An AI vision-based identification method for tracking and positioning of bamboo-plastic packaging boxes

Abstract

The application relates to the technical field of visual tracking, and discloses an AI vision-based bamboo-plastic packaging box tracking and positioning identification method, which comprises the following steps: performing scene light field analysis on an original image sequence of a bamboo-plastic packaging box to obtain environmental light field parameters; based on the material reflection characteristics of the bamboo-plastic packaging box, performing optical attribute extraction to obtain reflection characteristic parameters; based on the environmental light field parameters, performing optical response simulation on the reflection characteristic parameters of the bamboo-plastic packaging box to obtain visual samples; performing cross-domain feature alignment on the original image sequence and the visual samples to obtain joint feature expression; performing illumination interference decoupling on the joint feature expression to obtain illumination robust features; based on the illumination robust features, performing pose state evaluation on the spatial position and the attitude of the bamboo-plastic packaging box to obtain six-degree-of-freedom tracking and positioning data; and the application can improve the tracking and positioning identification efficiency of the bamboo-plastic packaging box.

Description

A recognition method for tracking and positioning bamboo-plastic packaging boxes based on AI vision Technical Field

[0001] This invention relates to the field of visual tracking technology, and in particular to a recognition method for tracking and locating bamboo-plastic packaging boxes based on AI vision. Background Technology

[0002] In logistics warehousing, intelligent manufacturing, and other scenarios, bamboo-plastic packaging boxes have become a core carrier for cargo storage and transportation due to their advantages such as lightweight, high strength, environmental friendliness, and recyclability. Their precise spatial pose tracking and positioning are the technological foundation for key processes such as automated sorting and intelligent warehousing, directly impacting logistics efficiency. Existing object tracking and positioning technologies are mostly based on image feature extraction and optical response analysis, deriving spatial pose by analyzing captured images. However, the surface reflectivity of bamboo-plastic packaging boxes and the complex changes in the ambient light field place higher demands on tracking and positioning accuracy and stability, necessitating targeted technical solutions.

[0003] In existing technologies, tracking and positioning solutions for bamboo-plastic packaging boxes struggle to accurately isolate the interference of dynamic changes in the ambient light field on feature extraction. This results in extracted object features being susceptible to fluctuations due to lighting conditions, thus affecting the accuracy of pose assessment. Furthermore, the lack of an effective cross-domain fusion mechanism between the visual features acquired through image acquisition and the inherent optical properties of the object's material makes it difficult to form a unified feature expression that comprehensively reflects the object's essential attributes. Consequently, the continuity and accuracy of the tracking and positioning process are limited. Therefore, improving the tracking, positioning, and recognition efficiency of bamboo-plastic packaging boxes has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides an AI vision-based identification method for tracking and locating bamboo-plastic packaging boxes to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, this invention provides a method for tracking and locating bamboo-plastic packaging boxes based on AI vision, comprising:

[0006] S1. Perform scene light field analysis on the original image sequence of the bamboo-plastic packaging box to obtain the ambient light field parameters of the bamboo-plastic packaging box;

[0007] S2. Based on the material reflection characteristics of the bamboo-plastic packaging box, optical properties of the bamboo-plastic packaging box are extracted to obtain the reflection characteristic parameters of the bamboo-plastic packaging box;

[0008] S3. Based on the ambient light field parameters, perform optical response simulation on the reflection characteristic parameters to obtain a visual sample of the bamboo-plastic packaging box;

[0009] S4. Perform cross-domain feature alignment between the original image sequence and the visual samples to obtain the joint feature representation of the bamboo-plastic packaging box;

[0010] S5. Perform illumination interference decoupling on the joint feature expression to obtain the illumination robust features of the bamboo-plastic packaging box;

[0011] S6. Based on the illumination robustness features, the spatial position and orientation of the bamboo-plastic packaging box are evaluated to obtain the six-degree-of-freedom tracking and positioning data of the bamboo-plastic packaging box.

[0012] In a preferred embodiment, scene light field analysis is performed on the original image sequence of the bamboo-plastic packaging box to obtain the ambient light field parameters of the bamboo-plastic packaging box, including:

[0013] Temporal extraction is performed on the original image sequence of the bamboo-plastic packaging box to obtain a stable image sequence of the bamboo-plastic packaging box;

[0014] The brightness of the image in the stable image sequence is equalized to obtain the brightness characterization of the bamboo-plastic packaging box.

[0015] The brightness characterization is subjected to multi-scale optical field feature analysis to obtain a multi-level descriptor of the optical field feature of the bamboo-plastic packaging box.

[0016] The scene light field description of the bamboo-plastic packaging box is obtained by scene fusion of the multi-level descriptor of the light field feature with the stable image sequence.

[0017] The ambient light field parameters of the bamboo-plastic packaging box are obtained by modal decomposition of the scene light field description.

[0018] In a preferred embodiment, the step of extracting optical properties of the bamboo-plastic packaging box based on its material reflectivity to obtain reflectivity parameters includes:

[0019] Multi-view collaborative analysis was performed on the original image sequence to obtain the reflection intensity distribution on the surface of the bamboo-plastic packaging box;

[0020] Based on the multispectral data in the original image sequence, the reflection intensity distribution is decomposed into spectral dimensions to obtain the high-dimensional spectral reflection profile of the bamboo-plastic packaging box.

[0021] The high-dimensional spectral reflectance profile is registered with a preset material optical property library to obtain the material optical properties of the bamboo-plastic packaging box.

[0022] Based on the optical properties of the material, the high-dimensional spectral reflectance profile is optically deconstructed to obtain the optical parameters of the bamboo-plastic packaging box;

[0023] The optical parameters are normalized to obtain the reflection characteristic parameters of the bamboo-plastic packaging box.

[0024] In a preferred embodiment, the step of optically deconstructing the high-dimensional spectral reflectance profile based on the material's optical properties to obtain the optical parameters of the bamboo-plastic packaging box includes:

[0025] Based on the optical properties of the material, the spectral composition of the high-dimensional spectral reflection profile is decoupled to obtain the diffuse reflection component and specular reflection component of the bamboo-plastic packaging box.

[0026] The diffuse reflection component and the specular reflection component are calibrated to obtain the reflection coefficient pair of the bamboo-plastic packaging box;

[0027] Based on the geometric distribution of the high-dimensional spectral reflectance profile, the reflectance coefficients are weighted and corrected to obtain the optical parameters of the bamboo-plastic packaging box.

