A mobile terminal-oriented scene self-adaptive real-time projection compensation system and method

Abstract

This invention relates to a scene-adaptive real-time projection compensation system for mobile devices, belonging to the field of projection compensation technology. It includes a mobile control terminal and a projection terminal. The mobile control terminal performs projection feedback acquisition, scene initialization, cache management, real-time compensation inference, display output, and benchmark testing. The projection terminal receives the compensated image or video frame output by the mobile control terminal and projects it onto a target projection surface. This system, with scene initialization caching and lightweight real-time inference as its core, decouples complex scene modeling from the real-time compensation path. It utilizes the mobile device's built-in camera, local fixed geometric transformations, mobile neural network inference, and remote benchmark testing to form a deployable, testable, and scalable mobile projection compensation closed loop. It also includes supporting methods for the system.

Description

Technical Field

[0001] This invention belongs to the field of projection compensation technology, and relates to a scene-adaptive real-time projection compensation system and method for mobile devices. Background Technology

[0002] Projection compensation aims to pre-correct input images or video frames based on the relationship between the projection device, the projection surface, and the viewing conditions, so that the image projected onto a non-ideal surface is as close as possible to the desired visual effect. This technology can be applied to scenarios such as portable projection, mobile displays, human-computer interaction, spatial augmented reality, conference presentations, and educational demonstrations.

[0003] In practical use, projected images are usually not projected onto a standard flat screen, but rather onto walls, wooden surfaces, colored background surfaces, textured surfaces, or surfaces with local undulations. Due to factors such as projection angle, surface morphology, surface reflectivity, and ambient lighting, the projected image seen by the observer is prone to problems such as geometric distortion, uneven brightness, color shift, texture interference, and decreased local contrast.

[0004] Existing projection compensation solutions typically rely on projector-camera systems for calibration and compensation, requiring an external fixed camera to acquire the projected feedback image and to perform geometric estimation, photometric modeling, and compensation inference on a desktop computing device. While these solutions can achieve relatively stable feedback, they increase the complexity of system setup and reduce the portability of deployment in mobile scenarios.

[0005] Some existing solutions utilize the camera of a mobile phone or other mobile device to capture the projected image and adjust the projector's operating parameters, four-corner correction parameters, focal length parameters, or perspective transformation parameters based on the captured image. While these solutions can improve some geometric distortions, their technical focus is typically on determining projector parameters or homography through feedback images. They struggle to simultaneously address issues related to color, texture, reflection, and ambient light compensation on non-ideal projection surfaces, and have not yet developed a compensation path suitable for real-time execution by mobile neural network acceleration units.

[0006] Other solutions propose color correction, Lambertian reflection modeling, nonlinear photometric compensation, or projection mapping methods for projector-camera systems. These solutions are typically based on desktop computing devices, external cameras, projector-camera calibration, or pixel-by-pixel photometric mapping, and do not address system design for mobile operating systems and other mobile device environments, such as operator support, memory usage, model export, edge acceleration unit scheduling, and real-time video links.

[0007] Furthermore, existing deep learning projection compensation models often include complex optical flow estimation, attention structures, and grid sampling operators that are not suitable for mobile runtime. In mobile environments, model deployment is also limited by factors such as neural network runtime, acceleration unit operator support, memory footprint, end-to-end latency, power consumption, and temperature control strategies. Summary of the Invention

[0008] In view of this, the purpose of the present invention is to provide a scene-adaptive real-time projection compensation system and method for mobile devices.

[0009] To achieve the above objectives, the present invention provides the following technical solution: On one hand, the present invention provides a scene-adaptive real-time projection compensation system for mobile devices, including a mobile control terminal and a projection terminal; the mobile control terminal is used to perform projection feedback acquisition, scene initialization, cache management, real-time compensation inference, display output, and benchmark testing; the projection terminal is used to receive the compensation image or compensation video frame output by the mobile control terminal and project it onto the target projection surface.

