Mobile terminal three-dimensional depth imaging method based on structured infrared light
By projecting a pseudo-random infrared speckle array onto a mobile terminal and using a deep learning model to predict the deformation state of missing points, the problem of data voids in 3D imaging under highly reflective or highly absorbent materials is solved, and efficient and complete 3D depth map generation is achieved.
Patent Information
- Application Number
- CN202610098497.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-26
- Publication Date
- 2026-02-24
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
When existing structured light schemes image on highly reflective or highly absorbent material surfaces, the infrared light spot is easily specularly reflected or absorbed, causing the imaging system to fail to capture effective signals, forming data holes, destroying the integrity of the 3D model, and traditional methods perform poorly at large holes or edges, increasing computational latency.
A mobile terminal 3D depth imaging method based on structured infrared light is adopted. A pseudo-randomly distributed infrared speckle array is projected through an infrared laser projection component. Deformed speckle images are acquired by an infrared camera. A feature matching algorithm is used to identify missing points, and a deep learning model is used to predict the deformation state of the missing points. Finally, a complete 3D depth map is generated by combining triangulation.
It effectively fills in missing areas, generates highly complete and accurate 3D models, improves the recognition success rate and reliability in complex scenes, and reduces computational latency.
Smart Images
Figure CN121564280A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of scanning imaging technology, specifically to a three-dimensional depth imaging method for mobile terminals based on structured infrared light. Background Technology
[0002] With the rapid development of mobile internet and artificial intelligence technologies, the functions of mobile terminals such as smartphones and tablets have far exceeded the traditional scope of communication and entertainment, placing higher demands on environmental perception and 3D interaction capabilities. High-precision 3D depth imaging technology has become the core foundation for realizing cutting-edge applications such as facial recognition, augmented reality (AR), 3D modeling, and gesture interaction.
[0003] Existing structured light solutions heavily rely on cameras successfully capturing each projected speckle. However, when the target surface has highly reflective or highly absorbent materials, the infrared speckles projected onto it will undergo specular reflection or be largely absorbed, causing the imaging system to fail to capture effective speckle signals. This results in data holes in the depth map, which can damage the integrity of the 3D model and lead to recognition failure. Traditional hole filling or missing data processing often uses post-processing methods based on geometric interpolation. These methods perform poorly when the holes are large or located at the edges of objects, and they usually need to be performed after the complete depth map is generated, increasing computational latency. Summary of the Invention
[0004] (a) Technical problems to be solved
[0005] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a three-dimensional depth imaging method for mobile terminals based on structured infrared light, which can effectively solve the problems of the existing technology.
[0006] (II) Technical Solution
[0007] To achieve the above objectives, the present invention provides the following technical solution: This invention discloses a three-dimensional depth imaging method for mobile terminals based on structured infrared light, comprising the following steps: Step 1: Using an infrared laser projection component mounted on a mobile terminal, an infrared speckle array with a preset spatial distribution structure is projected onto the target scene to form a structured light pattern on the surface of the target scene; the infrared laser projection component includes a VCSEL laser array and diffractive optical elements; Step 2: Receive and acquire infrared speckle images that have been distorted after being modulated on the surface of the target scene using the image acquisition device set on the mobile terminal, and compare the captured distorted speckle images with the standard reference speckle pattern pre-stored in the mobile terminal. Step 3: Using a feature matching algorithm, for each speckle feature point successfully captured in the deformed speckle image, find its corresponding original reference point in the standard reference speckle pattern. For feature points that fail to capture matching speckles in the deformed speckle image, their corresponding original reference points in the standard reference speckle pattern are marked as missing points. Step 4: For each marked missing point, extract its position data in the preset spatial distribution structure and the deformation state data of the captured speckle corresponding to all adjacent original reference points within the preset neighborhood. Input the position data and deformation state data into a speckle deformation state prediction model pre-trained by a deep learning algorithm. The model outputs the predicted deformation state data of the speckle corresponding to the missing point. The specific inputs of the model include the absolute coordinates of the missing point and the normalized offset vector relative to the neighborhood points. The output is the predicted displacement vector of the speckle at that point. The training data of the speckle deformation state prediction model comes from a large number of speckle images and their corresponding real depth data collected by pattern projection on various standard geometries and complex surfaces at different distances and angles. Step 5: Combine the speckle deformation state predicted in Step 4 with the speckle deformation state data that has been successfully matched in Step 3 to form a deformed speckle dataset. Calculate the corresponding depth value for each speckle based on its displacement relative to its original reference position. Step 6: Based on the calculated depth values of all speckles, generate 3D point cloud data and reconstruct the surface to finally construct a 3D depth image of the target scene.
