Picture identification and reconstruction method for whole house customized cabinet

Abstract

The invention discloses a picture recognition and reconstruction method for a whole house customized cabinet, and belongs to the technical field of picture defect detection, and the method comprises the steps: selecting a picture sample from a sample training set, and carrying out the image marking of different function partitions of a cabinet body in the picture sample through pyQt; inputting the marked picture into a YOLO model, training the YOLO model in combination with an AdamW optimizer, and outputting trained picture data; the loss weight of the image annotation is increased during training; performing post-processing operation on the picture data; and generating a cabinet model according to a data result of the post-processing operation. According to the invention, through combination of target detection, key point detection and semantic segmentation technologies, the recognition capability of a non-standard cabinet body and a complex scene is significantly improved. By automatically processing the picture to generate the model, manual measurement and design errors are reduced, the design period is remarkably shortened, and the technical threshold of customized furniture design is lowered.

Description

Technical Field

[0001] This invention belongs to the field of image defect detection technology, specifically a method for image recognition and reconstruction of custom-made cabinets for the whole house. Background Technology

[0002] In the traditional field of cabinet furniture design and manufacturing, customization often relies on repeated communication between designers and clients, on-site manual measurements, and 3D modeling by designers based on experience and 2D drawings (such as CAD). This process is not only time-consuming and labor-intensive, but also heavily dependent on the designer's personal skills and experience. Clients cannot easily and quickly see the final effect, and designers are prone to repeated design revisions or even errors due to communication mistakes or measurement deviations. This inefficient, "primitive" model significantly lengthens the design cycle, increases costs and error rates, and fails to meet the urgent needs of modern consumers for efficient, precise, and visual customization.

[0003] In recent years, to address the efficiency bottlenecks of traditional design, a technical solution based on image generation for cabinet models has emerged. The typical workflow of this type of solution is usually as follows:

[0004] (1) Acquire images of the cabinet;

[0005] (2) Use image processing or deep learning algorithms (such as edge detection, feature extraction, target recognition) to analyze images and identify cabinet components (such as door panels, drawers, handles) and their general structure;

[0006] (3) Based on the recognition results, combined with the preset standard cabinet component library and size rules, automatically assemble or generate a three-dimensional cabinet model.

[0007] However, existing technical solutions generally suffer from the following key shortcomings:

[0008] Highly reliant on preset rules and standard libraries: The model generation process is severely limited by predefined standard component libraries and matching rules. When faced with non-standard cabinets (irregular structures, special connection methods, complex decorations), parts occluded in the image, uneven lighting, or poor shooting angles, the system's recognition accuracy drops sharply. The generated models are often distorted, lack key details, or are simply unprocessable, resulting in poor flexibility and difficulty in meeting true personalized customization needs.

[0009] The generated models lack accuracy and manufacturability: Even when visually similar 3D models are generated, the results are usually "black box" closed meshes or voxels, lacking accurate, editable parametric dimensional information (such as precise sheet thickness, hole coordinates, and connector specifications). This makes it difficult to directly use the models to guide automated production (such as CNC machining and BOM generation), and subsequent modifications are also very difficult, making it impossible to seamlessly integrate with downstream manufacturing processes.

[0010] Based on this, an image recognition and reconstruction method for whole-house custom cabinets is provided. Summary of the Invention

[0011] The purpose of this invention is to provide an image recognition and reconstruction method for whole-house custom cabinets, addressing at least one aspect of the problems and shortcomings mentioned in the background art. This includes the significant limitations of existing technologies in handling highly personalized customization needs and generating accurate, editable production-grade models. This invention, by integrating advanced image understanding, geometric reasoning, and parametric modeling technologies, can efficiently and accurately generate parametric 3D cabinet models directly usable for manufacturing from user-provided images. This technology will bring significant macro-social benefits: greatly reducing the design threshold and cycle time for custom furniture, improving the level of automation and intelligence in the industry; reducing manual measurement and design errors, saving social resources; promoting the upgrade of the "what you see is what you get" personalized consumer experience, meeting people's growing demand for a better home life; and ultimately driving the furniture manufacturing industry towards efficiency, flexibility, and digitalization.

[0012] To achieve the above objectives, the present invention provides the following technical solution:

[0013] A method for image recognition and reconstruction for whole-house custom cabinets, comprising:

[0014] Image samples were selected from the training set, and image annotations of different functional areas of the cabinet in the image samples were performed using pyQt.

