A projector image positioning method, device, equipment and storage medium

Abstract

The present application relates to the field of image positioning technology, and discloses a projector image positioning method, device, equipment and storage medium. The method comprises: projecting a preset reference image onto a target projection surface through a projector, and collecting a first projection image; constructing a relationship graph positioning network of the first projection image, and performing image offset and rotation angle correction on the projector based on the relationship graph positioning network to obtain a first image offset and a first rotation angle; performing real-time image projection on the target projection surface through the projector based on the first image offset and the first rotation angle, and collecting a second projection image; performing deep feature matching and deviation analysis on the second projection image through a deep convolutional neural network model to obtain a second image offset and a second rotation angle. The present application improves the image positioning accuracy of the projector.

Description

Technical Field

[0001] The present application relates to the field of image positioning technology, and in particular to a projector image positioning method, device, equipment and storage medium. Background Art

[0002] With the growing demand for high-precision and high-resolution projection, how to ensure that the projector's image is accurately aligned with the target projection surface has become an urgent problem that needs to be solved.

[0003] Traditional projection technology often relies on manual adjustment or simple sensor feedback mechanisms, which are often incapable of handling complex environmental changes or dynamic scenes. It is easy to cause projection deviation due to environmental factors such as light changes or projection angle fine-tuning, affecting the final visual effect and the audience's viewing experience. In other words, the accuracy of existing technologies is low. Summary of the invention

[0004] The present application provides a projector image positioning method, device, equipment and storage medium, which are used to improve the image positioning accuracy of the projector.

[0005] In a first aspect, the present application provides a projector image positioning method, which includes: projecting a preset reference image onto a target projection surface through a projector, and collecting a first projection image; constructing a relationship graph positioning network for the first projection image, and performing image offset and rotation angle correction on the projector based on the relationship graph positioning network to obtain a first image offset and a first rotation angle; performing real-time image projection on the target projection surface based on the first image offset and the first rotation angle through the projector, and collecting a second projection image; performing deep feature matching and deviation analysis on the second projection image through a deep convolutional neural network model to obtain a second image offset and a second rotation angle.

[0006] In combination with the first aspect, in a first implementation method of the first aspect of the present application, the relationship graph positioning network of the first projection image is constructed, and the image offset and rotation angle correction of the projector are performed based on the relationship graph positioning network to obtain the first image offset and the first rotation angle, including: performing image preprocessing on the first projection image to obtain a standard projection image; performing image feature point detection on the standard projection image through a scale-invariant feature transformation algorithm to obtain multiple projection feature points, and calculating a first feature descriptor for each projection feature point; obtaining multiple reference feature points in the reference image, and creating a second feature descriptor for each reference feature point; constructing a relationship graph positioning network between the multiple projection feature points and the multiple reference feature points based on the first feature descriptor and the second feature descriptor; calculating the similarity transformation matrix of the projector based on the relationship graph positioning network, and performing image offset and rotation angle correction on the similarity transformation matrix to obtain the first image offset and the first rotation angle.

[0007] In combination with the first aspect, in a second implementation of the first aspect of the present application, the relationship graph positioning network between the multiple projected feature points and the multiple reference feature points is constructed based on the first feature descriptor and the second feature descriptor, including: performing Euclidean distance calculation on the first feature descriptor of each projected feature point to obtain a first Euclidean distance value of each first feature descriptor, and setting the first feature point weight of each projected feature point according to the first Euclidean distance value; performing Euclidean distance calculation on the second feature descriptor of each reference feature point to obtain a second Euclidean distance value of each second feature descriptor, and setting each reference feature point according to the second Euclidean distance value. A second feature point weight of a feature point; creating a corresponding plurality of first nodes according to the plurality of projection feature points, and creating a corresponding plurality of second nodes according to the plurality of reference feature points; calculating a third Euclidean distance value between the first feature descriptor and the second feature descriptor, and creating a plurality of edges according to the third Euclidean distance value; generating a first positioning network according to the plurality of first nodes, the plurality of second nodes and the plurality of edges; optimizing the propagation node features of the first positioning network through a graph convolutional network to obtain a second positioning network; optimizing the importance of the relationship between nodes of the second positioning network through a graph attention network to obtain a relationship graph positioning network.

[0008] In combination with the first aspect, in a third implementation method of the first aspect of the present application, the similarity transformation matrix of the projector is calculated based on the relationship graph positioning network, and the similarity transformation matrix is ​​corrected for image offset and rotation angle to obtain a first image offset and a first rotation angle, including: extracting key node position information from the relationship graph positioning network to obtain a key node position information set; performing rotation and translation rigid transformation on the key node position information set to obtain the similarity transformation matrix of the projector; performing singular value decomposition on the similarity transformation matrix to obtain a rotation matrix and a translation vector; performing image offset correction on the translation vector to obtain a first image offset; performing inverse trigonometric function calculation and rotation angle correction on the rotation matrix to obtain a first rotation angle.

[0009] In combination with the first aspect, in a fourth implementation method of the first aspect of the present application, the deep feature matching and deviation analysis of the second projection image through a deep convolutional neural network model to obtain a second image offset and a second rotation angle includes: performing feature point recognition on the second projection image to obtain a plurality of real-time feature points; inputting the plurality of real-time feature points and the plurality of reference feature points into a deep convolutional neural network model for deep feature matching to obtain feature matching results; constructing a target deviation matrix of the second projection image according to the feature matching results, and performing deviation analysis on the target deviation matrix to obtain a second image offset and a second rotation angle.