[0010] Furthermore, the mobile control terminal includes an acquisition and control module, a scene initialization module, a geometry pre-alignment module, a lightweight compensation inference module, and an output module; The acquisition and control module is used to control the projection end to output a reference image sequence, and to acquire the feedback image of the reference image sequence on the target projection surface through the built-in camera of the mobile control end. The scene initialization module is used to generate a geometric mapping cache and a surface feature cache corresponding to the current projection scene based on the feedback image, and bind and store them with scene identifier, projection area parameters, model version and input / output size information; The geometry pre-alignment module is used to read the geometry mapping cache during the real-time compensation phase and perform fixed geometry pre-alignment on the image or video frame to be compensated without re-performing the complete projector-camera geometry estimation frame by frame. The lightweight compensation inference module is used to compress, distill, replace or fix the pre-aligned input frame and surface feature cache input to obtain a lightweight projection compensation model, and call the mobile neural network acceleration unit to perform compensation inference. The output module is used to output the compensated image or compensated video frame to the projection end.

[0011] Furthermore, the mobile control terminal also includes a remote benchmark testing module, which is used to receive remote control commands, schedule initialization, single-frame compensation, mobile phone shooting compensation or video compensation tasks, and transmit compensation results and performance statistics.

[0012] On the other hand, the present invention provides a scene adaptive real-time projection compensation method for mobile devices, comprising: Initialization Phase: The mobile control terminal controls the projection terminal to project a sequence of reference images onto the target projection surface, and acquires feedback images through the mobile control terminal's built-in camera; the scene initialization module determines the projection area boundary, projection area pose, and local geometric relationships based on the feedback images, and generates a geometric mapping cache; the scene initialization module generates a surface feature cache based on multiple surface reference feedback images; the mobile control terminal binds and stores the geometric mapping cache, surface feature cache, scene identifier, projection area parameters, model version, and input / output size information for subsequent real-time compensation phase retrieval and reuse; Real-time compensation stage: The mobile control terminal reads the image or video frame to be compensated, the geometric pre-alignment module reads the geometric mapping cache and performs a fixed geometric transformation to obtain the pre-aligned input frame; the lightweight compensation inference module inputs the pre-aligned input frame and the surface feature cache into the lightweight projection compensation model to generate the compensated image or compensated video frame.

[0013] Furthermore, the reference image sequence may include a geometric reference image, a surface reference image, a grayscale reference image, a color reference image, or a combination thereof.

[0014] Furthermore, the geometric mapping cache is a geometric transformation mesh used to represent the sampling relationship between the input image or model input coordinates and the projection region coordinates.

[0015] Furthermore, the surface feature cache is used to describe the color, texture, reflection, and ambient light conditions of the current target projection surface.

[0016] Furthermore, the real-time compensation phase specifically includes the following steps: Read the image or video frame to be compensated and convert it into the input format required by the geometry pre-alignment module; The geometry prealignment module performs fixed geometry prealignment based on the geometry mapping cache; Read the surface feature cache and input it along with the pre-aligned input frame into the lightweight projection compensation model; The mobile lightweight inference runtime or hardware acceleration interface calls the mobile neural network processing unit or other available edge acceleration units to perform model inference and output compensation results. Perform color gamut, size, and buffer post-processing on the compensation results, and output them to the display layer, projection end, or video encoding and streaming module.

[0017] Furthermore, the geometry pre-alignment module is implemented through a local compilation module, texture sampling by the graphics processing unit, hardware image processing interface, or lookup table sampling, in order to avoid embedding complex geometry sampling operations into the real-time inference graph of the neural network.

[0018] The beneficial effects of this invention are as follows: This invention migrates the acquisition process, which relies on an external fixed camera in traditional projector-camera systems, to the built-in camera of a mobile terminal, making the mobile terminal an integrated platform for acquisition, initialization, inference, display control, and verification, thus improving the portability of deployment in mobile scenarios. The built-in camera acquisition process serves as the scene initialization input and is combined with subsequent dual buffers and edge-side inference paths.

[0019] This invention separates complex geometry estimation and surface modeling from the real-time path through an initialization caching mechanism, so that the real-time stage mainly performs fixed geometry pre-alignment, cache reading and lightweight compensation model inference, thereby reducing edge latency and operator adaptation difficulty.

[0020] This invention designs the model export and inference process around the mobile operating environment, such as the mobile lightweight inference runtime or hardware acceleration interface and the mobile neural network processing unit, and places the fixed geometry pre-alignment outside the real-time inference graph of the neural network, making the compensation model more suitable for running on real mobile devices.