[0028] In a preferred embodiment, the step of performing optical response simulation on the reflection characteristic parameters based on the ambient light field parameters to obtain a visual sample of the bamboo-plastic packaging box includes:

[0029] The illumination characteristics of the bamboo-plastic packaging box are obtained by extracting the illumination elements from the ambient light field parameters.

[0030] Based on the diffuse reflection coefficient and specular reflection coefficient in the reflection characteristic parameters, the light field illumination characteristics are simulated and traced in reverse to obtain the pixel-level irradiance distribution of the bamboo-plastic packaging box.

[0031] Based on the camera viewpoint parameters corresponding to the original image sequence, perspective projection mapping is performed on the pixel-level irradiance distribution, and the mapping result is rendered as a geometric projection image of the bamboo-plastic packaging box.

[0032] The geometric projection image is optimized for color gamut adaptation to obtain a visual sample of the bamboo-plastic packaging box.

[0033] In a preferred embodiment, cross-domain feature alignment is performed between the original image sequence and the visual samples to obtain a joint feature representation of the bamboo-plastic packaging box, including:

[0034] The original image sequence is characterized and converged to obtain the original features of the bamboo-plastic packaging box;

[0035] The visual samples are geometrically and topologically represented to obtain the simulated features of the bamboo-plastic packaging box;

[0036] Cross-domain feature alignment is performed on the original features and the simulated features to obtain the joint feature representation of the bamboo-plastic packaging box.

[0037] In a preferred embodiment, the joint feature representation is decoupled from illumination interference to obtain the illumination robust features of the bamboo-plastic packaging box, including:

[0038] The joint feature expression is subjected to light-sensitive decomposition to obtain the light-sensitive component and material intrinsic component of the bamboo-plastic packaging box;

[0039] Based on the ambient light field parameters, explicit interference quantization is performed on the light-sensitive components to obtain the light interference characteristics of the bamboo-plastic packaging box.

[0040] Interference suppression fusion is performed on the intrinsic components of the material and the illumination interference features to obtain the initial non-illuminated feature expression of the bamboo-plastic packaging box;

[0041] The initial unlit feature representation is optimized for invariance to obtain the illumination robust features of the bamboo-plastic packaging box.

[0042] In a preferred embodiment, the step of evaluating the spatial position and orientation of the bamboo-plastic packaging box based on the illumination robustness characteristics to obtain six-DOF tracking and positioning data of the bamboo-plastic packaging box includes:

[0043] The key geometric features of the bamboo-plastic packaging box are obtained by performing structure-aware extraction on the illumination robust features.

[0044] By deriving the constraint relationships of the key geometric features, the spatial constraints of the bamboo-plastic packaging box are obtained.

[0045] The spatial constraints are subjected to temporal smoothing to obtain the optimized pose sequence of the bamboo-plastic packaging box;

[0046] The optimized pose sequence is serialized and encapsulated into six-degree-of-freedom tracking and positioning data for the bamboo-plastic packaging box.

[0047] In a preferred embodiment, constraint relationships are derived from the key geometric features to obtain the spatial constraints of the bamboo-plastic packaging box, including:

[0048] Geometric correlation calibration is performed on the key geometric features to obtain feature matching pairs for the bamboo-plastic packaging box;

[0049] The geometric parameters of the feature matching pairs are calculated to obtain the preliminary constraint parameters of the bamboo-plastic packaging box.

[0050] The spatial constraints of the bamboo-plastic packaging box are obtained by effectively consolidating the preliminary constraint parameters.

[0051] In a preferred embodiment, the calculation formula for the preliminary constraint parameters is as follows:

[0052] ;

[0053] In the formula, These are the initial constraint parameters. For the first The unit normal vector of each of the key geometric features, For the first feature matching pair Coordinates of a point in three-dimensional space. The total number of the feature matching pairs. For Euclidean norm operators The preset numerical stability constant is... The coordinates of the three-dimensional space point The weighted centroid.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] 1. This invention obtains ambient light field parameters through scene light field analysis and reflection characteristic parameters through optical property analysis. Visual samples are generated through optical response simulation. Combined with the original image sequence, cross-domain feature alignment is completed to form joint features. Then, illumination robust features are obtained through illumination interference decoupling. This completely avoids the interference of dynamic changes in the ambient light field on feature extraction, ensures the reliability of the basic data for pose assessment, effectively removes illumination interference, and improves the accuracy of pose assessment.

[0056] 2. This invention obtains the reflective properties parameters of bamboo-plastic packaging box material through multi-view collaborative analysis and spectral dimension decomposition. The original image sequence is represented and converged to obtain visual features, and the geometric topology of visual samples is represented to obtain simulation features. The two are aligned and fused across domains to form a unified joint feature expression, which fully reflects the essential attributes of the object. It breaks through the limitations of insufficient fusion in existing technologies, realizes the cross-domain fusion of visual and material features, and ensures the continuity and accuracy of tracking. Attached Figure Description

[0057] Figure 1 is a flowchart illustrating an AI vision-based identification method for tracking and locating bamboo-plastic packaging boxes according to an embodiment of the present invention.

[0058] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0059] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0060] This application provides a method for tracking and locating bamboo-plastic packaging boxes based on AI vision. The execution subject of this AI vision-based bamboo-plastic packaging box tracking and locating method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the AI ​​vision-based bamboo-plastic packaging box tracking and locating method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0061] Referring to Figure 1, a flowchart illustrating a method for tracking and locating bamboo-plastic packaging boxes based on AI vision, according to an embodiment of the present invention, is shown. In this embodiment, the method for tracking and locating bamboo-plastic packaging boxes based on AI vision includes:

[0062] S1. Perform scene light field analysis on the original image sequence of the bamboo-plastic packaging box to obtain the ambient light field parameters of the bamboo-plastic packaging box;

[0063] In this embodiment of the invention, scene light field analysis is performed on the original image sequence of the bamboo-plastic packaging box to obtain the ambient light field parameters of the bamboo-plastic packaging box, including:

[0064] Temporal extraction is performed on the original image sequence of the bamboo-plastic packaging box to obtain a stable image sequence of the bamboo-plastic packaging box;

[0065] The brightness of the image in the stable image sequence is equalized to obtain the brightness characterization of the bamboo-plastic packaging box.

[0066] The brightness characterization is subjected to multi-scale optical field feature analysis to obtain a multi-level descriptor of the optical field feature of the bamboo-plastic packaging box.

[0067] The scene light field description of the bamboo-plastic packaging box is obtained by scene fusion of the multi-level descriptor of the light field feature with the stable image sequence.

[0068] The ambient light field parameters of the bamboo-plastic packaging box are obtained by modal decomposition of the scene light field description.