[0010] In combination with the first aspect, in a fifth implementation of the first aspect of the present application, the inputting the multiple real-time feature points and the multiple benchmark feature points into a deep convolutional neural network model for deep feature matching to obtain a feature matching result includes: inputting the multiple real-time feature points and the multiple benchmark feature points into a deep convolutional neural network model, the deep convolutional neural network model including: an input layer, an encoder layer, a bottleneck layer, a decoder layer and an output layer; performing feature point encoding on the multiple real-time feature points and the multiple benchmark feature points respectively through the input layer to obtain a real-time feature input vector and a benchmark feature input vector; and encoding the real-time feature input vector and the benchmark feature input vector respectively through the convolution layer in the encoder layer. The invention relates to a method for extracting local features of a feature input vector to obtain a real-time local feature vector and a benchmark local feature vector; performing feature compression on the real-time local feature vector and the benchmark local feature vector respectively through the fully connected network in the bottleneck layer to obtain a real-time compressed feature vector and a benchmark compressed feature vector; performing deconvolution operation on the real-time compressed feature vector and the benchmark compressed feature vector respectively through the transposed convolution layer in the decoder layer to obtain a real-time feature target vector and a benchmark feature target vector; performing feature vector similarity calculation on the real-time feature target vector and the benchmark feature target vector through the output layer to obtain a vector similarity value, and outputting a corresponding feature matching result according to the vector similarity value.

[0011] In combination with the first aspect, in a sixth implementation of the first aspect of the present application, the projector image positioning method also includes: performing projection feedback and dynamic projection tracking on the projector according to the second image offset and the second rotation angle to obtain a dynamic tracking image set; adjusting parameters of the deep convolutional neural network model according to the dynamic tracking image set to obtain an initial model hyperparameter range; optimizing the model hyperparameter values ​​of the initial model hyperparameter range through a particle swarm algorithm to obtain target model hyperparameters; and performing parameter gradient updating on the deep convolutional neural network model according to the target model hyperparameters to obtain a projection image dynamic positioning model.

[0012] In the second aspect, the present application provides a projector image positioning device, which includes: a first projection module, used to project a preset reference image onto a target projection surface through a projector, and collect a first projection image; a correction module, used to construct a relationship graph positioning network for the first projection image, and perform image offset and rotation angle correction on the projector based on the relationship graph positioning network to obtain a first image offset and a first rotation angle; a second projection module, used to perform real-time image projection on the target projection surface based on the first image offset and the first rotation angle through the projector, and collect a second projection image; an analysis module, used to perform deep feature matching and deviation analysis on the second projection image through a deep convolutional neural network model to obtain a second image offset and a second rotation angle.

[0013] The third aspect of the present application provides a projector image positioning device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the projector image positioning device executes the above-mentioned projector image positioning method.

[0014] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned projector image positioning method.

[0015] In the technical solution provided by the present application, the positioning accuracy of the projected image is effectively improved by using a relationship graph positioning network and a deep convolutional neural network. By accurately matching image feature points and using image processing algorithms to calculate image offsets and rotation angles, the precise alignment of the projected content on the target projection surface is ensured, greatly improving the image quality and viewing experience. It is possible to quickly perform deviation analysis and correction based on real-time captured image data. By analyzing the offset and rotation angle of the first projection image and the second projection image in real time, the projector settings can be adjusted immediately to ensure that the image always maintains the best display effect. The need for manual operation is reduced. Through the automated image feature matching and deviation correction process, not only the efficiency of the operation is improved, but also the risk of errors caused by improper human operation is reduced. Through the combined use of the relationship graph positioning network and the deep convolutional neural network, it can adapt to various different environments and changing conditions. By implementing dynamic feedback and tracking, the parameter settings can be continuously adjusted and optimized according to real-time data. This dynamic feedback mechanism ensures that the system always maintains a high-performance state during long-term operation, responds to various emergencies in a timely manner, and ensures the correctness and effectiveness of the projected content, thereby improving the image positioning accuracy of the projector. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0017] Figure 1 A schematic diagram of an embodiment of a projector image positioning method in an embodiment of the present application;

[0018] Figure 2 Schematic diagram of an embodiment of a projector image positioning device in an embodiment of the present application. DETAILED DESCRIPTION

[0019] The present application embodiment provides a projector image positioning method, device, equipment and storage medium. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described here can be implemented in an order other than the content illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0020] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , an embodiment of the projector image positioning method in the embodiment of the present application includes:

[0021] Step S101, projecting a preset reference image onto a target projection surface through a projector, and collecting a first projection image;

[0022] It is understandable that the execution subject of the present application may be a projector image positioning device, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0023] Specifically, a reference image suitable for image positioning is designed, which should contain various geometric figures and markers that can be automatically recognized, such as cross lines, dots, and color blocks of specific shapes. The reference image is loaded into the projector, and the resolution and brightness of the projector are ensured to be suitable for the selected projection surface, such as a white wall or a special projection screen, to ensure that the image is clear and visible with minimal distortion. The position and focal length of the projector are adjusted to ensure that the reference image can completely and evenly cover the entire target projection surface, while avoiding image distortion caused by the projection angle. After the image is projected, a high-resolution camera is fixed on the opposite side or in an appropriate position to ensure that the image of the entire projection surface can be captured and the first projection image is collected. The camera parameter settings such as exposure time, ISO, and focus should be adjusted in advance to adapt to the ambient light conditions and projection brightness to ensure that the collected image is of high quality and rich in details. After collecting the first projection image, the image processing software is used to check the clarity and accuracy of the image to confirm whether the markers and figures in the image are clear and recognizable.

[0024] Step S102: constructing a relationship graph positioning network for the first projection image, and performing image offset and rotation angle correction on the projector based on the relationship graph positioning network to obtain a first image offset and a first rotation angle;

[0025] Specifically, the first projection image is subjected to image preprocessing, including denoising, contrast enhancement and normalization, etc., to improve the image quality and obtain a standard projection image. The standard projection image is subjected to image feature point detection by a scale-invariant feature transformation algorithm to obtain a plurality of projection feature points, and a first feature descriptor of each projection feature point is calculated. These descriptors can reflect the local image characteristics of the feature points and remain invariant to rotation, scale transformation and brightness change, so as to be suitable for feature matching under different environmental conditions. At the same time, a plurality of reference feature points are extracted from the stored reference image, and a second feature descriptor is created for each reference feature point, and these descriptors also have a high degree of discrimination and stability. A relationship graph positioning network between a plurality of projection feature points and a plurality of reference feature points is constructed based on the first feature descriptor and the second feature descriptor, and the feature points in the projection image are accurately matched with the feature points in the reference image by a graph matching algorithm. The similarity transformation matrix of the projector is calculated based on the relationship graph positioning network, and the matrix describes the optimal geometric transformation from the projection image to the reference image, including translation, rotation and scaling. By applying the similarity transformation matrix, the offset and rotation angle of the image output of the projector are corrected in real time, and finally the accurate values ​​of the first image offset and the first rotation angle are obtained.