[0015] The labeled images are input into the YOLO model, and the AdamW optimizer is used to train the YOLO model, outputting the trained image data; and the loss weights of image annotation are added during training.

[0016] Perform post-processing operations on image data;

[0017] A cabinet model is generated based on the data results from the post-processing operation.

[0018] As a further aspect of the present invention, it also includes:

[0019] The YOLO model's detection results were evaluated using a confusion matrix.

[0020] As a further aspect of the present invention: post-processing operations are performed on the image data, including:

[0021] The system performs Region of Interest (ROI) identification on image data. When multiple ROIs exist in an image, they are categorized based on their location. It checks for the presence of baseboards or top panels; if present, these are added to the ROIs. Detection results of the same type are categorized by size, and their distribution is calculated using statistical methods, including automatic calculation using K-Means clustering. Elements in the same cluster are aligned to the same size, and their horizontal and vertical coordinates are matched. Other detected portions are filled into the ROIs, with NMS filtering for overlapping detections and fine-tuning the coordinates of remaining overlapping detections. The output image consists of several aligned rectangular regions, with all ROIs filled. For regions not yet filled, their types are defined according to requirements.

[0022] As a further aspect of the present invention: generating a cabinet model based on the data results of post-processing operations, including:

[0023] The rectangular areas in the post-processing output data are converted into rectangular wireframes, and collinear and touching line segments are merged; board material is generated at the line segments, and the corresponding cabinet functional components are filled in the areas; after all the generation is completed, the cabinet model is successfully generated.

[0024] According to the present invention, at least the following technical effects are achieved:

[0025] By combining object detection, key point detection, and semantic segmentation technologies, the system significantly improves its ability to recognize non-standard cabinets and complex scenes, handling issues such as occlusion and uneven lighting to meet personalized customization needs. The output 3D model includes editable parametric dimensional information, such as board thickness and cabinet functional components, enabling direct use in automated production and subsequent modifications, thus solving the problem of unproducible models generated by traditional technologies. Automated image processing for model generation reduces manual measurement and design errors, significantly shortening the design cycle and lowering the technical threshold for customized furniture design. This drives the furniture manufacturing industry towards efficiency, flexibility, and digitalization, while enhancing personalized consumer experiences and meeting people's demands for a better home life. Attached Figure Description

[0026] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0027] Figure 1 This is a flowchart of the method of the present invention;

[0028] Figure 2 This is a schematic diagram of the model training and evaluation data visualization of the present invention;

[0029] Figure 3 This is a visual diagram illustrating the data processing of the customized cabinet according to the present invention. Detailed Implementation

[0030] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0031] In the description of this invention, it should be understood that the orientation descriptions, such as up, down, front, back, left, right, etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.

[0032] In the description of this invention, "several" means one or more, "more than" means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. The use of "first" and "second" in the description is merely for distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0033] In the description of this invention, unless otherwise explicitly defined, terms such as "set up," "install," and "connect" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this invention in conjunction with the specific content of the technical solution.

[0034] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention; that is, the described embodiments are merely some, not all, of the embodiments of this invention. The components of the embodiments of this invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0035] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0036] like Figure 1-3 The embodiment of the present invention shown provides a method for image recognition and reconstruction for whole-house custom cabinets, comprising:

[0037] Image samples were selected from the training set, and bounding boxes were used to annotate the different functional areas of the cabinet in the image samples using pyQt.

[0038] It should be noted that pyQt is a toolkit for creating GUI applications, a fusion of the Python programming language and the Qt library. Bounding box annotation is an image annotation tool; bounding box annotation is the process of tightly enclosing a specific target object in a digital image with a rectangular frame.

[0039] The labeled images are input into the YOLO model, and the AdamW Optimizer is used to train the YOLO model, outputting the trained image data. During training, the loss weight of the bounding box is added to improve the detection accuracy.

[0040] It's important to note that the AdamW (Adam with Weight Decay Fix or Adam with DecoupledWeight Decay) optimizer decouples weight decay (L2 regularization) from the adaptive learning rate mechanism. Weight decay is no longer part of the gradient calculation in the adaptive learning rate calculation; instead, it's added directly to the parameters during parameter updates as an independent term. In the YOLO model, AdamW significantly improves the generalization ability and convergence stability of the YOLO model by modifying the implementation of weight decay.