[0069] When performing time-series extraction on the original image sequence of bamboo-plastic packaging boxes, the original image sequence is a collection of images containing bamboo-plastic packaging boxes taken continuously. According to the shooting time sequence of the images, the position and shape of the bamboo-plastic packaging boxes in each image are checked one by one. Images with no positional shift and no shape distortion of the bamboo-plastic packaging boxes are selected, while images whose information about the bamboo-plastic packaging boxes is distorted due to factors such as camera shake or temporary obstruction are removed. The final stable image sequence is a series of continuous images in which the key visual information of the bamboo-plastic packaging boxes remains consistent.

[0070] When equalizing the brightness of images in a stable image sequence, there are differences in brightness among the images in the sequence. For the pixel brightness data of each image, the brightness value of the bright area is adjusted so that it does not exceed the maximum brightness threshold that the human eye can clearly distinguish. At the same time, the brightness value of the dark area is increased so that it is not lower than the minimum brightness threshold that can present the details of the bamboo plastic packaging box. Through this adjustment, the brightness distribution of the entire image is made uniform, without areas that are too bright to cause loss of details or too dark to be distinguishable. The resulting brightness representation is an image representation that can accurately reflect the light intensity distribution of different areas on the surface of the bamboo plastic packaging box.

[0071] When performing multi-scale light field feature analysis on the light representation, the light representation clearly shows the light intensity distribution on the surface of the bamboo-plastic packaging box. According to the preset scale division standard, the light representation is divided into three levels: small scale, medium scale, and large scale. The small scale corresponds to the texture, edges and corners of the bamboo-plastic packaging box surface, the medium scale corresponds to the local overall area such as the sides and top of the bamboo-plastic packaging box, and the large scale corresponds to the overall scene area including the bamboo-plastic packaging box and a small amount of surrounding environment. The light intensity change trend, brightness gradient and other light field-related features at each scale are extracted. Then, these light field features at different scales are integrated to form a multi-level descriptor of light field features, which is a comprehensive descriptive information that can fully cover the light field characteristics of the bamboo-plastic packaging box at different scales.

[0072] When scene fusion is performed on the multi-level descriptor of light field features and the stable image sequence, the multi-level descriptor of light field features provides light field features at different scales, while the stable image sequence provides actual scene information of the bamboo-plastic packaging box. The light field features at each scale in the multi-level descriptor of light field features are accurately matched to the corresponding areas of the bamboo-plastic packaging box in the stable image sequence according to their corresponding scale range. This makes the light field features correspond to the scene information such as the outline and position of the bamboo-plastic packaging box, eliminating the disconnect between the light field features and the actual scene. The final scene light field description is a complete light field data record that integrates the specific scene information of the bamboo-plastic packaging box and the multi-scale light field features.

[0073] When performing modal decomposition on the scene light field description, the scene light field description contains comprehensive light field information of the environment in which the bamboo plastic packaging box is located. This light field information consists of different types of light field components, including direct light that directly illuminates the surface of the bamboo plastic packaging box, reflected light that illuminates the bamboo plastic packaging box after being reflected by other objects, and scattered light that is scattered when propagating in the air. Through a specific separation method, these different modes of light field components are separated one by one, and the relevant information of each mode of light field is extracted separately. The resulting environmental light field parameters are a set of specific information that can characterize the light field characteristics of each mode of light field, such as direct light, reflected light, and scattered light.

[0074] The beneficial effects are as follows: it removes distorted images from the original image sequence, ensuring a reliable and consistent image basis for subsequent light field analysis, providing high-quality initial data for scene light field analysis, solving the problem of uneven brightness in stable image sequences, clearly presenting the surface illumination intensity of bamboo-plastic packaging boxes, improving the accuracy of light field feature extraction, comprehensively capturing light field information at different scales, covering both detailed and overall scene light field characteristics, providing rich and hierarchical light field data, achieving accurate integration of light field features with actual scene information, making the light field description fit the application scenario, providing correlated and complete light field data for modal decomposition, separating different types of light field components in the environment, clarifying the characteristics of each light field, and providing accurate and detailed parameter basis for optical response simulation.

[0075] S2. Based on the material reflection characteristics of the bamboo-plastic packaging box, optical properties of the bamboo-plastic packaging box are extracted to obtain the reflection characteristic parameters of the bamboo-plastic packaging box;

[0076] In this embodiment of the invention, the step of extracting optical properties of the bamboo-plastic packaging box based on its material reflectivity to obtain reflectivity parameters includes:

[0077] Multi-view collaborative analysis was performed on the original image sequence to obtain the reflection intensity distribution on the surface of the bamboo-plastic packaging box;

[0078] Based on the multispectral data in the original image sequence, the reflection intensity distribution is decomposed into spectral dimensions to obtain the high-dimensional spectral reflection profile of the bamboo-plastic packaging box.

[0079] The high-dimensional spectral reflectance profile is registered with a preset material optical property library to obtain the material optical properties of the bamboo-plastic packaging box.

[0080] Based on the optical properties of the material, the high-dimensional spectral reflectance profile is optically deconstructed to obtain the optical parameters of the bamboo-plastic packaging box;

[0081] The optical parameters are normalized to obtain the reflection characteristic parameters of the bamboo-plastic packaging box.

[0082] The process of optically deconstructing the high-dimensional spectral reflectance profile based on the optical properties of the material to obtain the optical parameters of the bamboo-plastic packaging box includes:

[0083] Based on the optical properties of the material, the spectral composition of the high-dimensional spectral reflection profile is decoupled to obtain the diffuse reflection component and specular reflection component of the bamboo-plastic packaging box.

[0084] The diffuse reflection component and the specular reflection component are calibrated to obtain the reflection coefficient pair of the bamboo-plastic packaging box;

[0085] Based on the geometric distribution of the high-dimensional spectral reflectance profile, the reflectance coefficients are weighted and corrected to obtain the optical parameters of the bamboo-plastic packaging box.

[0086] The original image sequence is a continuous collection of images of bamboo-plastic packaging boxes taken from different angles. Multi-view collaborative analysis requires extracting the brightness data of each pixel on the surface of the bamboo-plastic packaging box from each viewpoint. This brightness data directly corresponds to the degree of light reflection at that point. Then, the brightness data of the corresponding pixels from all views are integrated and compared to eliminate the deviation caused by occlusion or uneven lighting from a single viewpoint. The final reflection intensity distribution is a complete data record that comprehensively presents the intensity of reflected light at various locations on the surface of the bamboo-plastic packaging box.