[0026] The first feature descriptor of each feature point detected in the projected image is subjected to Euclidean distance calculation to evaluate the similarity or difference between each feature point. The first Euclidean distance value obtained by calculation reflects the distance between each feature point and other feature points. Based on the distance value, a corresponding first feature point weight is set for each projected feature point. The setting of the weight depends on the relative distance between the feature points. The smaller the distance, the larger the weight, indicating that the feature point has a strong correlation with other feature points, which is helpful for subsequent feature matching and network construction. Similarly, the same processing is performed for each reference feature point in the reference image, and the Euclidean distance between the second feature descriptor of each reference feature point is calculated to obtain the second Euclidean distance value, and the second feature point weight of each reference feature point is set accordingly. The weight is evaluated and determined based on the distance value to ensure that the weight can effectively reflect the spatial relationship and importance between the feature points. Multiple corresponding first nodes are created based on multiple projected feature points, and multiple corresponding second nodes are created based on multiple reference feature points. These nodes represent the feature points in their respective images. The third Euclidean distance value between the first feature descriptor and the second feature descriptor is calculated. This distance value is used to measure the similarity or difference between the feature points of the projected image and the reference image. According to these distance values, multiple edges are created to connect the corresponding first nodes and second nodes. The weight of the edge is determined by the third Euclidean distance. The smaller the distance, the higher the similarity between the two nodes. A first positioning network is generated based on multiple first nodes, multiple second nodes, and multiple edges. This network reflects the preliminary matching relationship between the projection feature points and the reference feature points. The first positioning network is optimized by propagating node features through a graph convolutional network. By learning the features of each node in the network and their mutual relationships, the node feature expression is improved to obtain a more refined second positioning network. Finally, the importance of the relationship between nodes in the second positioning network is optimized through a graph attention network to obtain a relationship graph positioning network. The graph attention network can assign weights to the relationship between each node and its neighboring nodes, highlighting the feature relationships that are most critical to image positioning, and achieve more accurate projector image positioning in practical applications.

[0027] The key node position information of the relationship graph positioning network is extracted to obtain a set of key node position information. These key nodes represent the most important feature points in the projection image and the reference image. Their position information set is crucial for calculating the spatial relationship between the images. The position information of the key nodes is processed by rotation and translation rigid transformation. By finding the optimal transformation parameters to minimize the Euclidean distance between the feature points of the projection image and the feature points of the reference image, a similarity transformation matrix that can describe the transformation from the projection image to the reference image is determined. The matrix contains rotation and translation information, as well as a slight scaling factor. Its purpose is to align the feature points on the projection image to the corresponding feature points of the reference image as accurately as possible. The similarity transformation matrix is ​​subjected to singular value decomposition to obtain the rotation matrix and translation vector. Singular value decomposition is an effective matrix decomposition technique that extracts pure rotation and translation information from the similarity transformation matrix, where the rotation matrix describes the angle by which the image needs to be rotated, and the translation vector directly indicates the offset of the image in each direction. The image offset correction is performed on the translation vector. By applying the translation vector to the projected image, its position in space is adjusted, the offset difference with the reference image is reduced or eliminated, and finally the first image offset is obtained. This offset directly reflects the consistency of the position of the corrected projected image and the reference image. At the same time, the rotation matrix is ​​calculated by inverse trigonometric functions, and the specific rotation angle is extracted from the rotation matrix by calculating the arc cosine value of the matrix element. The extracted rotation angle is used to perform rotation correction on the projected image to ensure that its direction is completely consistent with the reference image, and the first rotation angle is obtained.

[0028] Step S103: Projecting a real-time image on the target projection surface based on the first image offset and the first rotation angle through a projector, and collecting a second projection image;

[0029] Specifically, first. Apply the first image offset and the first rotation angle to the settings of the projector. Adjust the position and projection angle of the projector to ensure that the projected image can be accurately aligned with the predetermined position of the target projection surface. The adjustment usually requires fine operation of the physical or electronic steering mechanism of the projector through the software control interface so that the projected image can be rotated to the correct angle and moved to the correct position in space. Start the real-time image projection process, and the reference image is re-projected onto the target projection surface according to the corrected offset and rotation angle. During the image projection process, a high-resolution camera is used synchronously to capture the second projection image. The camera settings such as exposure, focus and ISO are adjusted according to the lighting conditions and projection quality on site to ensure that the captured image can clearly reflect the actual projection effect of the projected image, including the accuracy of its position and angle. After the second projection image is captured, the image is used for further analysis and processing to verify the correction effect of the image offset and rotation angle, thereby providing necessary feedback information for subsequent correction steps.

[0030] Step S104: performing deep feature matching and deviation analysis on the second projection image through a deep convolutional neural network model to obtain a second image offset and a second rotation angle.

[0031] Specifically, feature points of the second projection image are identified. By using image processing techniques, such as edge detection, corner detection or feature detection algorithms, multiple real-time feature points are identified from the projected image, and these feature points reflect the key visual elements and structural information of the image. Multiple real-time feature points and multiple reference feature points are input into a deep convolutional neural network model for deep feature matching. The model uses its learned weights and biases to calculate the similarity between features, and outputs feature matching results, reflecting the correspondence and matching degree between the feature points of the second projection image and the reference feature points. A target deviation matrix of the second projection image is constructed based on the feature matching results. The deviation matrix describes the degree of spatial deviation of the second projection image relative to the reference image based on the spatial correspondence and difference in the matching results, including information such as offset and rotation. By performing deviation analysis on the target deviation matrix, the offset and rotation angle of the second image are calculated. These two parameters are the key to correcting the projection error. The offset analysis examines the average offset of the image feature points in the horizontal and vertical directions, while the rotation angle analysis is determined based on the angular change of the feature points relative to the center of the image.