[0008] Furthermore, in step 1, the preset spatial distribution structure is a pseudo-random distribution, and the number of spots in the infrared speckle array is between 10,000 and 40,000.
[0009] Furthermore, the image acquisition device in step 2 is an infrared camera whose spectral response band matches the emission band of the infrared laser projection component, and the standard reference speckle pattern is a speckle image acquired on a preset calibration plane.
[0010] Furthermore, the feature matching algorithm in step 3 simultaneously performs missing point identification during the matching process. When matching the captured deformed speckle image with the standard reference speckle pattern point by point, a missing point is marked for the reference point that meets any of the following conditions: No candidate speckles with a similarity exceeding a preset threshold were found in the deformed speckle image; The depth continuity check between the candidate speckle found in the deformed speckle image and other matched speckles in the neighborhood failed.
[0011] Furthermore, after completing the initial matching, the feature matching algorithm in step 3 performs neighborhood consistency verification: for each successfully matched speckle, the consistency of its displacement vector with other matching points in the preset neighborhood is calculated; if the displacement vector of a certain matching point deviates from the vector direction of more than a preset percentage of matching points in the neighborhood by more than a preset threshold or the difference in magnitude exceeds a preset threshold, then the point is determined to be a mismatched point, its matching result is revoked and it is remarked as a missing point.
[0012] Furthermore, the construction process of the speckle deformation state prediction model in step 4 is as follows: The infrared speckle array is projected onto various standard geometric objects and complex surface targets by the infrared laser projection component. The relative distance and angle between the target object and the mobile terminal are controlled synchronously. The three-dimensional coordinates of the measured area on the surface of the target object are recorded synchronously by the laser tracker as the true value. At the same time, the corresponding deformed speckle image is acquired by the image acquisition device. The acquired deformed speckle images are denoised and normalized. The displacement vector of speckle in each image is extracted as label data. Based on the standard reference speckle pattern, the speckle that was not successfully acquired is marked as missing and the displacement state of its neighboring speckle is recorded. A prediction model based on a graph neural network is constructed, which abstracts speckles into graph nodes. The node features include their spatial coordinates and displacement vectors. The adjacency relationship is determined based on the topological connection of adjacent points in a preset spatial distribution structure. The network parameters are optimized through backpropagation until convergence, with the coordinates of the missing point and the displacement vectors of the adjacent points as inputs and the actual displacement vector of the missing point as the output target.
[0013] Furthermore, the graph neural network includes a node embedding layer, a neighborhood aggregation layer, and a regression output layer, wherein: the node embedding layer encodes node coordinates into feature vectors through a fully connected layer; the neighborhood aggregation layer updates node features through an iterative message passing mechanism; and the regression output layer performs displacement vector regression on the node features corresponding to the missing points, with an output dimension of 2, corresponding to the displacement on the x and y axes of the image plane.
[0014] Furthermore, the depth value calculation process in step 5 is based on triangulation. According to the baseline distance determined by the fixed spatial positional relationship between the infrared laser projection component and the image acquisition device, and the optical focal length of the image acquisition device, the depth distance of the target surface position corresponding to each speckle feature point is obtained by calculating the parallax displacement generated by each speckle feature point relative to its standard reference position on the imaging plane.