[0041] The YOLO model's detection results were evaluated using the Confusion Matrix. A higher bounding box IOUthreshold value indicates a better model. For cabinet detection, Recall L is more important than Percision. Therefore, the confidence threshold can be appropriately lowered during prediction to increase Recall L. The specific amount to lower can be determined using the Percision-Recall Curve.

[0042] It should be noted that the confusion matrix is ​​an important tool for evaluating the performance of classification models.

[0043] Post-processing of image data: This includes combining ROI region identification results with other region identification results, including:

[0044] When there are multiple ROI regions in an image, different regions are classified according to their location.

[0045] Check for the presence of kickboard or top panel areas; if present, add them to the ROI area first.

[0046] Classifying similar detection results by size can be done by using statistical methods to calculate their distribution, or by using clustering methods such as K-Means to calculate it automatically.

[0047] Elements in the same cluster are of the same size and aligned to the horizontal and vertical coordinates.

[0048] Fill the other detected parts into the ROI region, during which NMS is used to filter overlapping detections, and the coordinates of the remaining overlapping detections need to be fine-tuned to avoid overlap.

[0049] It should be noted that NMS stands for Non-Maximum Suppression, which suppresses elements that are not maxima.

[0050] The output should now consist of several aligned rectangular regions, with the ROI regions mostly filled. For the unfilled regions, define their types according to your requirements.

[0051] The cabinet model is generated based on the data results from the post-processing operation. This includes: converting the rectangular areas in the data results output by the above method into rectangular wireframes, and merging collinear and touching line segments; generating panels at the line segments, and filling the areas with the corresponding cabinet functional components. Once all the generation is complete, the cabinet model is successfully generated.

[0052] By combining object detection, key point detection, and semantic segmentation technologies, the system significantly improves its ability to recognize non-standard cabinets and complex scenes, handling issues such as occlusion and uneven lighting to meet personalized customization needs. The output 3D model includes editable parametric dimensional information, such as board thickness and cabinet functional components, enabling direct use in automated production and subsequent modifications, thus solving the problem of unproducible models generated by traditional technologies. Automated image processing for model generation reduces manual measurement and design errors, significantly shortening the design cycle and lowering the technical threshold for customized furniture design. This drives the furniture manufacturing industry towards efficiency, flexibility, and digitalization, while enhancing personalized consumer experiences and meeting people's demands for a better home life.

[0053] In summary, the technology of this invention has significant advantages in terms of recognition accuracy, model generation quality, production integration capability, and social benefits.

[0054] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

Claims

1. A method for image recognition and reconstruction for whole-house custom cabinets, characterized in that, include: Image samples were selected from the training set, and image annotations of different functional areas of the cabinet in the image samples were performed using pyQt. The labeled images are input into the YOLO model, and the AdamW optimizer is used to train the YOLO model, outputting the trained image data; and the loss weights of image annotation are added during training. Perform post-processing operations on image data; A cabinet model is generated based on the data results from the post-processing operation.

2. The image recognition and reconstruction method for whole-house custom cabinets according to claim 1, characterized in that, Also includes: The YOLO model's detection results were evaluated using a confusion matrix.

3. The image recognition and reconstruction method for whole-house custom cabinets according to claim 1, characterized in that, Post-processing operations on image data include: The system performs Region of Interest (ROI) identification on image data. When multiple ROIs exist in an image, they are categorized based on their location. It checks for the presence of baseboards or top panels; if present, these are added to the ROIs. Detection results of the same type are categorized by size, and their distribution is calculated using statistical methods, including automatic calculation using K-Means clustering. Elements in the same cluster are aligned to the same size, and their horizontal and vertical coordinates are matched. Other detected portions are filled into the ROIs, with NMS filtering for overlapping detections and fine-tuning the coordinates of remaining overlapping detections. The output image consists of several aligned rectangular regions, with all ROIs filled. For regions not yet filled, their types are defined according to requirements.

4. The image recognition and reconstruction method for whole-house custom cabinets according to claim 3, characterized in that, A cabinet model is generated based on the data results from the post-processing operation, including: The rectangular areas in the post-processing output data are converted into rectangular wireframes, and collinear and touching line segments are merged; board material is generated at the line segments, and the corresponding cabinet functional components are filled in the areas; after all the generation is completed, the cabinet model is successfully generated.

Images