[0032] Multiple real-time feature points and multiple benchmark feature points are input into a deep convolutional neural network model for deep feature matching. The model consists of multiple layers, including an input layer, an encoder layer, a bottleneck layer, a decoder layer, and an output layer, which work together to achieve efficient feature analysis and matching. The input layer encodes the feature points of multiple real-time feature points and multiple benchmark feature points to obtain real-time feature input vectors and benchmark feature input vectors, and converts the feature points in the image into numerical data that can be processed by the network, ensuring that the key information of the feature points is effectively captured. The convolution layer in the encoder layer extracts local features from the real-time feature input vector and the benchmark feature input vector, respectively, and extracts the local information of the image feature points by applying multiple convolution kernels to generate real-time local feature vectors and benchmark local feature vectors. These vectors capture subtle changes in feature points and important visual information. The fully connected network in the bottleneck layer compresses the real-time local feature vector and the benchmark local feature vector, respectively, to reduce the data dimension while retaining key information, and obtain real-time compressed feature vectors and benchmark compressed feature vectors. Compression helps reduce the complexity of subsequent calculations while highlighting the most critical elements in the feature vector. Through the transposed convolution layer in the decoder layer, the real-time compressed feature vector and the benchmark compressed feature vector are deconvolved to reconstruct the feature vectors, and more detailed and expanded real-time feature target vectors and benchmark feature target vectors are obtained. These vectors are restored to a state close to the original input feature points through the operation of the transposed convolution layer, and the deep feature information learned by the network through training is incorporated to make the feature vectors more suitable for accurate matching. Finally, the real-time feature target vector and the benchmark feature target vector are deep feature matched through the output layer. Through optimized matching algorithms, such as distance metrics or similarity assessments, the network can accurately identify which feature points are matched and generate the final feature matching results.

[0033] According to the second image offset and the second rotation angle, the projector is projected with feedback and dynamically tracked. The projector adjusts its projection position and angle according to the real-time analysis results to compensate for the detected offset and rotation, ensuring that the projected content is consistent with the expected target position. Dynamic adjustment generates a set of dynamic tracking images, which reflect the response and adjustment effect of the projector to dynamic changes in the actual environment. The parameters of the deep convolutional neural network model are initially adjusted according to the dynamic tracking image set. The adjustment process involves analyzing the deviations and model responses in the image set to determine the initial hyperparameter range of the model parameters. These hyperparameters include learning rate, number of convolutional layers, convolution kernel size, etc., which directly affect the efficiency and accuracy of model training. In order to optimize the performance of the model, the initial model hyperparameter range is optimized by particle swarm algorithm. Particle swarm algorithm is an optimization technology based on swarm intelligence. It finds the best hyperparameter value by simulating the process of a group of particles searching for the optimal solution in the solution space. In this process, each particle represents a set of possible hyperparameter settings. The particles continuously adjust their flight direction and position based on their own and the group's experience, and eventually tend to the optimal solution. Finally, the parameter gradient of the deep convolutional neural network model is updated according to the target model hyperparameters. By adjusting the weights and biases in the network to adapt to the new hyperparameter settings, a fine-tuned projection image dynamic positioning model is obtained. The updated model can more accurately identify and match image features, improve the accuracy and response speed of dynamic projection tracking, and thus achieve smoother and more accurate image positioning and projection effects in actual projection applications.

[0034] In the embodiment of the present application, the positioning accuracy of the projected image is effectively improved by using a relationship graph positioning network and a deep convolutional neural network. By accurately matching image feature points and using image processing algorithms to calculate image offsets and rotation angles, the precise alignment of the projected content on the target projection surface is ensured, greatly improving the image quality and viewing experience. It is possible to quickly perform deviation analysis and correction based on real-time captured image data. By analyzing the offset and rotation angle of the first projection image and the second projection image in real time, the projector settings can be adjusted immediately to ensure that the image always maintains the best display effect. The need for manual operation is reduced. Through the automated image feature matching and deviation correction process, not only the efficiency of the operation is improved, but also the risk of errors caused by improper human operation is reduced. Through the combined use of the relationship graph positioning network and the deep convolutional neural network, it can adapt to various different environments and changing conditions. By implementing dynamic feedback and tracking, the parameter settings can be continuously adjusted and optimized according to real-time data. This dynamic feedback mechanism ensures that the system always maintains a high-performance state during long-term operation, responds to various emergencies in a timely manner, and ensures the correctness and effectiveness of the projected content, thereby improving the image positioning accuracy of the projector.

[0035] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0036] (1) performing image preprocessing on the first projection image to obtain a standard projection image;

[0037] (2) detecting image feature points of the standard projection image by using a scale-invariant feature transformation algorithm to obtain a plurality of projection feature points, and calculating a first feature descriptor of each projection feature point;

[0038] (3) obtaining a plurality of reference feature points in the reference image and creating a second feature descriptor for each reference feature point;

[0039] (4) constructing a relationship graph positioning network between a plurality of projected feature points and a plurality of reference feature points based on the first feature descriptor and the second feature descriptor;

[0040] (5) Calculate the similarity transformation matrix of the projector based on the relationship graph positioning network, and perform image offset and rotation angle correction on the similarity transformation matrix to obtain a first image offset and a first rotation angle.