[0015] Furthermore, the three-dimensional graphics construction process in step 6 is as follows: based on the calculated depth values of all speckles, combined with the intrinsic parameter matrix of the image acquisition device and the corresponding two-dimensional pixel coordinates, three-dimensional spatial point cloud data is generated through coordinate transformation calculation. Based on the depth value continuity constraint and spatial topological relationship, a continuous three-dimensional surface mesh model is generated from the generated three-dimensional point cloud data through a surface reconstruction algorithm.
[0016] Furthermore, the execution process of the surface reconstruction algorithm is as follows: establish a spatial index structure for the point cloud data, and perform a fast search for neighboring points based on a kd-tree or octree; calculate the normal vector direction for each sampling point, and fit the local tangent plane through principal component analysis; and generate a topologically continuous triangular mesh surface using the moving cube algorithm or Poisson reconstruction.
[0017] (III) Beneficial Effects
[0018] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects: 1. By intelligently predicting the speckle deformation state of missing regions through deep learning models, rather than simply relying on geometric interpolation for post-processing, it can predict the most likely shape of the missing parts based on the overall geometric continuity of the object's surface, thereby filling in the gaps at the front-end data level and ultimately outputting highly complete depth maps and 3D models, which greatly improves the success rate and reliability of functions such as face recognition and AR interaction in complex scenarios.
[0019] 2. By embedding the prediction process into the core workflow of depth calculation and seamlessly connecting it with the triangulation step, the system can quickly process both captured and predicted data simultaneously and output complete high-precision depth information. The system no longer needs to infinitely increase power in pursuit of capturing every perfect spot. Even if some spots are submerged by noise, the model can make a certain degree of compensation based on the captured effective information and prior knowledge. Attached Figure Description
[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.
[0021] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation
[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.
[0023] The present invention will be further described below with reference to embodiments.
[0024] The structured infrared light-based three-dimensional depth imaging method for mobile terminals in this embodiment, such as... Figure 1 As shown, it includes the following steps: Step 1: Using an infrared laser projection component mounted on a mobile terminal, an infrared speckle array with a preset spatial distribution structure is projected onto the target scene. The preset spatial distribution structure is a pseudo-random distribution. The number of speckles in the infrared speckle array is between 10,000 and 40,000, so as to form a structured light pattern on the surface of the target scene. The infrared laser projection component includes a VCSEL laser array and diffractive optical elements.
[0025] Step 2: Using the image acquisition device set on the mobile terminal, receive and acquire infrared speckle images that have been distorted after being modulated on the surface of the target scene, and compare the captured distorted speckle images with a standard reference speckle pattern pre-stored in the mobile terminal; the image acquisition device is an infrared camera whose spectral response band matches the emission band of the infrared laser projection component, and the standard reference speckle pattern is a speckle image acquired on a preset calibration plane.
[0026] Step 3: Using a feature matching algorithm, for each speckle feature point successfully captured in the deformed speckle image, its corresponding original reference point in the standard reference speckle pattern is found. For feature points that fail to capture matching speckles in the deformed speckle image, their corresponding original reference points in the standard reference speckle pattern are marked as missing points. The feature matching algorithm performs missing point identification operations simultaneously during the matching process. When matching the captured deformed speckle image with the standard reference speckle pattern point by point, reference points that meet any of the following conditions are marked as missing: No candidate speckles with a similarity exceeding a preset threshold were found in the deformed speckle image; The depth continuity check between the candidate speckle found in the deformed speckle image and other matched speckles in the neighborhood failed. After completing the initial matching, the feature matching algorithm performs neighborhood consistency verification: for each successfully matched speckle, it calculates the consistency of its displacement vector with other matching points in the preset neighborhood; if the displacement vector of a matching point deviates from the vector direction of more than a preset percentage of matching points in the neighborhood by more than a preset threshold or the difference in magnitude exceeds a preset threshold, then the point is determined to be a mismatched point, its matching result is revoked and it is remarked as a missing point.