[0021] This invention covers both fixed-scene deployment baseline and scene adaptive closed loop, which can not only verify the feasibility of the model on mobile devices, but also support rapid adaptation after changes in the projection surface or experimental settings.

[0022] This invention solidifies the real-machine testing process through a remote benchmarking module, which facilitates repeatable recording and horizontal comparison of compensation quality, latency, frame rate, and memory usage across different models, scenarios, and devices. Other advantages, objectives, and features of this invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of this invention can be realized and obtained through the following description. Attached Figure Description

[0023] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1 This is a diagram showing the overall structure of a scene-adaptive real-time projection compensation system for mobile devices. Figure 2 This is a block diagram of the functional modules of the mobile control terminal. Figure 3 A flowchart of the overall process for a mobile-oriented adaptive real-time projection compensation method. Figure 4 For real-time compensation of data flow graphs; Figure 5 This is a flowchart for remote benchmarking and closed-loop verification. Detailed Implementation

[0024] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0025] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Therefore, the drawings only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0026] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the invention. However, it will be apparent to those skilled in the art that embodiments of the invention may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the invention.

[0027] Example 1: like Figure 1 As shown, this invention provides a scene-adaptive real-time projection compensation system for mobile devices, including a mobile control terminal 10, a projection terminal 20, and a target projection surface 30. The mobile control terminal 10 can be a smartphone with a mobile operating system, a tablet computer, or other mobile devices with built-in cameras, mobile system-on-a-chip, neural network acceleration units, and video output capabilities. The projection terminal 20 can receive the compensated image or compensated video frame output by the mobile control terminal 10 via a wired video output interface, a wireless projection link, or a local area network streaming link. During the real-time compensation phase, the projection terminal 20 receives the compensated content generated by the mobile control terminal 10. This invention does not require real-time adjustment of the focal length, projection angle, four-corner correction parameters, or brightness parameters of the projection terminal 20 as a necessary compensation step. Wherein: The mobile control terminal 10 is used to perform projection feedback acquisition, scene initialization, cache management, real-time compensation inference, display output, and benchmark testing.

[0028] The projection end 20 is used to receive the compensated image or compensated video frame output by the mobile control end and project it onto the target projection surface.

[0029] Non-ideal projection surfaces 30 include walls, screens, wooden surfaces, colored substrates, textured surfaces, or projection surfaces with local undulations.

[0030] The input image or video frame 40 includes a single frame image to be compensated, a mobile phone captured image, a video frame, or a remote test input.

[0031] The initialization cache 50 includes at least a geometry mapping cache and a surface feature cache, and may further include scene identifiers, projection region parameters, model version, and input / output size information.

[0032] The lightweight real-time compensation path 60 includes fixed geometry pre-alignment, cached feature reading, lightweight projection compensation model inference, output post-processing, and projection display.

[0033] like Figure 2 As shown, the mobile control terminal 10 includes, in terms of software logic, an acquisition and control module 110, a scene initialization module 120, a geometry pre-alignment module 130, a lightweight compensation inference module 140, an output module, and an optional remote benchmarking module. These modules can be integrated into mobile operating system applications, system services, or mobile software development kits.

[0034] The acquisition and control module 110 is used to control the output of the reference image sequence by the projection end, and to acquire the feedback image of the reference image sequence on the target projection surface through the built-in camera of the mobile control end.

[0035] The scene initialization module 120 is used to generate a geometric mapping cache and a surface feature cache corresponding to the current projection scene based on the feedback image, and bind and store them with scene identifier, projection area parameters, model version and input / output size information.

[0036] The geometry pre-alignment module 130 is used to read the geometry mapping cache during the real-time compensation phase and perform fixed geometry pre-alignment on the image or video frame to be compensated without re-performing the full projector-camera geometry estimation frame by frame.

[0037] The lightweight compensation inference module 140 is used to compress, distill, replace or fix the pre-aligned input frame and surface feature cache input to obtain a lightweight projection compensation model, and call the mobile neural network acceleration unit to perform compensation inference.

[0038] The output module 150 is used to output the compensated image or video frame.