[0041] Specifically, the first projection image projected from the projector is preprocessed. This includes adjusting the contrast and brightness of the image, performing sharpening processing, and applying filters to remove noise to ensure that the clarity of the image and the recognizability of the features are maximized. After preprocessing, a standard projection image is obtained. Feature point detection is performed on the standard projection image using a scale-invariant feature transformation algorithm. The algorithm can identify and extract key points in the image and generate a unique descriptor for each feature point. These descriptors remain unchanged for image scaling and are highly stable for rotation and illumination changes. For example, a sign projected on a white wall may affect the clarity of the projected image due to changes in ambient light, but after being processed by a scale-invariant feature transformation algorithm, the edges and other significant features of the sign can be accurately detected from the image, and each detected projection feature point will be assigned a first feature descriptor containing its position, scale, and direction information. At the same time, the corresponding feature points are obtained from a stored reference image. The reference image is usually a pre-selected, representative image that provides a reference standard. The reference image may be an ideal or standard version image of the same sign. From the reference image, the scale-invariant feature transformation algorithm is also used to extract key feature points, and a second feature descriptor is created for each reference feature point. These descriptors record the key attributes of each feature point in the reference image. A relationship graph localization network is constructed between multiple projected feature points and multiple reference feature points based on the first feature descriptors and the second feature descriptors. The network connects all projected feature points and reference feature points, and determines the similarity between them by comparing their respective descriptors. The distances between descriptors are calculated, and the similarity between two descriptors is measured using methods such as Euclidean distance, and the edges in the network are constructed based on these distances. After the network is constructed, a graph matching algorithm is used to find the best matching method to identify the corresponding relationship between the projected image and the reference image. Finally, based on the relationship graph localization network, a similarity transformation matrix is ​​calculated. The matrix is ​​determined by the graph matching results, which describes how to adjust the projected image to a state that is as consistent as possible with the reference image through translation, rotation and scaling operations. After obtaining the similarity transformation matrix, it is parsed to separate the image offset and rotation angle, which are directly used to adjust the projector settings to ensure the correct position and orientation of the projected content.

[0042] In a specific embodiment, the process of executing the step of constructing a relationship graph positioning network between a plurality of projection feature points and a plurality of reference feature points based on the first feature descriptor and the second feature descriptor may specifically include the following steps:

[0043] (1) performing Euclidean distance calculation on the first feature descriptor of each projected feature point to obtain a first Euclidean distance value of each first feature descriptor, and setting a first feature point weight of each projected feature point according to the first Euclidean distance value;

[0044] (2) performing Euclidean distance calculation on the second feature descriptor of each reference feature point to obtain a second Euclidean distance value of each second feature descriptor, and setting a second feature point weight of each reference feature point according to the second Euclidean distance value;

[0045] (3) creating a corresponding plurality of first nodes according to the plurality of projection feature points, and creating a corresponding plurality of second nodes according to the plurality of reference feature points;

[0046] (4) calculating a third Euclidean distance value between the first feature descriptor and the second feature descriptor, and creating a plurality of edges according to the third Euclidean distance value;

[0047] (5) generating a first positioning network according to the plurality of first nodes, the plurality of second nodes, and the plurality of edges;

[0048] (6) Optimizing the propagation node features of the first positioning network through a graph convolutional network to obtain a second positioning network;

[0049] (7) The importance of the relationships between nodes in the second positioning network is optimized through the graph attention network to obtain a relational graph positioning network.

[0050] Specifically, the Euclidean distance calculation is performed on the first feature descriptor of each projected feature point, and the distance between each descriptor and all other descriptors is calculated to obtain information about the relative position of the feature point in the feature space. The first Euclidean distance value provides a quantitative way to measure the similarity or difference between each feature point and other points. The first feature point weight of each projected feature point is set according to the first Euclidean distance value. The shorter the distance, the greater the weight, which indicates a higher similarity. Similarly, for the second feature descriptor of each reference feature point, a similar Euclidean distance calculation is performed, and the weight is set according to the obtained second Euclidean distance value to ensure that the significance of the key feature points in the reference image can be appropriately emphasized. Create corresponding multiple first nodes according to multiple projected feature points, and create corresponding multiple second nodes according to multiple reference feature points. The first node represents the feature point in the projected image, and the second node represents the feature point in the reference image. A first positioning network is formed according to the first node and the second node. The edge of the network is created according to the third Euclidean distance value between the first feature descriptor and the second feature descriptor. This distance value measures the similarity between the projected feature points and the reference feature points, helping the system build a preliminary network that can reflect the relative position and feature similarity between the two images. The first localization network is optimized by propagating node features through a graph convolutional network. The graph convolutional network optimizes the feature representation of the entire network by propagating information between nodes, so that the network can more accurately reflect the actual similarity between nodes. The feature representation of nodes is enhanced by incorporating the relationship between nodes and their feature information into the network to obtain the second localization network. Finally, the importance of the relationship between nodes in the second localization network is optimized through a graph attention network to obtain a relational graph localization network. The graph attention network can highlight the node relationships that are most important in feature matching and image localization by assigning different attention weights to each node and their connections. The optimization based on the attention mechanism enhances the network's response to key features and enables the network to identify the most influential feature points and their connections among many features.

[0051] In a specific embodiment, the execution step calculates the similarity transformation matrix of the projector based on the relationship graph positioning network, and performs image offset and rotation angle correction on the similarity transformation matrix to obtain the first image offset and the first rotation angle. The process may specifically include the following steps:

[0052] (1) Extract key node location information from the relationship graph positioning network to obtain a set of key node location information;

[0053] (2) Performing rotation and translation rigid transformation on the key node position information set to obtain the similarity transformation matrix of the projector;

[0054] (3) Perform singular value decomposition on the similarity transformation matrix to obtain the rotation matrix and translation vector;

[0055] (4) performing image offset correction on the translation vector to obtain a first image offset;

[0056] (5) Perform inverse trigonometric function calculation and rotation angle correction on the rotation matrix to obtain a first rotation angle.