[0087] Multispectral data is image information corresponding to different wavelengths of light contained in the original image sequence. Based on this data, the spectral dimension decomposition of the reflection intensity distribution is performed. The reflection intensity data at each position in the reflection intensity distribution needs to be split according to different wavelengths of light, and the reflection intensity value at each position under each wavelength needs to be extracted. Then, the reflection intensity values ​​under all wavelengths are arranged in wavelength order to form a high-dimensional spectral reflection profile, which is a set of multi-dimensional data curves that reflect the reflection characteristics of different positions on the surface of bamboo plastic packaging boxes under the illumination of various wavelengths of light.

[0088] The pre-set material optical property library is a collection of standard spectral reflectance data of various known materials under different wavelengths of light. Each material corresponds to a unique standard spectral reflectance feature. The high-dimensional spectral reflectance profile is compared one by one with the standard spectral reflectance data of each material in the library to find the standard spectral reflectance data with the highest similarity to the high-dimensional spectral reflectance profile feature. The optical properties of the material corresponding to the standard spectral reflectance data are the material optical properties of the bamboo-plastic packaging box, specifically including the inherent optical properties of the material such as absorption, reflection, and transmission of light of different wavelengths.

[0089] Diffuse reflection is the phenomenon where light is uniformly reflected in all directions after it hits the surface of an object. Specular reflection is the phenomenon where light is reflected along a specific direction after it hits the surface of an object. Based on the inherent characteristics of light reflection in the optical properties of the material, the inherent performance rules of diffuse reflection and specular reflection of the material are clarified. Then, according to these rules, the reflection intensity data at each wavelength in the high-dimensional spectral reflection profile are separated. The part that can reflect the uniform reflection characteristics is divided into diffuse reflection components, and the part that can reflect the reflection characteristics in a specific direction is divided into specular reflection components. The two are spectral data sets that independently reflect the diffuse reflection and specular reflection effects of the bamboo-plastic packaging box surface.

[0090] Attribute calibration requires quantification and definition of diffuse reflection and specular reflection components separately. Based on the inherent intensity range of diffuse and specular reflection in the material's optical properties, the quantized value corresponding to the diffuse reflection component is determined. This value accurately reflects the intensity of diffuse reflection and is called the diffuse reflection coefficient. Similarly, the quantized value corresponding to the specular reflection component is determined. This value accurately reflects the intensity of specular reflection and is called the specular reflection coefficient. The pair of data composed of these two coefficients is the reflection coefficient pair.

[0091] The geometric distribution of high-dimensional spectral reflectance profiles refers to the spatial distribution characteristics of the data curves in each dimension, such as their shape, peak position, and trend of change. Based on these characteristics, the differences in reflectance characteristics of different areas on the surface of bamboo-plastic packaging boxes are judged. For areas with relatively stable reflectance characteristics, a lower correction weight is assigned to the reflectance coefficient pair. For areas with large changes in reflectance characteristics, a higher correction weight is assigned to the reflectance coefficient pair. Then, the diffuse reflectance coefficient and specular reflectance coefficient in the reflectance coefficient pair are numerically adjusted according to the corresponding weights so that the adjusted reflectance coefficient pair more accurately matches the actual reflectance of the surface of the bamboo-plastic packaging box. The adjusted reflectance coefficient pair is the optical parameter.

[0092] Normalization requires adjusting the values ​​of diffuse reflection coefficient and specular reflection coefficient in the optical parameters to a preset fixed range. During the adjustment process, the relative proportional relationship between the two coefficients remains unchanged. Specifically, this is achieved by calculating the ratio of each coefficient to the maximum value of that type of coefficient. The diffuse reflection coefficient and specular reflection coefficient obtained after adjustment are the reflection characteristic parameters. These parameters can accurately reflect the reflection characteristics of bamboo-plastic packaging boxes under a unified standard.

[0093] The beneficial effects include: integrating multi-view data to avoid single-view bias, obtaining accurate reflection intensity distribution, providing a reliable foundation for subsequent processing; splitting data by wavelength to clearly present multi-wavelength reflection characteristics, facilitating accurate material identification and subsequent decoupling; relying on attribute library comparison to quickly determine material optical properties, providing key basis for subsequent processing; separating two reflection components to avoid mutual interference, providing clean data for obtaining reflection coefficient pairs; quantifying and calibrating reflection components to accurately quantify reflection characteristics, providing a clear numerical basis for correction; being intuitive and operable; combining geometric distribution weighted correction to adapt to surface reflection differences, improving the fit between optical parameters and reality; and normalizing processing to unify the numerical range, eliminating scale differences, facilitating subsequent fusion, and improving parameter universality.

[0094] S3. Based on the ambient light field parameters, perform optical response simulation on the reflection characteristic parameters to obtain a visual sample of the bamboo-plastic packaging box;

[0095] In this embodiment of the invention, the step of performing optical response simulation on the reflection characteristic parameters based on the ambient light field parameters to obtain a visual sample of the bamboo-plastic packaging box includes:

[0096] The illumination characteristics of the bamboo-plastic packaging box are obtained by extracting the illumination elements from the ambient light field parameters.

[0097] Based on the diffuse reflection coefficient and specular reflection coefficient in the reflection characteristic parameters, the light field illumination characteristics are simulated and traced in reverse to obtain the pixel-level irradiance distribution of the bamboo-plastic packaging box.

[0098] Based on the camera viewpoint parameters corresponding to the original image sequence, perspective projection mapping is performed on the pixel-level irradiance distribution, and the mapping result is rendered as a geometric projection image of the bamboo-plastic packaging box.

[0099] The geometric projection image is optimized for color gamut adaptation to obtain a visual sample of the bamboo-plastic packaging box.

[0100] The ambient light field parameters were previously obtained through scene light field analysis, including light field information of different modes such as direct light, reflected light, and scattered light. Illumination element extraction is to extract the key components that affect the illumination effect on the surface of bamboo-plastic packaging boxes from these different modes of light field information. Specifically, these include the direction of light, the intensity of light, and the distribution range of light. The light field illumination characteristics are the collection of these key components, which can accurately reflect the actual illumination of bamboo-plastic packaging boxes by ambient light.

[0101] The reflection characteristic parameters were previously obtained through optical property analysis. The diffuse reflection coefficient reflects the intensity of light uniformly reflected from the surface of the bamboo-plastic packaging box, while the specular reflection coefficient reflects the intensity of light reflected along a specific direction. The reverse ray simulation traces the light propagation path from the surface of the bamboo-plastic packaging box, and combines the diffuse reflection coefficient to determine the propagation direction and energy attenuation of each reverse ray under uniform reflection. Combined with the specular reflection coefficient, the propagation path and energy change of each reverse ray under specific directional reflection are determined. Through this reverse derivation, the total light energy received by each pixel on the surface of the bamboo-plastic packaging box is determined. The pixel-level irradiance distribution is a specific data record of the received light energy corresponding to each pixel on the surface of the bamboo-plastic packaging box. Each pixel has a unique energy value, accurately reflecting the difference in light energy between different pixels.

