An AR glasses hidden camera system and a computational imaging processing method

Abstract

The application discloses an AR glasses hidden camera system and a computing imaging processing method, which can be applied to the technical fields of intelligent wearable devices, optical imaging and computational photography. The system is integrated in the frame structure, including: a decorative functional micropore array arranged on the outer surface of the frame, which is composed of a plurality of light transmission micropores and non-light transmission decorative points arranged according to a preset geometric texture, so as to realize visual traceless hiding; an image sensor arranged in a light-shielding accommodation space and on the back side of the array, which receives micropore transmitted light and generates a plurality of sub-aperture images; a light-shielding isolation structure for isolating the light paths of adjacent micropores; and an image processing unit with an AI perception enhanced photography processing pipeline, which performs geometric alignment, light field fusion, image quality enhancement and semantic optimization on the plurality of sub-aperture images and outputs a target imaging result. In this way, on the basis of realizing traceless hiding of the camera, crosstalk of the light paths is eliminated, the imaging clarity and scene optimization capability are improved, and the requirements of AR glasses for photographing and space perception are met.

Description

Technical Field

[0001] This application relates to the fields of smart wearable devices, optical imaging and computational photography, and in particular to an AR glasses hidden camera system and computational imaging processing method. Background Technology

[0002] Augmented Reality (AR) glasses are lightweight smart wearable devices that enable functions such as augmented reality, spatial awareness, and everyday image capture. With the rapid popularization of consumer-grade AR products, users' demands for integrated frame design, camera concealment, image quality, and intelligent shooting capabilities continue to rise.

[0003] Existing AR glasses with hidden cameras generally employ mechanical shielding or optical coating designs. However, visible structural traces or coating color differences easily remain on the device surface, resulting in poor concealment and raising concerns about intrusion and privacy. In terms of imaging, traditional single-aperture solutions limit light intake, making diffraction blur difficult to correct, leading to insufficient light field information and poor image quality. If multiple micro-apertures are used for light collection without independent isolation and geometric calibration of the light paths of each micro-aperture, crosstalk between adjacent apertures may occur, producing stray light and ghosting. Furthermore, these devices only possess basic shooting capabilities, lacking scene recognition, adaptive exposure, and intelligent composition optimization functions, resulting in poor imaging stability in complex scenes.

[0004] Therefore, how to eliminate optical crosstalk and improve image clarity and scene optimization capabilities while achieving seamless camera concealment is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] To address the aforementioned issues, this application provides an AR glasses hidden camera system and computational imaging processing method. While achieving seamless camera concealment, it eliminates optical crosstalk, improves image clarity and scene optimization capabilities, and adapts to the photography and spatial perception needs of AR glasses.

[0006] In a first aspect, embodiments of this application provide a hidden camera system for AR glasses, the system being integrated into the frame structure of AR glasses, including: a decorative functional micro-hole array, an image sensor, a light-shielding and isolation structure, and an image processing unit; The decorative functional micropore array is disposed on the outer surface of the frame structure, presenting an overall geometric texture pattern with a decorative effect. The decorative functional micropore array includes multiple light-transmitting micropores and multiple opaque decorative points. The light-transmitting micropores serve as sampling sub-apertures for the light field acquisition of the compound eye, and are mixed and arranged with the opaque decorative points according to a preset geometric texture pattern. The light-transmitting micropores and the opaque decorative points have the same appearance size, color, and / or surface treatment effect, so that an external observer cannot distinguish between the light-transmitting micropores and the opaque decorative points under preset observation conditions. The image sensor is disposed on the back side of the decorative functional micro-hole array and located in the light-shielding and receiving space formed inside the frame structure. It is used to receive ambient light transmitted through the light-transmitting micro-holes and generate a multi-path sub-aperture image corresponding to the sampling sub-aperture. The light-shielding and isolation structure is disposed between the decorative functional micro-hole array and the image sensor to isolate the light path of adjacent light-transmitting micro-holes; The image processing unit is connected to the image sensor signal. The image processing unit has a built-in artificial intelligence (AI) perception-enhanced photography processing pipeline, which is used to perform geometric alignment, light field fusion, image quality enhancement and semantic optimization processing on the multi-aperture images, and output the target imaging results.

[0007] Optionally, the geometric texture pattern is any one or more combinations of dot matrix patterns, hexagonal honeycomb patterns, and laser-etched grid patterns.

[0008] Optionally, the decorative functional micro-perforation array is disposed at any one or more of the head region, the front frame side panel, and the inner surface of the proximal end of the temple of the AR glasses.

[0009] Optionally, the diameter of the light-transmitting micro-holes is 50-300μm, and the number is 16-256; the effective spatial baseline between the light-transmitting micro-holes participating in light field acquisition within the decorative functional micro-hole array is 3-20mm.

[0010] Optionally, the image sensor has a pixel size of no more than 1.8 μm and a readout noise of less than 2.5 eΩ. - .

[0011] Optionally, the system further includes: a plano-convex microlens; The plano-convex microlens is disposed behind at least part of the light-transmitting micro-aperture, and the plano-convex microlens is disposed 1 mm behind any light-transmitting micro-aperture having the plano-convex microlens, and the focal length of the plano-convex microlens is 1-2 mm.

[0012] Optionally, the light-shielding and isolation structure includes at least one of a black light-shielding resin partition wall, a light-shielding microchannel array, a metal light-shielding partition plate, and an integrally molded light-shielding wall, wherein the light-shielding and isolation structure respectively defines the independent optical path corresponding to each light-transmitting micro-hole.

[0013] Optionally, the image processing unit stores or retrieves factory calibration data, which includes at least one of the following: a position mapping table of the light-transmitting micro-apertures on the photosensitive surface of the image sensor, baseline parameters between apertures, and point spread function parameters corresponding to each light-transmitting micro-aperture.

[0014] Secondly, embodiments of this application provide a computational imaging processing method for a hidden camera in AR glasses, including: The image acquisition frame acquired by the image sensor includes N sub-aperture imaging regions corresponding to N light-transmitting micro-apertures in the decorative functional micro-aperture array; where N is an integer greater than 1. Based on the pre-calibrated microaperture position mapping data, dark field correction, flat field correction and sub-aperture segmentation are performed on the image acquisition frame to obtain N independent sub-aperture image blocks, and the quality weight vector corresponding to each sub-aperture image block is determined. Based on geometric calibration data, disparity estimation and sub-pixel-level geometric alignment are performed on N-path sub-aperture image blocks to construct a light field tensor; combined with the quality weight vector, the light field tensor is weighted and fused using a multi-view fusion algorithm to obtain an initial fused image. Based on the pre-calibrated point spread function, the initial fused image is subjected to diffraction blur correction to obtain a deblurred fused image; The deblurred fused image is sequentially subjected to adaptive noise reduction, super-resolution reconstruction, and moiré pattern removal to obtain the target fused image; Identify the scene category corresponding to the target fused image; Based on the scene category, the corresponding virtual exposure strategy and tone enhancement parameters are matched, and the optimized target imaging result is output, while intelligent composition suggestions are generated simultaneously.