[0057] Specifically, the key node position information of the relationship graph positioning network is extracted to obtain a key node position information set. Those nodes marked as key nodes in the network are usually those feature points that play a decisive role in image matching, and their position information set is helpful for subsequent image transformation calculations. The key node position information set is subjected to rotation and translation rigid transformation to obtain the similarity transformation matrix of the projector. The similarity transformation matrix includes three basic elements: rotation, translation, and scaling. The key node position information extracted from the relationship graph positioning network is subjected to rotation and translation rigid transformation to find a transformation method so that the feature points on the projected image coincide with the corresponding feature points on the reference image as much as possible. The similarity transformation matrix is ​​subjected to singular value decomposition to obtain a rotation matrix and a translation vector. Singular value decomposition is a mathematical method commonly used in data dimensionality reduction and matrix simplification, which can decompose the rotation matrix and the translation vector, thereby extracting the rotation and translation information in the image transformation separately. The translation vector is subjected to image offset correction to obtain a first image offset. By applying the translation vector, the position of the projected image is directly adjusted to make it closer to or completely match the position of the reference image. The application of the translation vector is usually achieved by modifying the settings of the projector or adjusting the output coordinates of the image through software. Finally, the rotation matrix is ​​subjected to inverse trigonometric function calculation and rotation angle correction to obtain the first rotation angle. The extraction of the rotation angle usually depends on the properties of the rotation matrix. The rotation angle is obtained by calculating the arc cosine value of a specific element in the matrix.

[0058] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0059] (1) performing feature point recognition on the second projection image to obtain a plurality of real-time feature points;

[0060] (2) Inputting multiple real-time feature points and multiple reference feature points into a deep convolutional neural network model for deep feature matching to obtain feature matching results;

[0061] (3) A target deviation matrix of the second projection image is constructed according to the feature matching results, and a deviation analysis is performed on the target deviation matrix to obtain a second image offset and a second rotation angle.

[0062] Specifically, feature points of the second projection image are identified to obtain multiple real-time feature points. Advanced image processing techniques, such as scale-invariant feature transformation, accelerated robust feature and other algorithms, are used to detect stable and reliable feature points from the image. Multiple real-time feature points and multiple reference feature points are input into a deep convolutional neural network model for deep feature matching to obtain feature matching results. The model is trained to perform deep feature matching tasks. Through multiple layers of convolutional layers, activation layers and pooling layers, the network can extract and learn deep information of image features. In this process, the descriptor of each real-time feature point is compared with the descriptor of the corresponding feature point in the reference image, and the network evaluates the similarity and difference between the features by comparing the descriptors. A target deviation matrix of the second projection image is constructed based on the feature matching results. The matrix describes the spatial position deviation of the feature points in the second projection image relative to the corresponding feature points in the reference image based on the feature matching results. Constructing the target deviation matrix is ​​a computationally intensive process that requires accurate calculation and comparison of the spatial differences between each pair of matching feature points, including their position coordinate differences and angle deviations, which helps to identify possible projection errors and distortions in the image. Finally, the target deviation matrix is ​​subjected to deviation analysis to obtain the second image offset and the second rotation angle. The data in the deviation matrix is ​​statistically evaluated, for example, by calculating the mean and standard deviation of the deviation values ​​to quantify the overall shift and rotation requirements of the image. The calculation results are used to adjust the settings of the projection device to ensure that the projected image can be displayed in the correct position and at the correct angle, thereby maximizing the visual effect of the reference image.

[0063] In a specific embodiment, the execution step inputs multiple real-time feature points and multiple reference feature points into a deep convolutional neural network model for deep feature matching, and the process of obtaining feature matching results may specifically include the following steps:

[0064] (1) inputting a plurality of real-time feature points and a plurality of reference feature points into a deep convolutional neural network model, wherein the deep convolutional neural network model comprises: an input layer, an encoder layer, a bottleneck layer, a decoder layer, and an output layer;

[0065] (2) encoding the multiple real-time feature points and the multiple reference feature points through the input layer to obtain a real-time feature input vector and a reference feature input vector;

[0066] (3) performing local feature extraction on the real-time feature input vector and the reference feature input vector respectively through a convolution layer in the encoder layer to obtain a real-time local feature vector and a reference local feature vector;

[0067] (4) performing feature compression on the real-time local feature vector and the reference local feature vector through the fully connected network in the bottleneck layer to obtain a real-time compressed feature vector and a reference compressed feature vector;

[0068] (5) performing deconvolution operations on the real-time compressed feature vector and the reference compressed feature vector through a transposed convolution layer in the decoder layer to obtain a real-time feature target vector and a reference feature target vector;

[0069] (6) Perform deep feature matching on the real-time feature target vector and the reference feature target vector through the output layer to obtain the feature matching result.

[0070] Specifically, multiple real-time feature points and multiple benchmark feature points are input into a deep convolutional neural network model. Multiple feature points captured in real time from the scene and benchmark feature points extracted from historical drawings are received through the input layer. Each feature point is encoded into a vector form and converted into a format suitable for neural network processing. The original information of the real-time feature points and the benchmark feature points is converted into a real-time feature input vector and a benchmark feature input vector, and the conversion includes information such as position, size, shape, etc., and may also incorporate data about the pixel intensity and color distribution around the feature points. Local feature extraction is performed on the real-time feature input vector and the benchmark feature input vector respectively through the convolution layer in the encoder layer, and this layer is composed of multiple convolution layers. In the convolution layer, the network extracts local features from the input feature vector through multiple filters. These filters are designed to recognize and respond to various patterns and structures in the image data, such as edges, corners or more complex texture information. Through the convolution operation, the real-time feature vector and the benchmark feature vector are converted into a real-time local feature vector and a benchmark local feature vector respectively. Through the fully connected network in the bottleneck layer, the real-time local feature vector and the benchmark local feature vector are feature compressed to obtain a real-time compressed feature vector and a benchmark compressed feature vector. The bottleneck layer is composed of a fully connected network. The rich but high-dimensional features extracted by the previous convolutional layer are compressed through the fully connected layer to extract the most informative feature representation, while reducing the dimension of the data to reduce the computational burden and prevent overfitting. Through the transposed convolution layer in the decoder layer, the real-time compressed feature vector and the baseline compressed feature vector are deconvolved to obtain the real-time feature target vector and the baseline feature target vector. The compressed feature vector is decoded back to a higher-dimensional feature space to reconstruct the complete expression of the feature. Finally, the real-time feature target vector and the baseline feature target vector are deep-feature matched through the output layer to obtain the feature matching result. The feature matching result is determined by analyzing the similarity between the decoded feature vectors. The weights and bias parameters of the neural network are used to determine which feature points have the best match.

