resolution of 640×512; the data set contains 119 scenes in 49 views under 7 different lighting conditions, and each set of data consists of RGB images and phases and provides a true point cloud scanned using structured light; we selected 97 scenes as the training set of the network model and 22 scenes as the test set of the network model;

[0055] Step 2: Build a network model. We designed an efficient multi-view stereo reconstruction network model for outdoor scenes. The model architecture is as follows: Figure 2 As shown;

[0056] First, we use the feature pyramid as the backbone network to extract the input reference image I0∈R H×W×3 and the source image Multi-scale features Where i represents the i-th input view and N represents the number of input images; in order to expand the ability of long-term context aggregation and global feature interaction across scales, we design an adaptive feature extraction converter module based on large kernel attention, which uses a combination of adaptive large kernel attention modules and cross-attention modules within and across multi-view images at each stage. Secondly, the source image features are mapped to the reference image through homography transformation and fused to obtain a three-dimensional cost volume V, where V∈R H×W×C×D Then, we use the cost volume transformer based on 3D adaptive large kernel attention to perform regularization operation and generate a probability volume P, where P∈R H×W×DFinally, a Softmax operation is applied to the probability volume to estimate the depth of the reference image.

[0057] The specific contents of the above modules are described as follows:

[0058] like Figure 3 As shown in the figure, the adaptive feature extraction converter module based on large core attention consists of a self-attention module, an upsampling module and a cross-attention module; the self-attention module adopts a converter architecture, and uses the adaptive large core attention module and the hybrid feedforward neural network module proposed in this work to efficiently capture the global and local context features inside the image, so that the network pays more attention to challenging areas in multi-view stereo matching, such as weak textures and non-Lambertian surfaces; the upsampling module adopts bilinear interpolation upsampling to fuse multi-scale features; the cross-attention module adopts a linear cross-attention block to capture cross-image context information, and also uses a hybrid feedforward neural network module to aggregate global and local context information;

[0059] Among them, the specific composition diagram of the adaptive large core attention module is as follows Figure 4 As shown in the figure, a large kernel convolution can be lightweight decomposed into a depthwise separable convolution, a depthwise separable dilated convolution, and a pointwise convolution. Specifically, we use a 5×5 depthwise separable convolution, a 5×5 depthwise separable dilated convolution with a dilation factor of 3, and a 1×1 pointwise convolution to approximate a 13×13 large convolution kernel. Through the above method, we can capture long-range relationships with a slight computational cost and parameters, replace the multi-head self-attention module in the converter, and complete feature extraction;

[0060] Although the above decomposition method enables the network to extract local and global context information of the feature map while reducing the amount of computation and parameters, this decomposition method artificially fixes the convolution kernel calculation area, resulting in the network being unable to adaptively extract multi-view feature maps to reconstruct target features. To address the above issues, we introduce deformable convolution into the large-core attention module without significantly increasing the amount of computation and parameters of the model, further expanding the receptive field and the feature extraction range, so that the network can dynamically and adaptively extract image features during training, thereby improving the reconstruction accuracy of the model. The adaptive large-core attention module can be written as

[0061] F′ i =GELU(Conv 1×1 (F i ))

[0062] A=Conv 1×1 (DDW-D-Conv(DDW-Conv(F′ i )))

[0063]

[0064] First, we use a 1×1 convolution and a GELU activation function to transform the input feature F of the i-th view i Perform channel downsampling operation to obtain F' i ∈R H×W×C’ Then, through the adaptive large core attention module, we get the attention feature map A∈R H ×W×C′ The attention feature map indicates the importance of each feature. Finally, the attention feature map A and F' i After pixel-level multiplication, a 1×1 convolution is used to restore the feature channel dimension Output∈R H×W×C .