[0015] Optionally, the method further includes: Based on the inter-frame consistency constraint of optical flow, the inter-frame flicker of the target imaging result is suppressed; Based on the output application scenario, the target imaging result is converted into a target format that matches the output application scenario.

[0016] Optionally, before performing diffraction blur correction on the initial fused image based on a pre-calibrated point spread function to obtain a deblurred fused image, the method further includes: A depth map is generated based on the disparity information in the light field tensor.

[0017] Optionally, the diffraction blur correction of the initial fused image includes: The initial fused image is pre-filtered in the frequency domain using Wiener deconvolution, and / or deblurred using a deep deconvolution network.

[0018] Optionally, the step of matching the virtual exposure strategy and tone enhancement parameters corresponding to the scene category, outputting the optimized target imaging result, and simultaneously generating intelligent composition suggestions includes: Based on the scene category, dynamically adjust the equivalent international standardization (ISO) sensitivity parameters, gain curve, and high dynamic range (HDR) tone mapping parameters of the target fused image; Load the color lookup table corresponding to the scene category, perform targeted tone optimization on the target fusion image, and obtain the optimized target imaging result; The main region of the target fused image is identified by a saliency detection network, the target cropping box that conforms to the preset composition rules is calculated, and intelligent composition suggestions are generated by AR display overlay or automatic cropping. Output the target imaging results and the intelligent mapping suggestions.

[0019] As can be seen from the above technical solutions, compared with the prior art, this application has the following advantages: This application provides a hidden camera system for AR glasses, integrated into the frame structure of AR glasses, including: a decorative functional micro-hole array, an image sensor, a light-shielding and isolation structure, and an image processing unit. The decorative functional micro-hole array is disposed on the outer surface of the frame structure, presenting an overall geometric texture pattern with a decorative effect. The decorative functional micro-hole array includes multiple light-transmitting micro-holes and multiple opaque decorative points. The multiple light-transmitting micro-holes serve as multiple sampling sub-apertures for collecting the light field of the compound eye, and are mixed and arranged with the opaque decorative points according to a preset geometric texture pattern. The appearance, size, color, and / or surface treatment of the light-transmitting micro-holes and the opaque decorative points are consistent, and an external observer cannot distinguish between the light-transmitting micro-holes and the opaque decorative points under preset observation conditions. The image sensor is disposed within the light-shielding and receiving space formed by the inward enclosure of the frame structure, and is located on the back side of the decorative functional micro-hole array. It is used to receive ambient light transmitted through the light-transmitting micro-holes and generate multi-path sub-aperture images corresponding to the sampling sub-apertures. A light-shielding isolation structure is positioned between the decorative functional micro-aperture array and the image sensor to isolate the optical paths of adjacent light-transmitting micro-apertures. The image processing unit is signal-connected to the image sensor and incorporates an AI-powered perception-enhanced photographic processing pipeline. This pipeline performs geometric alignment, light field fusion, image quality enhancement, and semantic optimization on multi-aperture images, outputting the target imaging result. Thus, a concealed deployment is achieved through a decorative functional micro-aperture array composed of uniformly shaped light-transmitting micro-apertures and opaque decorative points. Combined with the light-shielding isolation structure blocking the optical paths of adjacent micro-apertures and the AI-powered perception-enhanced photographic processing pipeline's fusion and optimization of sub-aperture images, this achieves seamless camera concealment, eliminates optical crosstalk, improves image clarity and scene optimization capabilities, and adapts to the photography and spatial perception needs of AR glasses. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the structure of a hidden camera system for AR glasses provided in an embodiment of this application; Figure 2 This is a schematic diagram showing the position of a decorative functional microporous array provided in an embodiment of this application; Figure 3 A schematic diagram of a decorative functional microporous array provided in an embodiment of this application; Figure 4 A flowchart illustrating a computational imaging processing method for a hidden camera in AR glasses, provided as an embodiment of this application. Detailed Implementation

[0021] As mentioned earlier, existing AR glasses struggle to eliminate optical crosstalk and improve image clarity and scene optimization capabilities while seamlessly concealing the camera. Specifically, existing AR glasses typically employ mechanical shielding or optical coating designs for their hidden cameras, leaving visible structural traces or coating color differences on the device surface. This results in poor concealment and raises concerns about intrusiveness and privacy. In terms of imaging, traditional single-aperture solutions limit light intake, making diffraction blur difficult to correct, leading to insufficient light field information and poor image quality. If multiple micro-apertures are used for light collection without independent isolation and geometric calibration of the light paths of each micro-aperture, crosstalk between adjacent apertures can occur, producing stray light and ghosting. Furthermore, these devices only possess basic shooting capabilities, lacking scene recognition, adaptive exposure, and intelligent composition optimization functions, resulting in poor imaging stability in complex scenes.

[0022] To address the aforementioned issues, this application provides a hidden camera system for AR glasses. This system is integrated into the frame structure of AR glasses and includes: a decorative functional micro-aperture array, an image sensor, a light-shielding and isolation structure, and an image processing unit. The decorative functional micro-aperture array is disposed on the outer surface of the frame structure, presenting an overall geometric texture pattern with a decorative effect. The array includes multiple light-transmitting micro-apertures and multiple opaque decorative points. The light-transmitting micro-apertures serve as multiple sampling sub-apertures for collecting the light field of the compound eye, and are mixed and arranged with the opaque decorative points according to a preset geometric texture pattern. The light-transmitting micro-apertures and the opaque decorative points have the same appearance, size, color, and / or surface treatment effect, making it impossible for an external observer to distinguish between them under preset observation conditions. The image sensor is disposed within the light-shielding and receiving space formed by the inward enclosure of the frame structure and is located on the back side of the decorative functional micro-aperture array. It is used to receive ambient light transmitted through the light-transmitting micro-apertures and generate multi-path sub-aperture images corresponding to the sampling sub-apertures. A light-shielding isolation structure is positioned between the decorative functional microaperture array and the image sensor to isolate the optical paths of adjacent light-transmitting microapertures. The image processing unit is signal-connected to the image sensor and incorporates an AI-powered perception-enhanced photographic processing pipeline for geometric alignment, light field fusion, image quality enhancement, and semantic optimization of multi-aperture images, ultimately outputting the target imaging result.