[0039] The remote benchmark module 160 is used to receive remote control commands, schedule initialization, single-frame compensation, mobile phone shooting compensation or video compensation tasks, and transmit compensation results and performance statistics back.

[0040] Example 2: like Figure 3 As shown, this embodiment provides a scene-adaptive real-time projection compensation method for mobile devices, including a scene initialization phase and a real-time compensation phase.

[0041] The scene initialization phase includes: Step 1: The mobile control terminal controls the projection terminal to project a sequence of reference images onto the target projection surface, and acquires feedback images through the built-in camera of the mobile control terminal. The sequence of reference images may include geometric reference images, surface reference images, grayscale reference images, color reference images, or combinations thereof.

[0042] Step 2: The scene initialization module determines the projection region boundary, projection region pose, and local geometric relationships based on the feedback image, and generates a geometric mapping cache. In one embodiment, the geometric mapping cache is a geometric transformation mesh, used to represent the sampling relationship between the input image or model input coordinates and the projection region coordinates.

[0043] Step 3: The scene initialization module generates a surface feature cache based on multiple surface reference feedback images. The surface feature cache is used to describe the color, texture, reflection, and ambient light conditions of the current target projection surface.

[0044] Step 4: The system binds and stores the geometric mapping cache, surface feature cache, scene identifier, projection area parameters, model version, and input / output size information for subsequent real-time compensation stage reading and reuse, and avoids reprojecting the correction template or recalculating the complete projector-camera geometric correspondence on each image or video frame to be compensated.

[0045] Data stream during the real-time compensation phase, such as Figure 4 As shown, after the system reads the image or video frame to be compensated, the geometric pre-alignment module first reads the geometric mapping cache and performs a fixed geometric transformation to obtain the pre-aligned input frame. Subsequently, the lightweight compensation inference module inputs the pre-aligned input frame and the surface feature cache into the lightweight projection compensation model to generate the compensated image or compensated video frame. The real-time compensation stage primarily uses cache reading, fixed geometric pre-alignment, and end-side lightweight model inference as the main processing path, rather than repeatedly adjusting the projection end's operating parameters as a necessary step.

[0046] Step 1: The system reads the image or video frame to be compensated and converts it into the input format required by the geometric pre-alignment module.

[0047] Step 2: The geometry pre-alignment module performs fixed geometry pre-alignment based on the geometry mapping cache. In a preferred embodiment, this module is implemented through a local compilation module, texture sampling by the graphics processing unit, hardware image processing interface, or lookup table sampling, to avoid embedding complex geometry sampling operations into the real-time inference graph of the neural network.

[0048] Step 3: The system reads the surface feature cache and inputs it together with the pre-aligned input frame into the lightweight projection compensation model.

[0049] Step 4: The mobile lightweight inference runtime or hardware acceleration interface calls the mobile neural network processing unit or other available edge acceleration units to perform model inference and output the compensation results.

[0050] Step 5: The system performs necessary post-processing on the compensation results, including color gamut, size, and buffer settings, and outputs the results to the mobile phone display layer, projection end, or video encoding and streaming module.

[0051] This invention places complex geometry estimation and surface condition modeling before the scene initialization stage. It saves information related to the current projection scene through geometry mapping cache and surface feature cache, and binds the cache with scene identifier, projection area parameters, model version and input / output size information, so that the real-time compensation path no longer executes the heavy geometry estimation network frame by frame.

[0052] In the real-time compensation phase, a fixed geometric pre-alignment is performed on the image or video frame to be compensated based on the geometric mapping cache, and this fixed geometric pre-alignment is located outside the real-time inference graph of the lightweight projection compensation model. In other words, the system resamples the input frame to be compensated according to the sampling relationship stored in the geometric mapping cache to obtain the pre-aligned input frame. This description is only for illustrating the implementation mechanism and does not limit the specific interpolation method, grid resolution, or cache format.

[0053] The lightweight compensation inference module inputs the pre-aligned input frame and the surface feature cache into the lightweight projection compensation model to obtain the compensation output. The lightweight projection compensation model can be obtained through one or more of the following methods: teacher-student distillation, model compression, operator replacement, input-output fixation, quantization, and mobile runtime adaptation. It can also reduce or remove one or more mobile-unfriendly operators from complex optical flow estimation, attention modules, or grid sampling operators.