[0027] Step 4: For each marked missing point, extract its position data in the preset spatial distribution structure and the deformation state data of the captured speckle corresponding to all adjacent original reference points within the preset neighborhood. Input the position data and deformation state data into a speckle deformation state prediction model that has been pre-trained by a deep learning algorithm. The model outputs the predicted deformation state data of the speckle corresponding to the missing point. The specific inputs of the model include the absolute coordinates of the missing point and the normalized offset vector relative to the neighborhood points. The output is the predicted displacement vector of the speckle at that point. The training data of the speckle deformation state prediction model comes from a large number of speckle images and their corresponding real depth data collected by pattern projection on various standard geometries and complex surfaces at different distances and angles.
[0028] Step 5: Merge the speckle deformation state predicted in Step 4 with the speckle deformation state data successfully matched in Step 3 to form a deformed speckle dataset. Calculate the corresponding depth value for each speckle based on its displacement relative to its original reference position. The depth value calculation is based on triangulation. According to the baseline distance determined by the fixed spatial relationship between the infrared laser projection component and the image acquisition device, and the optical focal length of the image acquisition device, the depth distance of the target surface position corresponding to each speckle feature point is obtained by calculating the parallax displacement generated on the imaging plane relative to its standard reference position.
[0029] Step 6: Based on the calculated depth values of all speckles, generate 3D point cloud data and reconstruct the surface to finally construct a 3D depth image of the target scene.
[0030] Compared with existing technologies, by combining traditional structured light feature matching with a deep learning-based missing point prediction model, the problems of matching ambiguity and a large number of missing feature points caused by short baselines and high speckle density in miniaturized VCSEL speckle projectors for mobile terminals are solved.
[0031] This method not only effectively eliminates mismatched points by using algorithms such as neighborhood consistency verification, ensuring the reliability of the initial matching, but more importantly, it intelligently predicts the deformation state of missing points through deep learning models, greatly compensating for data loss caused by factors such as occlusion, noise, or weak texture. Thus, it can still generate complete and high-precision depth maps under high-density speckle arrays, effectively improving the integrity of 3D depth imaging and the final reconstruction effect while maintaining the compact hardware structure of mobile devices.
[0032] At other levels, in this embodiment, the construction process of the speckle deformation state prediction model is as follows: The infrared speckle array is projected onto various standard geometric objects and complex surface targets by the infrared laser projection component. The relative distance and angle between the target object and the mobile terminal are controlled synchronously. The three-dimensional coordinates of the measured area on the surface of the target object are recorded synchronously by the laser tracker as the true value. At the same time, the corresponding deformed speckle image is acquired by the image acquisition device. The acquired deformed speckle images are denoised and normalized. The displacement vector of speckle in each image is extracted as label data. Based on the standard reference speckle pattern, the speckle that was not successfully acquired is marked as missing and the displacement state of its neighboring speckle is recorded. A prediction model based on a graph neural network is constructed, which abstracts speckles into graph nodes. The node features include their spatial coordinates and displacement vectors. The adjacency relationship is determined based on the topological connection of adjacent points in a preset spatial distribution structure. The network parameters are optimized through backpropagation until convergence, with the coordinates of the missing point and the displacement vectors of the adjacent points as inputs and the actual displacement vector of the missing point as the output target. The graph neural network includes a node embedding layer, a neighborhood aggregation layer, and a regression output layer. The node embedding layer encodes node coordinates into feature vectors through a fully connected layer. The neighborhood aggregation layer updates node features through an iterative message passing mechanism. The regression output layer performs displacement vector regression on the node features corresponding to missing points, with an output dimension of 2, corresponding to the displacement on the x and y axes of the image plane.
[0033] Compared with existing technologies, graph neural networks are used to model the inherent spatial topological relationships between speckles. Traditional methods usually process each speckle independently or use ordinary fully connected networks, which cannot effectively utilize the high correlation of the deformation of neighboring points in high-density speckle arrays.