[0023] In this way, a decorative functional micro-hole array consisting of uniformly shaped light-transmitting micro-holes and opaque decorative dots is used to achieve concealed deployment. Combined with a light-shielding and isolation structure to block the light path of adjacent micro-holes, and combined with an AI perception-enhanced photography processing pipeline to perform fusion and optimization processing on sub-aperture images, the camera is seamlessly hidden, light path crosstalk is eliminated, and the image clarity and scene optimization capabilities are improved, making it suitable for AR glasses photography and spatial perception needs.

[0024] It should be noted that the AR glasses hidden camera system and computational imaging processing method provided in this application embodiment can be applied to the fields of smart wearable devices, optical imaging, and computational photography. The above are merely examples and do not limit the application field of the AR glasses hidden camera system and computational imaging processing method provided in this application.

[0025] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0026] Figure 1 This is a schematic diagram of the structure of a hidden camera system for AR glasses provided in an embodiment of this application. (Combined with...) Figure 1 As shown in the embodiment of this application, an AR glasses hidden camera system 100 is integrated into the frame structure of AR glasses and includes: a decorative functional micro-hole array 110, an image sensor 120, a light-shielding and isolation structure 130, and an image processing unit 140. The decorative functional micropore array 110 is disposed on the outer surface of the frame structure, presenting an overall geometric texture pattern with decorative effect; the decorative functional micropore array 110 includes multiple light-transmitting micropores and multiple opaque decorative points, the light-transmitting micropores serve as sampling sub-apertures for the light field acquisition of the compound eye, and are mixed and arranged with the opaque decorative points according to a preset geometric texture pattern; the light-transmitting micropores and the opaque decorative points have the same appearance size, color and / or surface treatment effect, so that an external observer cannot distinguish between the light-transmitting micropores and the opaque decorative points under preset observation conditions; The image sensor 120 is disposed on the back side of the decorative functional micro-hole array 110 and located in the light-shielding and receiving space formed inside the frame structure. It is used to receive ambient light transmitted through the light-transmitting micro-holes and generate a multi-path sub-aperture image corresponding to the sampling sub-aperture. The light-shielding and isolation structure 130 is disposed between the decorative functional micro-hole array 110 and the image sensor 120 to isolate the light path of adjacent light-transmitting micro-holes; The image processing unit 140 is signal-connected to the image sensor 120. The image processing unit 140 has a built-in AI perception-enhanced photography processing pipeline, which is used to perform geometric alignment, light field fusion, image quality enhancement and semantic optimization processing on the multi-path sub-aperture image, and output the target imaging result.

[0027] Specifically, this application provides an AR glasses hidden camera system 100, which is integrated into the frame structure of AR glasses. It is composed of a decorative-functional micro-aperture array 110, an image sensor 120, a light-shielding and isolation structure 130, and an image processing unit 140. The decorative-functional micro-aperture array (DFMA) 110 is disposed on the outer surface of the frame structure and includes several light-transmitting micro-apertures and opaque decorative points. Multiple light-transmitting micro-apertures and opaque decorative points are arranged in a mixed pattern according to a preset geometric texture, presenting an overall decorative geometric texture pattern with modern industrial aesthetics. It is understood that the number of light-transmitting micro-apertures is greater than one, serving as the sampling sub-aperture for the compound eye's light field acquisition, thus satisfying the compound eye's imaging requirements. The light-transmitting micro-apertures and opaque decorative points maintain consistency in appearance, size, color, and / or surface treatment. Under preset observation conditions, external observers cannot distinguish between the two, perceiving only the decorative texture of the glasses, thereby achieving the hidden camera design. For example, during processing, all points on the entire DFMA substrate can be simultaneously machined using femtosecond laser or precision stamping. The outer diameter and recess size of the light-transmitting micropores and opaque decorative points are completely consistent. After uniform degreasing, blasting, and sandblasting pretreatment, the entire area undergoes 8-12μm black hard anodizing, followed by spraying with a matte black polyurethane (PU) protective coating of the same batch. Throughout the process, there is no regional differentiation in coloring and polishing. The metal substrate can be integrally molded, while the plastic substrate is injection molded using the same color masterbatch in two colors, with only the micropores being opened. Relying on uniform molding and synchronous surface coating processes, color differences and reflection differences between points are eliminated, achieving visual indistinguishability between the two under preset observation conditions. Among them, the preset observation conditions can be "daily social observation conditions with an observation distance ≥30cm, ambient illuminance of 100lux-10000lux, and an angle between the line of sight and the normal of the frame surface of 0°-60°". The preset geometric texture follows a pre-defined point arrangement rule, uniformly specifying the size, spacing, and array arrangement of the light-transmitting micro-holes and opaque decorative points. Furthermore, the substrate material of the decorative functional micro-hole array 110 is typically a thin black anodized aluminum alloy plate (thickness ≤ 0.4 mm) with an opacity ≥ 99.9%, integrally formed with or precisely attached to the frame. The image sensor 120 is located within the light-shielding space formed by the inward enclosure of the frame structure, and is situated on the back side parallel to the decorative functional micro-hole array 110. It receives ambient light transmitted through each light-transmitting micro-hole and generates sub-aperture images corresponding to each micro-hole. The output of the image sensor 120 is in the form of raw (RAW) frames. Each RAW frame contains multiple sub-aperture images corresponding to all light-transmitting micro-holes, i.e., the raw data of all sub-aperture regions superimposed.The light-shielding isolation structure 130 can be a black resin tube array, positioned between the decorative functional micro-aperture array 110 and the image sensor 120, and corresponding to the optical path of each light-transmitting micro-aperture. This isolates the independent optical paths between adjacent light-transmitting micro-apertures, preventing crosstalk between sub-aperture images. The image processing unit 140 is signal-connected to the image sensor 120 and incorporates an AI-Perception-Enhanced Photography Pipeline (AIPEPP) for aligning and fusing multi-path sub-aperture images, enhancing image quality, and optimizing semantics, ultimately outputting the target imaging result.

[0028] As one implementation, regarding the location of the decorative functional microperforation array 110, the decorative functional microperforation array 110 is disposed at any one or more of the head region, the front frame side panel, and the inner surface of the proximal end of the temple of the AR glasses.