[0102] The original image sequence is a continuous set of images obtained by photographing bamboo plastic packaging boxes. The camera viewpoint parameters are the specific position, shooting angle, and shooting distance of the camera when these images were taken. Perspective projection mapping projects the light energy data of each pixel in the pixel-level irradiance distribution onto a two-dimensional plane consistent with the shooting angle of the original image sequence, according to the shooting angle and distance determined by the camera viewpoint parameters. This ensures that the position and light energy correspondence of each point after projection matches the actual shooting scene. Rendering transforms the light energy data on the two-dimensional plane obtained after projection into a visualized image form, maintaining the brightness performance corresponding to the light energy at each position during the transformation process. The geometric projection image is a two-dimensional image with the geometric outline of the bamboo plastic packaging box and the corresponding light intensity distribution, and its geometric shape and perspective are consistent with the original image sequence.

[0103] Geometric projection images are two-dimensional images with geometric contours and illumination distributions. Color gamut adaptation optimization first determines the color range and color performance characteristics presented by the original image sequence, and then adjusts the color performance of the geometric projection image to be consistent with the color range of the original image sequence. This ensures that the color style of the geometric projection image matches the original image sequence, while correcting any color deviations that may exist in the geometric projection image. This results in natural color transitions and accurate color reproduction in the image. The visual sample is a two-dimensional image that, after color adjustment, retains accurate geometric contours and illumination distributions and is highly consistent with the color performance of the original image sequence. It can realistically simulate the visual presentation effect of bamboo plastic packaging boxes under actual ambient light.

[0104] The beneficial effects are as follows: key lighting elements are extracted, and the light field illumination characteristics comprehensively reflect the illumination situation, providing a precise and comprehensive lighting foundation for reverse ray simulation tracing, ensuring that the simulated lighting is consistent with reality. By combining two types of reflection coefficients for reverse ray simulation tracing, the surface reflection law is restored, and pixel-level irradiance distribution provides detailed and accurate lighting data, avoiding simulation result deviations. Perspective projection mapping is completed based on camera viewpoint parameters, ensuring that the viewpoint and shape of the geometric projection image are consistent with the original image sequence. The rendered visualization image provides a consistent carrier for color adaptation optimization. Color gamut adaptation optimization achieves a high degree of color matching between the geometric projection image and the original image sequence, correcting deviations. The visual samples realistically simulate the actual visual effect, improving the accuracy of subsequent cross-domain feature alignment.

[0105] S4. Perform cross-domain feature alignment between the original image sequence and the visual samples to obtain the joint feature representation of the bamboo-plastic packaging box;

[0106] In this embodiment of the invention, cross-domain feature alignment is performed between the original image sequence and the visual samples to obtain the joint feature representation of the bamboo-plastic packaging box, including:

[0107] The original image sequence is characterized and converged to obtain the original features of the bamboo-plastic packaging box;

[0108] The visual samples are geometrically and topologically represented to obtain the simulated features of the bamboo-plastic packaging box;

[0109] Cross-domain feature alignment is performed on the original features and the simulated features to obtain the joint feature representation of the bamboo-plastic packaging box.

[0110] The original image sequence is a collection of previously captured images containing bamboo-plastic packaging boxes. First, the key visual content that can be used for identification, such as the outline, surface texture, and color distribution of the bamboo-plastic packaging box, is extracted from each image. Then, these key visual contents of all images are integrated in chronological order of shooting time, and duplicate information is removed. The core content that can fully reflect the inherent visual attributes of the bamboo-plastic packaging box is retained. The final original feature is a collection of all effective key visual information, which fully presents the inherent visual characteristics of the bamboo-plastic packaging box in the actual shooting scene.

[0111] The visual sample is a two-dimensional image of a bamboo-plastic packaging box obtained through optical response simulation, which is consistent with the color representation of the original image sequence. First, the geometric structure of the bamboo-plastic packaging box in the visual sample is analyzed to determine the specific positions of each vertices, edges, planes and other geometric elements and their connection relationships. Then, based on the positions and connection relationships of these geometric elements, a feature set that can reflect the spatial structure organization of the bamboo-plastic packaging box is constructed. In the construction process, the adjacent, containment and connection relationships between geometric elements are accurately captured. The final simulation features are a set that accurately reflects the spatial structure features of the bamboo-plastic packaging box and fully presents the geometric topological properties of the bamboo-plastic packaging box.

[0112] The original feature set contains the inherent key visual information of the bamboo-plastic packaging box as captured by actual photography. The simulation features accurately reflect the spatial structural features of the bamboo-plastic packaging box. First, the visual information related to the outline of the bamboo-plastic packaging box in the original features and the corresponding geometric outline features in the simulation features are identified. The outline features of the two are accurately matched. Then, based on the matching relationship of the outline features, the visual information such as surface texture and color distribution in the original features are sequentially associated with the corresponding geometric structural areas in the simulation features. This ensures that each piece of visual information can be accurately matched to the corresponding geometric structural position. The final joint feature expression is a unified feature set that integrates the actual visual information and spatial structural features of the bamboo-plastic packaging box, achieving an organic combination of features from two different sources.

[0113] The beneficial effects are as follows: extracting effective visual information from the original image sequence and removing redundancy, so that the original features can fully and accurately reflect the inherent visual attributes of bamboo plastic packaging boxes, providing high-quality data for cross-domain alignment, ensuring alignment accuracy, analyzing the geometric structure and element relationships of visual samples, and clearly presenting spatial topological attributes with simulation features, laying the foundation for matching with the original features, improving alignment feasibility, associating the original features and simulation features with contour features, eliminating domain differences, and combining the advantages of both feature sets, providing comprehensive and unified feature support for subsequent decoupling of illumination interference.

[0114] S5. Perform illumination interference decoupling on the joint feature expression to obtain the illumination robust features of the bamboo-plastic packaging box;

[0115] In this embodiment of the invention, the joint feature representation is decoupled from illumination interference to obtain the illumination robust features of the bamboo-plastic packaging box, including:

[0116] The joint feature expression is subjected to light-sensitive decomposition to obtain the light-sensitive component and material intrinsic component of the bamboo-plastic packaging box;

[0117] Based on the ambient light field parameters, explicit interference quantization is performed on the light-sensitive components to obtain the light interference characteristics of the bamboo-plastic packaging box.