[0054] The lightweight projection compensation model can be exported as a lightweight inference runtime or a hardware-accelerated interface executable model for mobile devices, and can call mobile neural network processing units, digital signal processing units, graphics processing units or other edge-side heterogeneous computing units to perform the main neural network calculations on mobile devices.

[0055] like Figure 5 As shown, the present invention also performs remote benchmark testing and closed-loop verification. The remote browser or control terminal can send tasks to the mobile terminal control service through network control commands. After the system executes the task, it sends the compensation results, video preview stream, end-to-end frame rate, latency at each stage, and memory usage back to the remote control terminal.

[0056] This invention generates compensated image or video frames and outputs them as input to the projection end, which then projects and displays the received content. This mechanism differs from correction methods that only adjust the projection end's focal length, projection angle, corner correction parameters, or brightness parameters based on feedback images.

[0057] Example 3: In this embodiment, when the relative positions of the mobile phone, the projector, and the target projection surface remain unchanged, the system adopts a fixed scene deployment mode. In this mode, scene-related parameters are pre-defined in configuration files, cache files, or model inputs. The mobile control terminal directly loads the fixed scene model and corresponding cache to compensate for the input image or video frames.

[0058] This fixed-scene deployment mode can be used to verify the feasibility of the projection compensation model running on real mobile devices, including model format conversion, model loading, edge inference, single-frame compensation, video compensation, output display, latency statistics, frame rate statistics, and memory statistics.

[0059] In another implementation, the mobile control terminal and the projection terminal are connected via a wired video interface, with the target projection surface being the conference room wall. Upon initial system use, the mobile control terminal controls the projection terminal to project a sequence of reference images, and the mobile phone camera captures the feedback images. The scene initialization module generates corresponding geometric mapping caches and surface feature caches, binding them to scene identifiers, model versions, and input / output dimensions. During the real-time compensation phase, the system reads presentation images or video frames, first performing local fixed geometric transformations, then calling the mobile terminal's lightweight model for compensation inference, and finally outputting the compensation result as video input to the projection terminal.

[0060] In another implementation, the mobile control terminal and the projection terminal are connected via wireless projection or a local area network streaming link, and the target projection surface is a wooden surface or a wall with a colored primer. When the system detects a significant change in ambient light or the projection area, it can re-acquire the surface reference image and update the surface feature cache; when it detects a change in the projection area boundary or device posture exceeding a threshold, it can regenerate the geometric mapping cache.

[0061] When the model version, scene identifier, input / output dimensions, or cache file are inconsistent, the system will prevent entry into the real-time compensation process and prompt the user to reload the matching cache or re-execute the new scene initialization.

[0062] When the system detects that changes in the projection area boundary, camera viewpoint, or device posture exceed a preset threshold, it marks the current geometry mapping cache as invalid and prompts the user to re-execute the new scene initialization or partially update the geometry cache.

[0063] When there are significant changes in the brightness distribution, color shift, or surface texture statistics of the feedback image, the system can re-acquire the surface reference image and update the surface feature cache; if necessary, the geometric mapping cache remains unchanged to shorten the update latency.

[0064] When the mobile lightweight inference backend fails to initialize, the target operator is not supported, or the device's temperature control strategy restricts the use of the neural network processing unit, the system can switch to the central processing unit, graphics processing unit, or other backup models, or prompt the user to select a lighter model version.

[0065] When the end-to-end frame rate is lower than a preset threshold, the system can reduce the input resolution, reduce the video frame processing frequency, switch to a fixed scene model, turn off remote preview, or adjust encoding parameters to maintain stable compensation output.

[0066] When the network control connection, browser control terminal, or video streaming link is interrupted, the mobile operating system application retains the local compensation and display process and records the abnormal state, allowing the remote end to read the log after the connection is restored.

[0067] 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 present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A scene-adaptive real-time projection compensation system for mobile devices, characterized in that: It includes a mobile control terminal and a projection terminal; the mobile control terminal is used to perform projection feedback acquisition, scene initialization, cache management, real-time compensation inference, display output and benchmark testing; the projection terminal is used to receive the compensation image or compensation video frame output by the mobile control terminal and project it onto the target projection surface.