[0029] Figure 2 This is a schematic diagram showing the location of a decorative functional microporous array provided in an embodiment of this application. (Combined with...) Figure 2 As shown, the decorative functional microperforation array 110 can be positioned in, but is not limited to, the head region 111 of the AR glasses, the front frame side panel 112, and the inner proximal side of the temple 113. The decorative functional microperforation array 110 can be positioned in any of the above-mentioned locations, or it can be positioned in multiple locations simultaneously. For example, it can be positioned only in the head region 111, or simultaneously in the head region 111 and the front frame side panel 112. It is understood that this application only lists the above three positions as examples, and the actual possible positions are not limited to these. Furthermore, regarding the head region 111, its location is the structural component area where the AR glasses' front frame connects to the temple, a key decorative area in the glasses' appearance design, with a structural thickness ≥3mm, making it the optimal integration location for the decorative functional microperforation array 110. Figure 2 Only the case where the decorative functional microperforation array 110 is set in the pile head region 111 is shown; for the front frame side panel 112, its position is on both sides of the front of the frame, which provides the best light field coverage and is the second preferred integration position of the decorative functional microperforation array 110; for the inner side of the temple near the end 113, its position is the inner side of the temple near the end of the AR glasses, which can supplement the lateral angle information and is the third preferred integration position of the decorative functional microperforation array 110.

[0030] As one implementation, regarding how to design the geometric texture pattern of the decorative functional microporous array 110, the geometric texture pattern is any one or more combinations of dot matrix patterns, hexagonal honeycomb patterns, and laser-etched grid patterns.

[0031] Figure 3This is a schematic diagram illustrating a decorative functional microporous array provided in an embodiment of this application. (Combined with...) Figure 3 As shown, the geometric texture pattern of the decorative functional microporous array 110 is decorative, including but not limited to dot matrix patterns (such as...). Figure 3 a) Hexagonal honeycomb pattern (such as Figure 3 (b) and laser-etched mesh patterns (such as...) Figure 3 (c) The geometric texture pattern of the decorative functional micro-perforation array 110 on the AR glasses can be any of the above-mentioned patterns, or multiple geometric texture patterns can be combined. For example, a dot matrix pattern can be used alone, or a combination of a dot matrix pattern and a hexagonal honeycomb pattern can be used. It is understood that this application only lists the above three geometric texture patterns as examples, and the actual geometric texture patterns that can be used are not limited to these. Furthermore, for the dot matrix pattern, its appearance is a regularly arranged array of dots, which blends with the industrial / tech-savvy eyewear design. The light-transmitting micro-holes 114 are several "dots" in the dot matrix, consistent with the appearance of the opaque decorative dots 115. For the honeycomb pattern, its appearance is a hexagonal honeycomb geometric pattern, with a strong modern feel and high unit density. The light-transmitting micro-holes 114 and the opaque decorative dots 115 are both hidden in the center or node positions of the honeycomb. For the laser-etched grid pattern, its appearance is a geometric grid laser-etched on the surface of the metal / alloy frame. The light-transmitting micro-holes 114 are light-transmitting holes at the grid nodes, consistent with the opaque decorative dots 115, and integrated with the etched texture.

[0032] As one implementation method, regarding the design of the light-transmitting micropores in the decorative functional micropore array 110, the diameter of the light-transmitting micropores is 50-300μm and the number is 16-256; the effective spatial baseline between the light-transmitting micropores participating in light field acquisition in the decorative functional micropore array 110 is 3-20mm.

[0033] Specifically, the light-transmitting micro-apertures are formed using femtosecond laser micro-aperture processing or precision stamping, with a diameter of 50-300 μm and a roundness error of less than 5 μm. Preferably, the diameter can be 80-250 μm. This size range avoids excessive light intake and local overexposure caused by excessively large apertures, while preventing insufficient light transmission and reduced signal-to-noise ratio caused by excessively small apertures. Simultaneously, the strict roundness error avoids image blurring caused by micro-aperture edge distortion. The effective number of light-transmitting micro-apertures is set according to the frame area and the required light field angle sampling density, typically 16-256. A sufficient number of micro-apertures enriches multi-angle sampling information, providing a data foundation for subsequent multi-image fusion and image quality improvement. In addition, in order to ensure the effectiveness of parallax information and support depth estimation and super-resolution reconstruction, this application sets the effective spatial baseline between the light-transmitting micro-apertures (effective light-transmitting micro-apertures) participating in light field acquisition within the decorative functional micro-aperture array 110 to be 3-20mm. This baseline range takes into account both the depth resolution accuracy of near-distance objects and the parallax detection capability of far-distance targets, ensuring the stability of light field tensor construction and diffraction deblurring processing, and effectively improving the final imaging clarity and 3D scene restoration effect.

[0034] As one implementation method, regarding the design of the image sensor 120, the pixel size of the image sensor 120 is no greater than 1.8 μm, and the readout noise is less than 2.5 eΩ. - .

[0035] Specifically, Complementary Metal-Oxide-Semiconductor (CMOS) is a semiconductor technology widely used in modern electronic devices. The image sensor 120 provided in this application can be a back-side illumination CMOS (BSI-CMOS) image sensor with a pixel size no larger than 1.8 μm. This allows for the arrangement of more photosensitive units within a limited chip area, improving the light field sampling density; the readout noise is less than 2.5 eΩ. - This reduces image noise in low-light scenarios with tiny light-perforated apertures. During assembly, the distance between the light-perforated aperture and the BSI-CMOS image sensor is 5-12mm to match the micro-aperture imaging optical path and avoid image defocus and edge distortion.

[0036] As one implementation method, regarding how to design the AR glasses hidden camera system 100, the AR glasses hidden camera system 100 further includes: a plano-convex microlens; The plano-convex microlens is disposed behind at least part of the light-transmitting micro-aperture, and the plano-convex microlens is disposed 1 mm behind any light-transmitting micro-aperture having the plano-convex microlens, and the focal length of the plano-convex microlens is 1-2 mm.

[0037] Specifically, this system is also equipped with plano-convex microlenses, with at least some of the light-transmitting micro-apertures corresponding to the rear side of the micro-aperture. These plano-convex microlenses are positioned approximately 1 mm behind each light-transmitting micro-aperture equipped with a plano-convex microlens. It can be understood that the distance between the two can be specifically set to 1 mm, and the focal length of the plano-convex microlens can be set to 1-2 mm. Preferably, one plano-convex microlens can be positioned approximately 1 mm behind each light-transmitting micro-aperture. Thus, by adding plano-convex microlenses, traditional pinhole imaging is optimized into a microlens converging imaging mode. Without changing the overall system architecture, this significantly increases the effective light intake of a single aperture, effectively improving the problems of insufficient light intake from small light-transmitting micro-apertures and low signal-to-noise ratio in low-light images. It further improves the brightness and imaging uniformity of each sub-aperture image, providing higher-quality image material for subsequent light field fusion and depth calculation.