[0065] The specific composition diagram of the hybrid feedforward neural network module is as follows Figure 5 As shown in the figure, traditional converters usually use fully connected layers as feedforward neural networks, which completely rely on attention to capture the dependencies between pixels, resulting in the network being poor at learning local information of two-dimensional images, affecting the final depth estimation results. In order to solve the above problems, the present invention designs a new feedforward neural network to replace the fully connected layers in the self-attention module and the cross-attention module, thereby improving the network's ability to extract significant features across images. Specifically, we first use a 1×1 convolutional layer to expand the input features. Figure X ∈R H×W×C The channel dimension is X'∈R H×W×3C . Secondly, the feature map is evenly divided into three parts along the channel direction, and the channels are spliced ​​after a 1×1 convolution, a 3×3 depth convolution and a 5×5 depth convolution respectively, so as to fully extract the multi-scale spatial morphological feature information. Then, the jump connection is further optimized through three layers of normalization to adaptively retain the foreground target features and suppress background clutter interference. Finally, the GELU activation function and 1×1 convolution are used to restore the channel dimension of the feature map.

[0066] In the process of cost volume regularization, the global receptive field has a significant impact on the final probability volume prediction; as the receptive field expands, the voxels of the probability volume with the highest probability become smoother, more complete, and more confident in the depth dimension; in actual use, we upgrade the two-dimensional adaptive large kernel attention module to a three-dimensional representation and propose a new cost volume converter based on three-dimensional adaptive large kernel attention;

[0067] like Figure 6As shown in the figure, the cost volume converter based on 3D adaptive large kernel attention adopts a U-shaped network architecture, which consists of an encoder, a bottleneck layer, a decoder and a skip connection; given a cost volume, three 3D feature extraction stages are used consecutively in the encoder, each of which contains a 3D self-attention block and a downsampling layer. The 3D feature map with depth information is input into the 3D self-attention block composed of a 3D adaptive large kernel attention block and a hybrid feedforward neural network block to regularize and aggregate the global cost information. Compared with the 3D convolutional layer and the non-regularized module, the 3D self-attention block we proposed can produce higher quality probability volumes. The downsampling layer reduces the spatial resolution of the feature map to half of the original while doubling the channel size. The central bottleneck layer uses another 3D self-attention block to perceive the image semantic information. The structure of the decoder is symmetrical to that of the encoder. In order to reduce the channel dimension and improve the feature resolution at the same time, the upsampling layer uses transposed convolution to replace the original bilinear interpolation upsampling method to adjust the spatial and channel dimensions of the cost volume. In order to reduce the spatial information loss of the network, we use skip connections to connect shallow features and deep features in series, fuse the multi-scale features in the encoder and decoder, and retain the multi-scale features from the encoder and decoder. Finally, a 1×1×1 convolution kernel is used to aggregate the channel information of each spatial pixel to obtain the final probability volume P.

[0068] Step 3, training the network model: input the data set prepared in step 1 into the network model constructed in step 2 for training;

[0069] By minimizing the loss between the network output and the label, better depth prediction results are achieved; we use focal loss as the loss function in the network training process and treat depth estimation as a classification task to strengthen single hotspot supervision in blurred areas; focal loss can be expressed as:

[0070]

[0071] in represents the predicted probability of depth hypothesis d at pixel p, represents the depth value closest to the ground truth among all hypotheses, {p v} represents the subset of pixels with valid ground truth values, γ represents the focus parameter;

[0072] Set the input image batch size to 2, select the Adam optimizer, set the number of training iterations to 16, gradually decrease the learning rate from 0.0005 to 0, and save the weights of the iteration results from 8 to 16 rounds;

[0073] Step 4: Select the optimal network model; use the test data set as input to evaluate and select the optimal network model parameters; two official indicators are selected for evaluation, namely accuracy evaluation and completeness evaluation; accuracy evaluates the absolute distance between the average point cloud and point cloud of the reconstruction result and the ground truth, while completeness evaluates the distance between the reconstruction result and the ground truth. Overall, it is the average value of accuracy and completeness, which represents the overall performance of the model.

[0074] Step 5, fine-tune the network model; fine-tune the network model with the second multi-view stereo reconstruction dataset to generalize the model from indoor scenes with controllable lighting to large-scale outdoor scenes. The second multi-view stereo reconstruction dataset uses the BlendedMVS dataset; this dataset contains 113 outdoor scenes, each with 30 to 40 multi-view images, including buildings, street scenes, sculptures, and small objects; compared with the DTU dataset, the scenes in the BlendedMVS dataset contain a variety of different camera trajectories; unstructured camera trajectories can better model different image capture styles, so that the network has better generalization performance for real-world reconstruction. We used 106 scenes for training and 7 scenes for testing.