[0118] Interference suppression fusion is performed on the intrinsic components of the material and the illumination interference features to obtain the initial non-illuminated feature expression of the bamboo-plastic packaging box;

[0119] The initial unlit feature representation is optimized for invariance to obtain the illumination robust features of the bamboo-plastic packaging box.

[0120] The joint feature expression is a unified feature set that integrates the actual visual information and spatial structural features of bamboo-plastic packaging boxes. First, a reference standard for illumination changes is determined. This standard is based on the feature change patterns of bamboo-plastic packaging boxes under different known light intensities and directions. Then, each feature in the joint feature expression is examined one by one to determine whether the feature changes with changes in light intensity and direction. Features that change due to illumination changes are classified as light-sensitive components. This component is the set of features in the joint feature expression that changes only due to illumination factors and can directly reflect the influence of illumination on the joint feature expression. Features that are not affected by illumination changes and remain stable are classified as material intrinsic components. This component is the set of features in the joint feature expression that reflects the inherent properties of the bamboo-plastic packaging box material itself and does not change with changes in illumination conditions.

[0121] Ambient light field parameters are a set of specific information characterizing the properties of each mode of light field, such as direct light, reflected light, and scattered light. Illumination-sensitive components are a set of features in the joint feature expression that change only due to illumination factors. First, key information such as illumination intensity and direction of illumination of different modes of light field is extracted from the ambient light field parameters. Then, these key information are correlated with each feature in the illumination-sensitive components. Using the value of the feature under the original illumination conditions as a benchmark, the difference between the value of the feature under other illumination conditions and the benchmark value is calculated. The magnitude of the difference directly reflects the degree of illumination interference of the feature. The quantitative results of the interference degree of all illumination-sensitive features are integrated to form an illumination interference feature, which is a set of specific quantitative data that can accurately reflect the interference caused by illumination on the joint feature expression. Each data corresponds to the degree of interference of an illumination-sensitive feature.

[0122] The intrinsic material components are a set of features that reflect the inherent properties of the bamboo-plastic packaging box itself. The light interference features are a set of specific quantitative data that reflect the interference caused by light on the joint feature expression. Based on the intrinsic material components, for each intrinsic material feature, the corresponding interference quantitative data in the light interference features is found. If the interference quantitative data shows that the intrinsic material feature is greatly affected by light interference, that is, the difference is greater than the preset interference level judgment value, which is obtained through a large number of light interference experiments on bamboo-plastic packaging boxes, the intrinsic material feature is corrected according to the reverse value of the interference quantitative data. If light causes the feature value to increase, the value is adjusted to decrease in the opposite direction according to the increase of the interference quantitative value, and vice versa. If the interference quantitative data shows that the intrinsic material feature is affected by light interference to zero, that is, the difference is equal to zero, the original value of the intrinsic material feature remains unchanged. All the corrected features and the uncorrected features are integrated together to form the initial light-free feature expression, which is a preliminary feature set that excludes the influence of light interference and only reflects the inherent properties of the bamboo-plastic packaging box itself.

[0123] The initial no-light feature representation is a preliminary feature set that excludes the influence of light interference and only reflects the inherent properties of the bamboo-plastic packaging box itself. First, the numerical changes of each feature in the initial no-light feature representation under different scenarios and time points are analyzed. The preset fixed threshold is determined based on a large amount of material feature data of bamboo-plastic packaging boxes. This threshold ensures that the features below can stably reflect the material properties. Features with numerical changes less than the preset fixed threshold are selected as basic features. Then, these basic features are enhanced by highlighting the unique identification information of the features to enhance the recognizability of the features in different environments. At the same time, unstable features with numerical changes greater than the preset fixed threshold in the initial no-light feature representation are removed. The enhanced basic features are integrated to form the light-robust features. This feature set is the final feature set that can always remain stable and accurately reflect the material and structural properties of the bamboo-plastic packaging box itself under various lighting conditions.

[0124] The beneficial effects are: accurately distinguishing between light-sensitive features and intrinsic material features, clarifying the objects to be processed, laying the foundation for obtaining stable features, quantifying light interference based on ambient light field parameters to make the interference measurable, providing accurate data basis, ensuring precise implementation of interference suppression, reverse correction to eliminate light interference, ensuring that the initial features truly reflect the inherent properties of bamboo-plastic packaging boxes, screening and strengthening stable features, improving feature reliability, providing high-quality support for pose assessment, and ensuring accurate tracking and positioning.

[0125] S6. Based on the illumination robustness features, the spatial position and orientation of the bamboo-plastic packaging box are evaluated to obtain the six-degree-of-freedom tracking and positioning data of the bamboo-plastic packaging box.

[0126] In this embodiment of the invention, the step of evaluating the spatial position and orientation of the bamboo-plastic packaging box based on the illumination robustness features to obtain the six-degree-of-freedom tracking and positioning data of the bamboo-plastic packaging box includes:

[0127] The key geometric features of the bamboo-plastic packaging box are obtained by performing structure-aware extraction on the illumination robust features.

[0128] By deriving the constraint relationships of the key geometric features, the spatial constraints of the bamboo-plastic packaging box are obtained.

[0129] The spatial constraints are subjected to temporal smoothing to obtain the optimized pose sequence of the bamboo-plastic packaging box;

[0130] The optimized pose sequence is serialized and encapsulated into six-degree-of-freedom tracking and positioning data for the bamboo-plastic packaging box.

[0131] By deriving the constraint relationships of the key geometric features, the spatial constraints of the bamboo-plastic packaging box are obtained, including:

[0132] Geometric correlation calibration is performed on the key geometric features to obtain feature matching pairs for the bamboo-plastic packaging box;

[0133] The geometric parameters of the feature matching pairs are calculated to obtain the preliminary constraint parameters of the bamboo-plastic packaging box.

[0134] The spatial constraints of the bamboo-plastic packaging box are obtained by effectively consolidating the preliminary constraint parameters.

[0135] The calculation formula for the preliminary constraint parameters is as follows:

[0136] ;

[0137] In the formula, These are the initial constraint parameters. For the first The unit normal vector of each of the key geometric features, For the first feature matching pair Coordinates of a point in three-dimensional space. The total number of the feature matching pairs. For Euclidean norm operators The preset numerical stability constant is... The coordinates of the three-dimensional space point The weighted centroid.