[0071] In a specific embodiment, the projector image positioning method further includes the following steps:

[0072] (1) performing projection feedback and dynamic projection tracking on the projector according to the second image offset and the second rotation angle to obtain a dynamic tracking image set;

[0073] (2) adjusting the parameters of the deep convolutional neural network model according to the dynamic tracking image set to obtain the initial model hyperparameter range;

[0074] (3) Using the particle swarm algorithm to optimize the model hyperparameter values ​​within the initial model hyperparameter range, the target model hyperparameters are obtained;

[0075] (4) The parameter gradient of the deep convolutional neural network model is updated according to the target model hyperparameters to obtain the projection image dynamic positioning model.

[0076] Specifically, according to the second image offset and the second rotation angle, the projector is projected with feedback and dynamically tracked to obtain a set of dynamic tracking images. The projector dynamically adjusts the projection parameters (such as projection angle, focal length, etc.) according to the offset and rotation angle data collected in real time to ensure that the projected image is optimally aligned with the predetermined projection target (such as a display surface or a specific area). As the projector parameters are adjusted, the system continuously collects new projection images to obtain a set of dynamic tracking images. According to the set of dynamic tracking images, the parameters of the deep convolutional neural network model are adjusted to obtain the initial model hyperparameter range. The change trend in the image set, such as image stability, clarity, and alignment accuracy, is analyzed to adjust the hyperparameters of the network model, such as learning rate, number of layers, filter size, etc. After determining the initial model hyperparameter range, the particle swarm optimization algorithm is used to find the optimal values ​​of these hyperparameters. The particle swarm optimization algorithm is an optimization tool that simulates social behavior and is performed by simulating a group of particles (solutions) to find the optimal solution in the solution space. Each particle adjusts its position in the solution space based on its own experience and the experience of the group to find the point that minimizes or maximizes specific performance indicators (such as error rate, response time, etc.) to obtain the target model hyperparameters. Finally, the deep convolutional neural network model is updated with parameter gradients according to the target model hyperparameters to obtain the projection image dynamic positioning model. By adjusting the network weights and biases using techniques such as back propagation, the network model is ensured to be able to adapt to the actual situation in the dynamic tracking image collection to the greatest extent possible.

[0077] The above describes the projector image positioning method in the embodiment of the present application. The following describes the projector image positioning device in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the projector image positioning device includes:

[0078] The first projection module 201 is used to project a preset reference image onto a target projection surface through a projector and collect a first projection image;

[0079] A correction module 202 is used to construct a relationship graph positioning network of the first projection image, and perform image offset and rotation angle correction on the projector based on the relationship graph positioning network to obtain a first image offset and a first rotation angle;

[0080] The second projection module 203 is used to project a real-time image on the target projection surface based on the first image offset and the first rotation angle through a projector, and to collect a second projection image;

[0081] The analysis module 204 is used to perform deep feature matching and deviation analysis on the second projection image through a deep convolutional neural network model to obtain a second image offset and a second rotation angle.

[0082] Through the collaboration of the above components, the positioning accuracy of the projected image is effectively improved through the relationship graph positioning network and the deep convolutional neural network. By accurately matching the image feature points and using the image processing algorithm to calculate the image offset and rotation angle, the precise alignment of the projected content on the target projection surface is ensured, greatly improving the image quality and viewing experience. It can quickly perform deviation analysis and correction based on the real-time captured image data. By analyzing the offset and rotation angle of the first and second projected images in real time, the projector settings can be adjusted immediately to ensure that the image always maintains the best display effect. The need for manual operation is reduced. Through the automated image feature matching and deviation correction process, not only the efficiency of operation is improved, but also the risk of errors caused by improper human operation is reduced. Through the combined use of the relationship graph positioning network and the deep convolutional neural network, it can adapt to various different environments and changing conditions. By implementing dynamic feedback and tracking, the parameter settings can be continuously adjusted and optimized according to real-time data. This dynamic feedback mechanism ensures that the system always maintains high performance during long-term operation, responds to various emergencies in a timely manner, ensures the correctness and effectiveness of the projected content, and thus improves the image positioning accuracy of the projector.

[0083] The present application also provides a projector image positioning device, which includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes the steps of the projector image positioning method in the above-mentioned embodiments.

[0084] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the projector image positioning method.

[0085] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0086] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0087] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A projector image positioning method, characterized in that: The projector image positioning method comprises: Projecting a preset reference image onto a target projection surface through a projector, and collecting a first projection image; Construct a relationship graph positioning network for the first projection image, and perform image offset and rotation angle correction on the projector based on the relationship graph positioning network to obtain a first image offset and a first rotation angle; specifically comprising: performing image preprocessing on the first projection image to obtain a standard projection image; performing image feature point detection on the standard projection image through a scale-invariant feature transformation algorithm to obtain multiple projection feature points, and calculating a first feature descriptor for each projection feature point; obtaining multiple reference feature points in the reference image, and creating a second feature descriptor for each reference feature point; performing Euclidean distance calculation on the first feature descriptor of each projection feature point to obtain a first Euclidean distance value of each first feature descriptor, and setting a first feature point weight of each projection feature point according to the first Euclidean distance value; performing Euclidean distance calculation on the second feature descriptor of each reference feature point to obtain a second Euclidean distance value of each second feature descriptor, and setting a second feature point weight of each reference feature point according to the second Euclidean distance value; creating a pair according to the multiple projection feature points The method comprises the steps of: calculating a plurality of first nodes corresponding to the first feature descriptor and creating a plurality of corresponding second nodes according to the plurality of reference feature points; calculating a third Euclidean distance value between the first feature descriptor and the second feature descriptor, and creating a plurality of edges according to the third Euclidean distance value; generating a first positioning network according to the plurality of first nodes, the plurality of second nodes and the plurality of edges; optimizing the propagation node features of the first positioning network through a graph convolution network to obtain a second positioning network; optimizing the importance of the relationship between nodes of the second positioning network through a graph attention network to obtain a relationship graph positioning network; extracting key node position information of the relationship graph positioning network to obtain a key node position information set; performing rotation and translation rigid transformation on the key node position information set to obtain a similarity transformation matrix of the projector; performing singular value decomposition on the similarity transformation matrix to obtain a rotation matrix and a translation vector; performing image offset correction on the translation vector to obtain a first image offset; performing inverse trigonometric function calculation and rotation angle correction on the rotation matrix to obtain a first rotation angle; Using the projector, projecting a real-time image onto the target projection surface based on the first image offset and the first rotation angle, and collecting a second projection image; The second projection image is subjected to deep feature matching and deviation analysis through a deep convolutional neural network model to obtain a second image offset and a second rotation angle.