[0075] Step 6, save the network model; after the network training is completed, save the optimal network parameters, and then input the multi-view scene image to be three-dimensionally reconstructed into the network to obtain the predicted depth map, and obtain the reconstructed three-dimensional point cloud through depth filtering and fusion operations.

[0076] from Figure 7 It can be seen that while ensuring that the accuracy and completeness indicators are not reduced, the method proposed in the present invention has less memory usage, less calculation amount and less reasoning time than the existing method; where the memory usage unit is MB, the calculation amount unit is G, and the reasoning time unit is seconds; these indicators also further illustrate that the method proposed in the present invention has better 3D reconstruction quality and lower calculation cost;

[0077] like Figure 8 As shown, the present invention also provides a multi-view stereo reconstruction system for outdoor scenes, which mainly includes an image acquisition module, a model building module, a data set and model loading module, a model training module, a depth estimation module, a depth filtering and fusion module, and a point cloud visualization module;

[0078] The image acquisition module is used to acquire training data and test data; the training data and test data include multi-view images and corresponding depth images.

[0079] A model building module is used to build the network model proposed by this method and initialize the model weights;

[0080] A data set and model loading module, used to load input images and network models onto a hardware device for model training; the hardware device refers to an image processor;

[0081] A model training module is used to perform supervised model training on the network model constructed by this method for the multi-view images of each scene in the training data to obtain the optimal model weight;

[0082] A depth estimation module is used to perform depth estimation on the multi-view images of each scene in the test data through a network model equipped with optimal model weights to obtain a multi-view depth map of each scene;

[0083] The depth filtering and fusion module is used to obtain a 3D point cloud file for each scene after performing depth filtering and fusion operations on the multi-view depth map predicted by the network;

[0084] The point cloud visualization module is used to analyze and process the 3D point cloud file through the MeshLab software platform to realize 3D point cloud visualization; the visualization results are as follows Fig. 9 As shown;

[0085] like Fig.10 As shown, the present invention also provides a multi-view stereoscopic reconstruction electronic device for outdoor scenes, which mainly includes a memory, a processor, a communication interface and a bus; wherein the memory, the processor and the communication interface are connected to each other through the bus;

[0086] A Samsung 1TB solid-state drive is used as a memory to store data sets, programs, and intermediate products during program execution. When the program stored in the memory is executed by a processor, the processor and the communication interface are used to perform each step of the multi-view stereo reconstruction training method of an embodiment of the present invention;

[0087] Using Intel core i7-14700KF as a processor to execute relevant programs to implement the functions required to be performed by the units in the multi-view stereoscopic reconstruction training system of the present invention, or to perform the multi-view stereoscopic reconstruction training method of the present invention;

[0088] The communication interface uses a transceiver system such as, but not limited to, a transceiver to achieve communication between the system and other devices or a communication network; for example, an image to be processed or an initial feature map of the image to be processed can be obtained through the communication interface;

[0089] The bus may include a pathway for transferring information between various components of the system (e.g., memory, processor, communication interface);

[0090] The embodiments of the present application are described above. However, these embodiments are only for the purpose of illustration, and are not intended to limit the scope of the present application. Although the embodiments are described above, this does not mean that the measures in the various embodiments cannot be used in combination. The scope of the present application is limited by the attached claims and their equivalents. Without departing from the scope of the present application, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. An efficient multi-view stereo reconstruction method for outdoor scenes, characterized in that: The steps include: Step 1, preparing a data set: preparing a first multi-view stereo reconstruction data set and a second multi-view stereo reconstruction data set, and dividing the two data sets into a training data set and a test data set respectively; The training data set is used to train the network model, and the test data set is used to evaluate the model training effect; Step 2: Build the network model: The network consists of a backbone network, a large-core attention-based adaptive feature extraction converter module, a cost regularization module, and a depth estimation module; Step 3, training the network model: input the data set prepared in step 1 into the network model built in step 2 for training; by minimizing the loss between the network output and the label, better depth prediction results can be achieved; Step 4: Screen the optimal network model; use the test data set as input to evaluate and screen the optimal network model parameters; Step 5, fine-tuning the network model: fine-tuning the network model using the second multi-view stereo reconstruction dataset to generalize the model from indoor scenes with controllable lighting to large-scale outdoor scenes; Step 6, save the network model; after the network training is completed, save the optimal network parameters, and then input the multi-view scene image to be three-dimensionally reconstructed into the network to obtain the predicted depth map, and obtain the reconstructed three-dimensional point cloud through depth filtering and fusion operations.