[0138] Illumination robust features are a set of features that remain stable and accurately reflect the material and structural properties of bamboo-plastic packaging boxes under various lighting conditions. Structural perception extraction first identifies the feature parts directly related to the spatial structure of bamboo-plastic packaging boxes from the illumination robust features. These parts cover features that reflect the geometric shapes of the packaging box, such as the outline, corners, and surface protrusions or depressions. Then, the importance of the features in determining the spatial position and orientation is selected based on whether the feature can clearly identify the unique spatial structural points of the packaging box, such as the corner features that can distinguish different sides of the packaging box. Finally, the extracted key geometric features are a set of features that accurately represent the core spatial structure of bamboo-plastic packaging boxes. Each feature corresponds to a specific geometric position or shape on the packaging box.

[0139] Key geometric features are a set of features that represent the core spatial structure of bamboo-plastic packaging boxes. Geometric correlation calibration first analyzes the actual geometric position of the packaging box corresponding to each key geometric feature, and then determines the spatial correspondence between different key geometric features, such as the connection relationship between a corner feature and the contour feature of an adjacent face. Subsequently, two key geometric features that have a direct spatial relationship are combined together to form a pair of features. Each pair of combinations can reflect the relationship between two geometric positions. These paired feature combinations are feature matching relationship pairs.

[0140] A feature matching pair is a combination that reflects the spatial relationship between two key geometric features. The geometric parameter calculation is based on the actual physical dimensions of the bamboo-plastic packaging box. Based on the actual geometric positions of the two features in each feature matching pair, the specific values ​​of the distance, angle and other constraints between the two positions are calculated to ensure that the calculated values ​​can accurately reflect the real spatial constraints. Each feature matching pair corresponds to a calculated constraint value, and these constraint values ​​are the preliminary constraint parameters.

[0141] The initial constraint parameters are the constraint values ​​corresponding to each feature matching relationship. The effective constraint collection first sets a reasonable range of values ​​based on the standard size and normal spatial structure of the bamboo-plastic packaging box. The initial constraint parameters that exceed this range are judged as invalid and removed. Then, the remaining initial constraint parameters within the effective range are integrated. The integrated parameter set can fully reflect the spatial constraint relationship between the key geometric features of the bamboo-plastic packaging box. This integrated parameter set is the spatial constraint.

[0142] Spatial constraints are a set of parameters reflecting the spatial limitations of key geometric features of bamboo-plastic packaging boxes. Temporal smoothing first analyzes the spatial constraints corresponding to each time point according to the shooting time sequence of the original image sequence. Then, based on the frame rate of the shooting device and the speed of normal movement or posture change of the bamboo-plastic packaging box, a fixed change threshold is set. The change amplitude between spatial constraints of adjacent time points is checked. If the change amplitude exceeds the threshold, the spatial constraints of the next time point are adjusted to control the change amplitude within the threshold range, ensuring that the change of spatial constraints is continuous and stable. The pose information determined by the spatial constraints corresponding to each time point after adjustment, arranged in chronological order, forms the optimized pose sequence.

[0143] The optimized pose sequence is a stable pose information sequence arranged in chronological order. Serialization encapsulation first organizes the pose information of each time point in the optimized pose sequence according to a unified format. The pose information of each time point clearly includes three dimensions of information representing spatial position and three dimensions of information representing attitude. Then, the organized pose information is combined into a complete data set in chronological order. This data set is the six-degree-of-freedom tracking and positioning data, which can continuously and accurately reflect the spatial position and attitude of bamboo plastic packaging boxes at different time points.

[0144] In the formula, These are the preliminary constraint parameters to be determined, used to quantize the first... The degree to which key geometric features constrain the spatial orientation of bamboo-plastic packaging boxes provides a foundation for subsequent integration of spatial constraints. It is a unit normal vector further obtained from the extracted k-th key geometric feature, used to characterize the spatial orientation of the geometric feature, and serves as the directional reference for constraint calculation. The coordinates of the i-th three-dimensional point in the feature matching pair obtained through geometric correlation calibration are the core location data that carries geometric constraint information. The total number of feature matching pairs limits the range of the summation operation in the formula, ensuring that the constraint contributions of all matching points are included. Based on all The weighted centroid of the calculation serves as a unified geometric benchmark, reducing computational bias caused by discrete points. yes arrive The Euclidean norm, or straight-line distance in three-dimensional space, is used to quantify the positional offset of a matching point relative to a reference point. It is a numerical stability constant preset based on the resolution of the original image sequence and the feature matching error threshold, used to avoid and To prevent computational overflow caused by the denominator approaching zero when the result is too close, this method ensures the reliability of the result.

[0145] The beneficial effects are as follows: redundant information is eliminated, key geometric features are accurate, providing effective basic data for subsequent constraint relationship derivation, ensuring the accuracy of subsequent steps; the spatial correspondence of key geometric features is clarified, forming feature matching relationship pairs, providing clear computational objects, improving the efficiency of geometric parameter solution; constraint values ​​are calculated based on actual geometric positions and physical dimensions, the initial constraint parameters are accurate, providing a reliable foundation for effective constraint aggregation; invalid parameters are eliminated within a reasonable range, the integrated spatial constraints are comprehensive and accurate, avoiding interference with subsequent pose evaluation; abnormal spatial constraints are adjusted by setting change thresholds, optimizing the continuous and stable pose sequence, eliminating abrupt changes in pose information, improving reliability; pose information is organized in a unified format, the six-degree-of-freedom tracking and positioning data is complete, arranged in chronological order, meeting the needs of actual tracking applications.