[0038] As one implementation method, regarding how to design the light-shielding isolation structure 130, the light-shielding isolation structure 130 includes at least one of a black light-shielding resin partition wall, a light-shielding microchannel array, a metal light-shielding partition plate, and an integrally molded light-shielding wall, wherein the light-shielding isolation structure defines an independent optical path corresponding to each light-transmitting micro-hole.

[0039] Specifically, the light-shielding isolation structure 130 includes at least one of the following: a black light-shielding resin partition wall, a light-shielding microchannel array, a metal light-shielding partition plate, and an integrally molded light-shielding wall. It is arranged between each light-transmitting micro-hole and divides a dedicated propagation space for each light-transmitting micro-hole by light-shielding isolation, thereby limiting the independent optical path corresponding to each light-transmitting micro-hole, blocking the beams of adjacent micro-holes from interfering with each other, eliminating optical interference caused by cross-light and stray light reflection between micro-holes, ensuring that the light rays of each optical path are transmitted independently and do not affect each other, and improving the accuracy of optical imaging or optical detection.

[0040] As one implementation, regarding the design of the image processing unit 140, the image processing unit 140 stores or recalls factory calibration data, which includes at least one of the following: a position mapping table of the light-transmitting micro-apertures on the photosensitive surface of the image sensor, baseline parameters between apertures, and point spread function parameters corresponding to each light-transmitting micro-aperture.

[0041] Specifically, the image processing unit's built-in storage area can pre-save factory calibration data, and this factory calibration data can also be read and called during image correction and other stages. Factory calibration data is a fixed set of standard parameters obtained through optical calibration before the device leaves the factory. As an example, the complete factory optical calibration process is as follows: Using a standard optical calibration stage, the first step involves acquiring images from multiple angles using a checkerboard calibration plate and fitting the microaperture imaging coordinates and baseline distance parameters; the second step involves acquiring diffraction images one aperture at a time using a monochromatic parallel point light source and fitting the independent point spread function (PSF) parameters for each microaperture; the third step involves acquiring images using a globally uniform diffuse light source to obtain dark-field and flat-field correction reference frames. After all calibrations are completed, the parameters are written to the device's storage. This fixed standard parameter set includes at least one of three core parameters: first, a DFMA (Digital-Defined Mapping Table) mapping table recording the position coordinates of all light-transmitting micro-apers on the image sensor's photosensitive surface, used to match the correspondence between micro-apers and imaging pixels; second, baseline distance parameters between different light-transmitting micro-apers (effective spatial baselines between light-transmitting micro-apers participating in light field acquisition within the decorative functional micro-aperture array 110), used for 3D reconstruction and parallax calculation; and third, point spread function parameters unique to each light-transmitting micro-aperture, used to remove imaging blur and restore a clear image. The image processing unit uses the aforementioned factory calibration data to correct, calculate, and restore the original acquired image, eliminating imaging errors caused by micro-aperture optical differences and installation / assembly deviations, thereby improving imaging accuracy and detection reliability.

[0042] In summary, this application provides a hidden camera system for AR glasses, integrated into the frame structure of AR glasses, comprising: a decorative functional micro-perforation array, an image sensor, a light-shielding isolation structure, and an image processing unit. The decorative functional micro-perforation array is disposed on the outer surface of the frame structure, presenting an overall geometric texture pattern with a decorative effect. The array includes multiple light-transmitting micro-holes and multiple opaque decorative points. The light-transmitting micro-holes serve as multiple sampling sub-apertures for collecting the light field of the compound eye, and are mixed and arranged with the opaque decorative points according to a preset geometric texture pattern. The light-transmitting micro-holes and the opaque decorative points have the same appearance size, color, and / or surface treatment effect, making it impossible for an external observer to distinguish between them under preset observation conditions. The image sensor is disposed within the light-shielding and containing space formed by the inward enclosure of the frame structure, and is located on the back side of the decorative functional micro-perforation array. It is used to receive ambient light transmitted through the light-transmitting micro-holes and generate multi-path sub-aperture images corresponding to the sampling sub-apertures. A light-shielding isolation structure is positioned between the decorative functional micro-aperture array and the image sensor to isolate the optical paths of adjacent light-transmitting micro-apertures. The image processing unit is signal-connected to the image sensor and incorporates an AI-powered perception-enhanced photographic processing pipeline. This pipeline performs geometric alignment, light field fusion, image quality enhancement, and semantic optimization on multi-aperture images, outputting the target imaging result. Thus, a concealed deployment is achieved through a decorative functional micro-aperture array composed of uniformly shaped light-transmitting micro-apertures and opaque decorative points. Combined with the light-shielding isolation structure blocking the optical paths of adjacent micro-apertures and the AI-powered perception-enhanced photographic processing pipeline's fusion and optimization of sub-aperture images, this achieves seamless camera concealment, eliminates optical crosstalk, improves image clarity and scene optimization capabilities, and adapts to the photography and spatial perception needs of AR glasses.

[0043] Figure 4 This is a flowchart illustrating a computational imaging processing method for a hidden camera in AR glasses, provided as an embodiment of this application. (Combined with...) Figure 4 As shown in the embodiments of this application, a computational imaging processing method for hidden cameras in AR glasses may include: S401: Acquire an image acquisition frame from an image sensor, the image acquisition frame including N sub-aperture imaging regions corresponding to N light-transmitting micro-apertures in the decorative functional micro-aperture array; where N is an integer greater than 1.

[0044] In practical applications, the imaging process of AR glasses is mainly completed by the image processing unit. Specifically, the N light-transmitting micro-apertures on the decorative functional micro-aperture array serve as independent sampling sub-apertures, enabling N light paths to be transmitted into the image sensor. The image sensor receives the ambient light transmitted through each light-transmitting micro-aperture and generates an image acquisition frame, i.e., a RAW frame. Each image acquisition frame contains multiple sub-aperture imaging regions corresponding to the multiple light-transmitting micro-apertures. These multiple sub-aperture imaging regions are distributed at pre-calibrated positions on the photosensitive surface of the image sensor and are arranged side-by-side to form the raw data. Subsequently, the image sensor sends the image acquisition frame to the image processing unit through a signal connection, and the image processing unit receives it synchronously. That is, the image processing unit acquires the image acquisition frame acquired by the image sensor, which includes the N sub-aperture imaging regions corresponding to the N light-transmitting micro-apertures in the decorative functional micro-aperture array. Here, N is an integer greater than 1.

[0045] S402: Based on the pre-calibrated micro-aperture position mapping data, perform dark field correction, flat field correction and sub-aperture segmentation on the image acquisition frame to obtain N independent sub-aperture image blocks, and determine the quality weight vector corresponding to each sub-aperture image block.