2. The projector image positioning method according to claim 1, characterized in that: The performing deep feature matching and deviation analysis on the second projection image by using a deep convolutional neural network model to obtain a second image offset and a second rotation angle includes: Performing feature point recognition on the second projection image to obtain a plurality of real-time feature points; Inputting the multiple real-time feature points and the multiple reference feature points into a deep convolutional neural network model for deep feature matching to obtain a feature matching result; A target deviation matrix of the second projection image is constructed according to the feature matching result, and a deviation analysis is performed on the target deviation matrix to obtain a second image offset and a second rotation angle.

3. The projector image positioning method according to claim 2, characterized in that: The step of inputting the multiple real-time feature points and the multiple reference feature points into a deep convolutional neural network model for deep feature matching to obtain a feature matching result includes: Inputting the multiple real-time feature points and the multiple reference feature points into a deep convolutional neural network model, wherein the deep convolutional neural network model includes: an input layer, an encoder layer, a bottleneck layer, a decoder layer, and an output layer; Performing feature point encoding on the multiple real-time feature points and the multiple reference feature points respectively through the input layer to obtain a real-time feature input vector and a reference feature input vector; Performing local feature extraction on the real-time feature input vector and the reference feature input vector respectively through a convolution layer in the encoder layer to obtain a real-time local feature vector and a reference local feature vector; Through the fully connected network in the bottleneck layer, feature compression is performed on the real-time local feature vector and the reference local feature vector respectively to obtain a real-time compressed feature vector and a reference compressed feature vector; Through the transposed convolution layer in the decoder layer, deconvolution operations are performed on the real-time compressed feature vector and the reference compressed feature vector to obtain a real-time feature target vector and a reference feature target vector; The real-time feature target vector and the reference feature target vector are deep-feature matched through the output layer to obtain a feature matching result.

4. The projector image positioning method according to claim 2, characterized in that: The projector image positioning method also includes: According to the second image offset and the second rotation angle, performing projection feedback and dynamic projection tracking on the projector to obtain a dynamic tracking image set; Adjusting parameters of the deep convolutional neural network model according to the dynamic tracking image set to obtain an initial model hyperparameter range; Using a particle swarm algorithm to optimize the model hyperparameter value within the initial model hyperparameter range, and obtain the target model hyperparameter; The parameter gradient of the deep convolutional neural network model is updated according to the target model hyperparameters to obtain a projection image dynamic positioning model.

5. A projector image positioning device, characterized in that: The projector image positioning device comprises: A first projection module, used for projecting a preset reference image onto a target projection surface through a projector, and collecting a first projection image; A correction module is used to construct a relationship graph positioning network for the first projection image, and based on the relationship graph positioning network, perform image offset and rotation angle correction on the projector to obtain a first image offset and a first rotation angle; specifically comprising: performing image preprocessing on the first projection image to obtain a standard projection image; performing image feature point detection on the standard projection image through a scale-invariant feature transformation algorithm to obtain multiple projection feature points, and calculating a first feature descriptor for each projection feature point; acquiring multiple reference feature points in the reference image, and creating a second feature descriptor for each reference feature point; performing Euclidean distance calculation on the first feature descriptor of each projection feature point to obtain a first Euclidean distance value of each first feature descriptor, and setting a first feature point weight of each projection feature point according to the first Euclidean distance value; performing Euclidean distance calculation on the second feature descriptor of each reference feature point to obtain a second Euclidean distance value of each second feature descriptor, and setting a second feature point weight of each reference feature point according to the second Euclidean distance value; Create a corresponding plurality of first nodes, and create a corresponding plurality of second nodes according to the plurality of reference feature points; calculate the third Euclidean distance value between the first feature descriptor and the second feature descriptor, and create a plurality of edges according to the third Euclidean distance value; generate a first positioning network according to the plurality of first nodes, the plurality of second nodes and the plurality of edges; optimize the propagation node features of the first positioning network through a graph convolution network to obtain a second positioning network; optimize the importance of the relationship between nodes of the second positioning network through a graph attention network to obtain a relationship graph positioning network; extract key node position information from the relationship graph positioning network to obtain a key node position information set; perform rotation and translation rigid transformation on the key node position information set to obtain a similarity transformation matrix of the projector; perform singular value decomposition on the similarity transformation matrix to obtain a rotation matrix and a translation vector; perform image offset correction on the translation vector to obtain a first image offset; perform inverse trigonometric function calculation and rotation angle correction on the rotation matrix to obtain a first rotation angle; A second projection module, configured to project a real-time image onto the target projection surface through the projector based on the first image offset and the first rotation angle, and to collect a second projection image; An analysis module is used to perform deep feature matching and deviation analysis on the second projection image through a deep convolutional neural network model to obtain a second image offset and a second rotation angle.

6. A projector image positioning device, characterized in that: The projector image positioning device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory to enable the projector image positioning device to perform the projector image positioning method according to any one of claims 1 to 4.

7. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the projector image positioning method according to any one of claims 1 to 4 is implemented.

Images