[0146] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0147] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0148] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for tracking and locating bamboo-plastic packaging boxes based on AI vision, characterized in that, The method includes: S1, performing scene light field analysis on the original image sequence of the bamboo-plastic packaging box to obtain the ambient light field parameters of the bamboo-plastic packaging box; S2, based on the material reflection characteristics of the bamboo-plastic packaging box, performing optical property extraction on the bamboo-plastic packaging box to obtain the reflection characteristic parameters of the bamboo-plastic packaging box, including: performing multi-view collaborative analysis on the original image sequence to obtain the reflection intensity distribution on the surface of the bamboo-plastic packaging box; and performing spectral dimension decomposition on the reflection intensity distribution based on the multispectral data in the original image sequence to obtain the high-dimensional spectral reflectance of the bamboo-plastic packaging box. The high-dimensional spectral reflectance profile is matched with a preset material optical property library to obtain the material optical properties of the bamboo-plastic packaging box. Based on the material optical properties, the high-dimensional spectral reflectance profile is optically deconstructed to obtain the optical parameters of the bamboo-plastic packaging box. The optical parameters are normalized to obtain the reflectance characteristic parameters of the bamboo-plastic packaging box, including: decoupling the spectral components of the high-dimensional spectral reflectance profile based on the material optical properties to obtain the diffuse reflection component and specular reflection component of the bamboo-plastic packaging box; and then, the diffuse reflection component and specular reflection component are normalized. Attribute labeling is performed to obtain the reflectance coefficient pairs of the bamboo-plastic packaging box; based on the geometric distribution of the high-dimensional spectral reflectance profile, the reflectance coefficient pairs are weighted and corrected to obtain the optical parameters of the bamboo-plastic packaging box; S3, based on the ambient light field parameters, optical response simulation is performed on the reflectance characteristic parameters to obtain visual samples of the bamboo-plastic packaging box; S4, cross-domain feature alignment is performed between the original image sequence and the visual samples to obtain the joint feature representation of the bamboo-plastic packaging box; S5, illumination interference decoupling is performed on the joint feature representation to obtain the illumination robust features of the bamboo-plastic packaging box. S6. Based on the illumination robust features, perform pose state evaluation on the spatial position and orientation of the bamboo-plastic packaging box to obtain six-degree-of-freedom tracking and positioning data of the bamboo-plastic packaging box, including: extracting the key geometric features of the bamboo-plastic packaging box by structural perception from the illumination robust features; deriving the constraint relationship of the key geometric features to obtain the spatial constraints of the bamboo-plastic packaging box; performing temporal smoothing processing on the spatial constraints to obtain the optimized pose sequence of the bamboo-plastic packaging box; and serializing and encapsulating the optimized pose sequence into six-degree-of-freedom tracking and positioning data of the bamboo-plastic packaging box.

2. The identification method for tracking and locating bamboo-plastic packaging boxes based on AI vision as described in claim 1, characterized in that, The process involves performing scene light field analysis on the original image sequence of the bamboo-plastic packaging box to obtain its ambient light field parameters. This includes: performing temporal extraction on the original image sequence of the bamboo-plastic packaging box to obtain a stable image sequence; equalizing the imaging brightness in the stable image sequence to obtain a brightness characterization of the bamboo-plastic packaging box; performing multi-scale light field feature analysis on the brightness characterization to obtain a multi-level descriptor of the light field features of the bamboo-plastic packaging box; performing scene fusion on the multi-level descriptor of the light field features and the stable image sequence to obtain a scene light field description of the bamboo-plastic packaging box; and performing modal decomposition on the scene light field description to obtain the ambient light field parameters of the bamboo-plastic packaging box.

3. The identification method for tracking and locating bamboo-plastic packaging boxes based on AI vision as described in claim 1, characterized in that, The step of performing optical response simulation on the reflection characteristic parameters based on the ambient light field parameters to obtain a visual sample of the bamboo-plastic packaging box includes: extracting illumination elements from the ambient light field parameters to obtain the light field illumination characteristics of the bamboo-plastic packaging box; performing reverse ray simulation tracing on the light field illumination characteristics according to the diffuse reflection coefficient and specular reflection coefficient in the reflection characteristic parameters to obtain the pixel-level irradiance distribution of the bamboo-plastic packaging box; performing perspective projection mapping on the pixel-level irradiance distribution according to the camera viewpoint parameters corresponding to the original image sequence, and rendering the mapping result as a geometric projection image of the bamboo-plastic packaging box; and performing color gamut adaptation optimization on the geometric projection image to obtain a visual sample of the bamboo-plastic packaging box.

4. The identification method for tracking and locating bamboo-plastic packaging boxes based on AI vision as described in claim 1, characterized in that, The method of performing cross-domain feature alignment between the original image sequence and the visual samples to obtain the joint feature representation of the bamboo-plastic packaging box includes: performing representation convergence on the original image sequence to obtain the original features of the bamboo-plastic packaging box; performing geometric topological representation on the visual samples to obtain the simulated features of the bamboo-plastic packaging box; and performing cross-domain feature alignment between the original features and the simulated features to obtain the joint feature representation of the bamboo-plastic packaging box.

5. The identification method for tracking and positioning bamboo-plastic packaging boxes based on AI vision as described in claim 3, characterized in that, The illumination-robust features of the bamboo-plastic packaging box are obtained by decoupling the joint feature expression from illumination interference, including: performing illumination-sensitive decomposition on the joint feature expression to obtain the illumination-sensitive component and the material intrinsic component of the bamboo-plastic packaging box; performing explicit interference quantization on the illumination-sensitive component based on the ambient light field parameters to obtain the illumination interference features of the bamboo-plastic packaging box; performing interference suppression fusion on the material intrinsic component and the illumination interference features to obtain the initial no-illumination feature expression of the bamboo-plastic packaging box; and performing invariant optimization on the initial no-illumination feature expression to obtain the illumination-robust features of the bamboo-plastic packaging box.

6. The identification method for tracking and locating bamboo-plastic packaging boxes based on AI vision as described in claim 1, characterized in that, The process of evaluating the spatial position and orientation of the bamboo-plastic packaging box based on the illumination robust features to obtain six-degree-of-freedom tracking and positioning data of the bamboo-plastic packaging box includes: extracting the key geometric features of the bamboo-plastic packaging box through structure-aware extraction of the illumination robust features; deriving the constraint relationships of the key geometric features to obtain the spatial constraints of the bamboo-plastic packaging box; performing temporal smoothing processing on the spatial constraints to obtain the optimized pose sequence of the bamboo-plastic packaging box; and serializing and encapsulating the optimized pose sequence into six-degree-of-freedom tracking and positioning data of the bamboo-plastic packaging box.

7. The identification method for tracking and locating bamboo-plastic packaging boxes based on AI vision as described in claim 1, characterized in that, The spatial constraints of the bamboo-plastic packaging box are derived by performing constraint relationship derivation on the key geometric features, including: performing geometric association calibration on the key geometric features to obtain feature matching relationship pairs of the bamboo-plastic packaging box; performing geometric parameter calculation on the feature matching relationship pairs to obtain preliminary constraint parameters of the bamboo-plastic packaging box; and performing effective constraint aggregation on the preliminary constraint parameters to obtain the spatial constraints of the bamboo-plastic packaging box.

8. The identification method for tracking and locating bamboo-plastic packaging boxes based on AI vision as described in claim 7, characterized in that, The calculation formula for the preliminary constraint parameters is as follows: In the formula, These are the initial constraint parameters. For the first The unit normal vector of each of the key geometric features, For the first feature matching pair Coordinates of a point in three-dimensional space. The total number of the feature matching pairs. For Euclidean norm operators The preset numerical stability constant is... The coordinates of the three-dimensional space point The weighted centroid.

Images