2. The scene-adaptive real-time projection compensation system for mobile devices according to claim 1, characterized in that: The mobile control terminal includes an acquisition and control module, a scene initialization module, a geometry pre-alignment module, a lightweight compensation inference module, and an output module; The acquisition and control module is used to control the projection end to output a reference image sequence, and to acquire the feedback image of the reference image sequence on the target projection surface through the built-in camera of the mobile control end. The scene initialization module is used to generate a geometric mapping cache and a surface feature cache corresponding to the current projection scene based on the feedback image, and bind and store them with scene identifier, projection area parameters, model version and input / output size information; The geometry pre-alignment module is used to read the geometry mapping cache during the real-time compensation phase and perform fixed geometry pre-alignment on the image or video frame to be compensated without re-performing the complete projector-camera geometry estimation frame by frame. The lightweight compensation inference module is used to compress, distill, replace or fix the pre-aligned input frame and surface feature cache input to obtain a lightweight projection compensation model, and call the mobile neural network acceleration unit to perform compensation inference. The output module is used to output the compensated image or compensated video frame to the projection end.

3. The scene-adaptive real-time projection compensation system and method for mobile terminals according to claim 2, characterized in that: The mobile control terminal also includes a remote benchmark testing module, which is used to receive remote control commands, schedule initialization, single-frame compensation, mobile phone shooting compensation or video compensation tasks, and transmit compensation results and performance statistics.

4. A scene-adaptive real-time projection compensation method for mobile devices, characterized in that: The system based on any one of claims 1-3 includes: Initialization Phase: The mobile control terminal controls the projection terminal to project a sequence of reference images onto the target projection surface, and acquires feedback images through the mobile control terminal's built-in camera; the scene initialization module determines the projection area boundary, projection area pose, and local geometric relationships based on the feedback images, and generates a geometric mapping cache; the scene initialization module generates a surface feature cache based on multiple surface reference feedback images; the mobile control terminal binds and stores the geometric mapping cache, surface feature cache, scene identifier, projection area parameters, model version, and input / output size information for subsequent real-time compensation phase retrieval and reuse; Real-time compensation stage: The mobile control terminal reads the image or video frame to be compensated, the geometric pre-alignment module reads the geometric mapping cache and performs a fixed geometric transformation to obtain the pre-aligned input frame; the lightweight compensation inference module inputs the pre-aligned input frame and the surface feature cache into the lightweight projection compensation model to generate the compensated image or compensated video frame.

5. The scene adaptive real-time projection compensation method for mobile devices according to claim 4, characterized in that: The reference image sequence may include geometric reference images, surface reference images, grayscale reference images, color reference images, or combinations thereof.

6. The scene adaptive real-time projection compensation method for mobile devices according to claim 4, characterized in that: The geometric mapping cache is a geometric transformation mesh used to represent the sampling relationship between the input image or model input coordinates and the projection region coordinates.

7. The scene adaptive real-time projection compensation method for mobile devices according to claim 4, characterized in that: The surface feature cache is used to describe the color, texture, reflection, and ambient light conditions of the current target projection surface.

8. The scene adaptive real-time projection compensation method for mobile devices according to claim 4, characterized in that: The real-time compensation phase specifically includes the following steps: Read the image or video frame to be compensated and convert it into the input format required by the geometry pre-alignment module; The geometry prealignment module performs fixed geometry prealignment based on the geometry mapping cache; Read the surface feature cache and input it along with the pre-aligned input frame into the lightweight projection compensation model; The mobile lightweight inference runtime or hardware acceleration interface calls the mobile neural network processing unit or other available edge acceleration units to perform model inference and output compensation results. Perform color gamut, size, and buffer post-processing on the compensation results, and output them to the display layer, projection end, or video encoding and streaming module.

9. The scene adaptive real-time projection compensation method for mobile devices according to claim 8, characterized in that: The geometry pre-alignment module is implemented through a local compilation module, texture sampling by the graphics processing unit, hardware image processing interface, or lookup table sampling, in order to avoid embedding complex geometry sampling operations into the real-time inference graph of the neural network.

Images