[0046] In practical applications, the image acquisition frame is first preprocessed, including dark field correction, flat field correction, sub-aperture segmentation, and sub-aperture quality assessment. Dark field correction and flat field correction involve subtracting the dark field frame from the image acquisition frame, dividing by the flat field frame, and eliminating fixed pattern noise from the image sensor and the non-uniformity of light intake at each aperture. Sub-aperture segmentation is performed based on the factory-calibrated DFMA aperture position mapping table (DFMA calibrates the pixel coordinates of each aperture on the photosensitive surface of the image sensor using a standard light source before leaving the factory, storing the coordinate correspondence as a mapping table), precisely segmenting the image acquisition frame into N sub-aperture image blocks. Sub-aperture quality assessment evaluates the signal-to-noise ratio (SNR) of each sub-aperture image based on local variance and gradient statistics, generating a quality weight vector corresponding to each sub-aperture image block. Through the above preprocessing process, the imaging deviation caused by the sensor and the micro-aperture array can be uniformly corrected, and high signal-to-noise ratio effective sub-aperture data can be screened. In subsequent light field fusion and depth estimation, the interference of low-quality image patches can be weakened based on the quality weight vector, which can greatly improve the accuracy of 3D scene reconstruction and the final imaging clarity.

[0047] S403: Based on geometric calibration data, perform disparity estimation and sub-pixel level geometric alignment on N-path sub-aperture image blocks to construct a light field tensor; combine the quality weight vector and perform weighted fusion on the light field tensor through a multi-view fusion algorithm to obtain an initial fused image.

[0048] Furthermore, the method also includes generating a depth map based on the disparity information in the light field tensor.

[0049] In practical applications, after completing the preprocessing of sub-aperture images and obtaining the quality weight vectors corresponding to each sub-aperture image block, the process proceeds to multi-view sub-aperture alignment and compound eye optical field reconstruction. This application relies on the factory-preset DFMA geometric calibration data to accurately match the optical path geometric parameters of each light-transmitting microaperture and optional plano-convex microlens. It can combine optical flow algorithms to perform pixel-by-pixel disparity calculations on all N-path sub-aperture image blocks, solving for the disparity offset field between each sub-aperture image. This achieves sub-pixel-level geometric alignment of multi-path sub-aperture images, effectively eliminating positional offset and geometric distortion problems between multi-view images and ensuring the registration accuracy of multi-path sub-aperture images. For disparity estimation, two optional algorithms are provided: First, a 4-layer pyramid Farneback dense optical flow is used by default to solve the pixel-by-pixel floating-point disparity from coarse to fine, combined with a Random Sample Consensus (RANSAC) algorithm to remove outlier disparity points. Second, for low-computing-power scenarios, a lightweight sparse feature matching algorithm (Oriented FAST and Rotated BRIEF, ORB) is used for feature matching and bicubic interpolation to generate a dense disparity map. During calculation, the geometrically calibrated microaperture baseline and sensor intrinsic parameters are read, and pixel displacements are converted into true disparity. For the sub-pixel level geometric alignment process, the floating-point disparity output by the optical flow can be used for reverse mapping, and 0.01-pixel-level accurate registration is achieved through 4×4 neighborhood bicubic interpolation resampling. After alignment, invalid edge regions are cropped, and quality weights for each sub-aperture are bound for subsequent weighted fusion. After geometric alignment, all aligned sub-aperture images are organized according to angular coordinates (u, v) and spatial coordinates (x, y) to construct a complete multidimensional light field tensor L(x, y, u, v), thus preserving the light field imaging information from different viewpoints and spatial locations. Here, x represents the horizontal spatial dimension of the image, y represents the vertical spatial dimension of the image, u represents the horizontal angular offset dimension between the multiple sub-apertures, and v represents the vertical angular offset dimension between the multiple sub-apertures. Furthermore, this application combines the quality weight vectors of each sub-aperture image obtained in the preprocessing stage to apply weighted constraints to the corresponding sub-aperture data within the light field tensor. Based on the reference multi-view FourierNet (SV-FourierNet) architecture, a multi-view fusion algorithm (MVS-SR) is used to complete the weighted fusion processing of the light field tensor. This fully utilizes the sub-pixel offset complementary information of the multi-aperture images to compensate for the low resolution and lack of detail in a single sub-aperture image. The final output is an initial fused image with effectively improved resolution. This initial fused image retains complete scene details with only a small amount of diffraction blur and slight noise.Furthermore, the SV-FourierNet architecture can be replaced with a NonlinearActivation Free Network (NAFNet) or a Restoration Transformer (Restormer). MVS-SR can be downgraded to simple weighted fusion, sacrificing the gains from Super-Resolution (SR) processing in exchange for reduced latency, making it suitable for real-time Simultaneous Localization and Mapping (SLAM) scenarios. Simultaneously, based on the dense disparity information carried by the constructed light field tensor, high-precision dense depth maps of the scene can be generated in parallel, providing reliable depth data support for core applications such as SLAM localization, virtual object occlusion rendering, and 3D scene reconstruction for AR devices.

[0050] S404: Based on the pre-calibrated point spread function, the initial fused image is subjected to diffraction blur correction to obtain a deblurred fused image.

[0051] Furthermore, the diffraction blur correction of the initial fused image includes: The initial fused image is pre-filtered in the frequency domain using Wiener deconvolution, and / or deblurred using a deep deconvolution network.

[0052] In practical applications, after obtaining the initial fused image through multi-view fusion, a diffraction correction process is performed. This application first obtains the pre-calibrated (factory-calibrated) point spread function (PSF), which characterizes the diffraction blurring features produced by the light-transmitting microaperture optical path, or the combined optical path of the light-transmitting microaperture and plano-convex microlens. In one optional embodiment, the correction process can be divided into two stages: first, Wiener frequency domain pre-filtering is performed, and a Fourier transform is applied to the point spread function to obtain the frequency domain transfer function H(f), according to the formula... The Wiener filter function is obtained by solving for it. This filter function is then used to perform deconvolution on the initial fused image to initially eliminate low-pass blur caused by optical path diffraction. Here, W(f) is the Wiener filter function, f represents the two-dimensional frequency domain coordinates of the image, and H(f) is the frequency domain transfer function. For the complex conjugate of H(f), Let H(f) represent the square of the modulus, and SNR be the signal-to-noise ratio of the sub-aperture image. The Wiener-filtered image is then fed into a deep deconvolutional network with a pre-calibrated PSF as conditional input. This network, finely tuned based on the NAFNet and Restormer architectures, effectively suppresses the ringing effect that easily occurs during deconvolution, restores lost high-frequency texture details, improves the modulation transfer function (MTF) response, recovers high-frequency texture details, and ultimately outputs a significantly sharper, deblurred, fused image.