2. The efficient multi-view stereoscopic reconstruction method for outdoor scenes according to claim 1, characterized in that: The first multi-view stereo reconstruction dataset in step 1 is a DTU dataset, which obtains multi-view images, depth maps and corresponding camera internal and external parameters of each scene.

3. The efficient multi-view stereoscopic reconstruction method for outdoor scenes according to claim 1, characterized in that: The backbone network in step 2 is a feature pyramid feature extraction network, which is used to extract multi-scale features of the input reference image and the source image; The large core attention based adaptive feature extraction converter module includes an adaptive large core attention block, a hybrid feedforward neural network block, a linear cross attention block and an upsampling module; The cost regularization module, by replacing the 3D convolution kernel with a 3D adaptive large kernel attention, can expand the receptive field without increasing model parameters and computational complexity, so that the model can better capture the global information in the scene, thereby improving the accuracy of stereo reconstruction; The depth estimation module uses the maximum probability depth estimation to select, for each pixel position, a depth value with the highest probability value from the probability body and use it as the depth value of the pixel position in the depth estimation map.

4. The efficient multi-view stereoscopic reconstruction method for outdoor scenes according to claim 3, characterized in that: The adaptive large-core attention block in step 2 uses multi-scale convolution kernels with different expansion coefficients to form a large-size convolution kernel, so that the model can effectively reduce the computational cost and improve the computational efficiency of the model without sacrificing performance. On this basis, by introducing deformable convolution, the large-size convolution kernel can adaptively learn multi-view targets of different scales and obtain a larger receptive field, so that the model can better understand the target context information in outdoor scenes, thereby improving the ability to extract complex features; The hybrid feedforward neural network block better aggregates global and multi-scale local information and extends the design of layer normalization to a general form, which promotes the optimization of skip connections and improves the network feature extraction capability. The linear cross attention block enables the network to learn the feature association between images from different perspectives, helping the model to better understand the spatial structure of the object and improve the accuracy and stability of stereo reconstruction.

5. The efficient multi-view stereoscopic reconstruction method for outdoor scenes according to claim 1, characterized in that: In step 3, the preset thresholds in the training network model include a preset value of the loss function, a preset value of the number of iterations, and a preset value of the optimized learning rate.

6. The efficient multi-view stereoscopic reconstruction method for outdoor scenes according to claim 1, characterized in that: The loss function in step 3 is focal loss to strengthen single hot spot supervision in the blurred area.

7. The efficient multi-view stereoscopic reconstruction method for outdoor scenes according to claim 1, characterized in that: The step 4 also includes evaluating the accuracy and completeness of the algorithm reconstruction results through evaluation indicators during the training of the network model.

8. The efficient multi-view stereoscopic reconstruction method for outdoor scenes according to claim 1, characterized in that: The second multi-view stereo reconstruction dataset in step 5 is a BlendedMVS dataset.

9. A multi-view stereo reconstruction system for outdoor scenes, characterized in that: The system comprises: An image acquisition module, used to acquire training data and test data; the training data and test data include multi-view images and corresponding depth images; A model building module is used to build the network model proposed by this method and initialize the model weights; A data set and model loading module, used to load input images and network models onto a hardware device for model training; the hardware device refers to an image processor; A model training module is used to perform supervised model training on the network model constructed by this method for the multi-view images of each scene in the training data to obtain the optimal model weight; A depth estimation module is used to perform depth estimation on the multi-view images of each scene in the test data through a network model equipped with optimal model weights to obtain a multi-view depth map of each scene; The depth filtering and fusion module is used to obtain a 3D point cloud file for each scene after performing depth filtering and fusion operations on the multi-view depth map predicted by the network; Point cloud visualization module, used to parse and process 3D point cloud files through the MeshLab software platform to achieve 3D point cloud visualization; A storage medium for storing a multi-view stereo reconstruction system; Applied to the efficient multi-view stereo reconstruction method for outdoor scenes as described in any one of claims 1 to 8.

Images