[0053] S405: Perform adaptive noise reduction, super-resolution reconstruction and moiré pattern removal on the deblurred fused image in sequence to obtain the target fused image.

[0054] In practical applications, after completing diffraction correction to obtain a deblurred and fused image, AI noise reduction and super-resolution enhancement processing are performed. First, adaptive AI noise reduction is performed. The deblurred and fused image is input into an automatic noise level estimation noise reduction network designed based on the Fast Flexible Denoising Network (FFDNet) architecture. This network can autonomously identify and suppress various residual noises amplified by deconvolution operations within the image. Then, super-resolution reconstruction is performed on the denoised image. A Real-Enhanced Super-Resolution Generative Adversarial Network (RealESRGAN) or a Residual Channel Attention Network (RCAN) finely tuned to the characteristics of microaperture imaging is used for processing, resulting in a super-resolution image with a resolution improvement of 2-4 times. Finally, moiré pattern removal is performed. The frequency components of moiré patterns in the image are located through spectral analysis, and targeted removal of moiré patterns is achieved using a wavelet-based dual-branch network (WDNet)-like network. Ultimately, a high-resolution, low-noise target fused image is obtained, with imaging visual quality approaching that of a traditional lens-type camera.

[0055] S406: Identify the scene category corresponding to the target fused image.

[0056] In practical applications, after obtaining the target fused image, scene category recognition is performed on the image. A lightweight scene classification network (Mobile Network Version 3, MobileNetV3) is used to complete the recognition operation. This network can run on the device's Neural Processing Unit (NPU) with a processing latency of less than 5ms. It can distinguish various scenes such as landscapes, portraits, food, documents, night scenes, and motion, and output the corresponding scene category results, providing a basis for subsequent scene-specific adaptive adjustment of imaging processing parameters.

[0057] S407: Based on the scene category, match the corresponding virtual exposure strategy and tone enhancement parameters, output the optimized target imaging result, and simultaneously generate intelligent composition suggestions.

[0058] Furthermore, the step of matching the virtual exposure strategy and tone enhancement parameters corresponding to the scene category, outputting the optimized target imaging result, and simultaneously generating intelligent composition suggestions specifically includes: Based on the scene category, dynamically adjust the equivalent ISO sensitivity parameter, gain curve, and HDR tone mapping parameter of the target fused image; Load the color lookup table corresponding to the scene category, perform targeted tone optimization on the target fusion image, and obtain the optimized target imaging result; The main region of the target fused image is identified by a saliency detection network, the target cropping box that conforms to the preset composition rules is calculated, and intelligent composition suggestions are generated by AR display overlay or automatic cropping. Output the target imaging results and the intelligent mapping suggestions.

[0059] In practical applications, after identifying the scene category corresponding to the target fused image, the corresponding virtual exposure strategy and (semantic) tone enhancement parameters are matched based on the scene category to complete the image semantic optimization processing, output the optimized target imaging result, and generate intelligent composition suggestions. Specifically, firstly, the Virtual Exposure Strategy (VES) is adjusted, dynamically adapting the equivalent ISO sensitivity parameters, gain curve, and HDR tone mapping parameters according to the identified scene category. For example, it enhances the details in dark areas of night scenes and protects the original skin tone from color cast in portrait scenes. Then, semantic tone enhancement is performed, retrieving the scene-specific color look-up table (LUT) for color optimization. For example, it enhances the color saturation of blue sky and green plants in landscape scenes and performs fine-grained skin tone correction in portrait scenes. At the same time, the main subject region within the image is identified through a saliency detection network, and the optimal cropping box is calculated. The cropping box can be visualized and overlaid on the AR interface or automatically cropped directly. Finally, the semantically optimized target imaging result is output, along with scene labels and intelligent composition suggestions.

[0060] It should be noted that the SV-FourierNet, FFDNet, RealESRGAN, and WDNet used above are all existing open-source image restoration networks. The original network architecture, training strategies, and basic parameters can all be obtained completely from the corresponding academic papers and open-source code. This application does not modify the kernel structure of each network, but only adapts the input dimension for DFMA microaperture diffraction imaging features: the network input channels are adapted to multi-path sub-aperture light field tensors, the input pixel size is matched to the output resolution of the BSI-CMOS sensor, and the device's factory PSF calibration parameters are loaded as network conditional inputs during the inference stage, without the need to redesign the network layers and loss function.

[0061] Furthermore, since the methods for ensuring stable timing output are not entirely the same, the embodiments of this application can be described for one possible processing method.

[0062] In one instance, the method further includes: Based on the inter-frame consistency constraint of optical flow, the inter-frame flicker of the target imaging result is suppressed; Based on the output application scenario, the target imaging result is converted into a target format that matches the output application scenario.

[0063] In practical applications, after obtaining the target imaging results with frame-by-frame semantic optimization, inter-frame temporal stabilization and format adaptation output processing are performed. First, video mode temporal smoothing is applied, using an optical flow-based inter-frame consistency constraint strategy to perform temporal correction on continuous image frame sequences. This effectively eliminates inter-frame brightness, color, and detail jitter, suppresses video flicker, and improves the smoothness and consistency of video playback. Then, differentiated format adaptation output is performed according to different output application scenarios. For example, for AR perception scenarios, low-latency RAW data or interface data adapted for SLAM operations is output; for daily life recording scenarios, Joint Photographic Experts Group (JPEG) image format or H.265 video encoding format is output; for video call scenarios, real-time compressed video streams are output. Finally, the high-quality images or videos that have undergone end-to-end AI image enhancement processing are uniformly output to upper-layer applications through the AR main processor.

[0064] In summary, the computational imaging processing method for a hidden camera in AR glasses provided in this application first acquires an image frame from an image sensor. After dark field correction and flat field correction, the image frame is segmented into N independent sub-aperture image blocks, and the mass weight vector corresponding to each sub-aperture image block is determined. The image frame includes N sub-aperture imaging regions corresponding to N light-transmitting micro-apertures in a decorative functional micro-aperture array; N is an integer greater than 1. Then, based on geometric calibration data, disparity estimation and sub-pixel-level geometric alignment are performed on the N sub-aperture image blocks to construct a light field tensor. Combined with the mass weight vector, a multi-view fusion algorithm is used to weighted fuse the light field tensor to obtain an initial fused image. Next, based on a pre-calibrated point spread function, diffraction blur correction is applied to the initial fused image to obtain a deblurred fused image. Finally, adaptive noise reduction, super-resolution reconstruction, and moiré pattern removal are performed on the deblurred fused image to obtain the target fused image. Finally, the scene category corresponding to the target fused image is identified; based on the scene category, the corresponding virtual exposure strategy and tone enhancement parameters are matched, and the optimized target imaging result is output, while intelligent composition suggestions are generated simultaneously. In this way, by performing frame-by-channel correction, geometric alignment weighted fusion and correction of diffraction blur on the original images simultaneously acquired by multiple light-transmitting micro-apers, combined with scene-adaptive exposure and tone optimization processing, and relying on the decorative functional micro-aperture array to achieve seamless camera concealment and isolation of optical path crosstalk, the imaging clarity and scene optimization capabilities are greatly improved, adapting to the needs of AR glasses photography and spatial perception.

[0065] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A hidden camera system for AR glasses, characterized in that, The system is integrated into the frame structure of AR glasses and includes: a decorative functional micro-perforation array, an image sensor, a light-blocking and isolation structure, and an image processing unit; The decorative functional micropore array is disposed on the outer surface of the frame structure, presenting an overall geometric texture pattern with a decorative effect. The decorative functional micropore array includes multiple light-transmitting micropores and multiple opaque decorative points. The light-transmitting micropores serve as sampling sub-apertures for the light field acquisition of the compound eye, and are mixed and arranged with the opaque decorative points according to a preset geometric texture pattern. The light-transmitting micropores and the opaque decorative points have the same appearance size, color, and / or surface treatment effect, so that an external observer cannot distinguish between the light-transmitting micropores and the opaque decorative points under preset observation conditions. The image sensor is disposed on the back side of the decorative functional micro-hole array and located in the light-shielding and receiving space formed inside the frame structure. It is used to receive ambient light transmitted through the light-transmitting micro-holes and generate a multi-path sub-aperture image corresponding to the sampling sub-aperture. The light-shielding and isolation structure is disposed between the decorative functional micro-hole array and the image sensor to isolate the light path of adjacent light-transmitting micro-holes; The image processing unit is connected to the image sensor signal. The image processing unit has a built-in AI perception-enhanced photography processing pipeline, which is used to perform geometric alignment, light field fusion, image quality enhancement and semantic optimization processing on the multi-aperture images, and output the target imaging results.

2. The system according to claim 1, characterized in that, The geometric texture pattern is any one or more combinations of dot matrix patterns, hexagonal honeycomb patterns, and laser-etched grid patterns.

3. The system according to claim 1, characterized in that, The decorative functional micro-perforation array is disposed at any one or more of the following locations: the head region of the AR glasses, the side panel of the front frame, and the inner surface of the proximal end of the temple.

4. The system according to claim 1, characterized in that, The diameter of the light-transmitting micropores is 50-300μm, and the number is 16-256; the effective spatial baseline between the light-transmitting micropores participating in light field acquisition within the decorative functional micropore array is 3-20mm.

5. The system according to claim 1, characterized in that, The image sensor has a pixel size of no more than 1.8 μm and a readout noise of less than 2.5 e.g. - .

6. The system according to claim 1, characterized in that, The system also includes: plano-convex microlenses; The plano-convex microlens is disposed behind at least part of the light-transmitting micro-aperture, and the plano-convex microlens is disposed 1 mm behind any light-transmitting micro-aperture having the plano-convex microlens, and the focal length of the plano-convex microlens is 1-2 mm.

7. The system according to claim 1, characterized in that, The light-shielding and isolation structure includes at least one of a black light-shielding resin partition wall, a light-shielding microchannel array, a metal light-shielding partition plate, and an integrally molded light-shielding wall, wherein the light-shielding and isolation structure defines an independent optical path corresponding to each light-transmitting micro-hole.

8. The system according to claim 1, characterized in that, The image processing unit stores or retrieves factory calibration data, which includes at least one of the following: a position mapping table of the light-transmitting micro-apertures on the photosensitive surface of the image sensor, baseline parameters between apertures, and point spread function parameters corresponding to each light-transmitting micro-aperture.

9. A computational imaging processing method for a hidden camera in AR glasses, characterized in that, The method includes: The image acquisition frame acquired by the image sensor includes N sub-aperture imaging regions corresponding to N light-transmitting micro-apertures in the decorative functional micro-aperture array; where N is an integer greater than 1. Based on the pre-calibrated microaperture position mapping data, dark field correction, flat field correction and sub-aperture segmentation are performed on the image acquisition frame to obtain N independent sub-aperture image blocks, and the quality weight vector corresponding to each sub-aperture image block is determined. Based on geometric calibration data, disparity estimation and sub-pixel-level geometric alignment are performed on N-path sub-aperture image blocks to construct a light field tensor; combined with the quality weight vector, the light field tensor is weighted and fused using a multi-view fusion algorithm to obtain an initial fused image. Based on the pre-calibrated point spread function, the initial fused image is subjected to diffraction blur correction to obtain a deblurred fused image; The deblurred fused image is sequentially subjected to adaptive noise reduction, super-resolution reconstruction, and moiré pattern removal to obtain the target fused image; Identify the scene category corresponding to the target fused image; Based on the scene category, the corresponding virtual exposure strategy and tone enhancement parameters are matched, and the optimized target imaging result is output, while intelligent composition suggestions are generated simultaneously.

10. The method according to claim 9, characterized in that, The method further includes: Based on the inter-frame consistency constraint of optical flow, the inter-frame flicker of the target imaging result is suppressed; Based on the output application scenario, the target imaging result is converted into a target format that matches the output application scenario.

11. The method according to claim 9, characterized in that, Before performing diffraction blur correction on the initial fused image based on the pre-calibrated point spread function to obtain the deblurred fused image, the method further includes: A depth map is generated based on the disparity information in the light field tensor.

12. The method according to claim 9, characterized in that, The diffraction blur correction of the initial fused image includes: The initial fused image is pre-filtered in the frequency domain using Wiener deconvolution, and / or deblurred using a deep deconvolution network.

13. The method according to claim 9, characterized in that, The virtual exposure strategy and tone enhancement parameters corresponding to the scene category are matched to output the optimized target imaging result, and intelligent composition suggestions are generated simultaneously, including: Based on the scene category, dynamically adjust the equivalent ISO sensitivity parameter, gain curve, and HDR tone mapping parameter of the target fused image; Load the color lookup table corresponding to the scene category, perform targeted tone optimization on the target fusion image, and obtain the optimized target imaging result; The main region of the target fused image is identified by a saliency detection network, the target cropping box that conforms to the preset composition rules is calculated, and intelligent composition suggestions are generated by AR display overlay or automatic cropping. Output the target imaging results and the intelligent mapping suggestions.

Images