[0034] This method treats the speckle array as a graph structure and performs neighborhood aggregation through a message passing mechanism. This enables the model to learn and infer the deformation state of missing points from the real displacements of surrounding measured points, greatly improving the physical rationality and accuracy of the prediction. This GNN-based modeling approach is more in line with the spatial continuity of the structured light deformation field, thereby significantly improving the accuracy and generalization ability of deep learning models for missing value prediction.
[0035] In this embodiment, a three-dimensional graphics construction process is provided, which is as follows: based on the depth values of all the speckles obtained by calculation, combined with the intrinsic parameter matrix of the image acquisition device and the corresponding two-dimensional pixel coordinates, three-dimensional spatial point cloud data is generated by coordinate transformation calculation; and based on the depth value continuity constraint and spatial topological relationship, a continuous three-dimensional surface mesh model is generated by surface reconstruction algorithm for the generated three-dimensional point cloud data. The execution process of the surface reconstruction algorithm is as follows: Establish a spatial index structure for point cloud data, and perform fast neighborhood point search based on kd-tree or octree; Calculate the normal vector direction for each sampling point, and fit the local tangent plane using principal component analysis; Based on the principle of normal vector consistency, the moving cube algorithm or Poisson reconstruction is used to generate a topologically continuous triangular mesh surface; The reconstructed mesh is repaired for holes and smoothed by filtering to eliminate surface distortion caused by missing depth data.
[0036] In summary, this invention utilizes a red light projection component mounted on a mobile device to construct a projection pattern using a red light spot array with a preset range structure, and projects the pattern onto a designated target. An image acquisition device receives images of the deformed spots on the target surface after projection, analyzes the state of the deformed spots on the target object's surface, compares the captured deformed spot images with a pre-stored standard reference pattern, and uses feature matching to find the original point corresponding to each captured spot in the reference pattern. Original points that were not captured as image spots are marked. Based on the position data of the original point in the preset projection pattern and the deformation state of the corresponding spots of adjacent original points within the preset range of the original point, the data is input into a deformation prediction model that has been pre-constructed and trained using a deep learning algorithm. The model predicts and outputs the deformation state of the spots corresponding to the marked original points. The predicted spots are combined with all captured spots to calculate the displacement of each spot relative to its reference position. Through triangulation, the depth value corresponding to a spot is calculated based on the displacement of a spot. Based on the depth values of each spot, a three-dimensional graphic is constructed. By projecting high-density pseudo-random infrared speckle, combined with feature matching and deep learning techniques, the problem of missing and mismatched feature points caused by occlusion and noise is effectively solved. The introduction of graph neural network to predict the displacement of missing points improves the integrity of point cloud and the continuity of depth data. It is suitable for integration into mobile terminals and can still achieve stable and reliable 3D reconstruction in complex scenarios, taking into account both real-time requirements and high detail restoration capabilities.
[0037] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A three-dimensional depth imaging method for mobile terminals based on structured infrared light, characterized in that, Includes the following steps: Step 1: Project an infrared speckle array with a preset spatial distribution structure onto the target scene using an infrared laser projection component mounted on the mobile terminal, so as to form a structured light pattern on the surface of the target scene; Step 2: Receive and acquire infrared speckle images that have been distorted after being modulated on the surface of the target scene using the image acquisition device set on the mobile terminal, and compare the captured distorted speckle images with the standard reference speckle pattern pre-stored in the mobile terminal. Step 3: Using a feature matching algorithm, for each speckle feature point successfully captured in the deformed speckle image, find its corresponding original reference point in the standard reference speckle pattern. For feature points that fail to capture matching speckles in the deformed speckle image, their corresponding original reference points in the standard reference speckle pattern are marked as missing points. Step 4: For each marked missing point, extract its position data in the preset spatial distribution structure and the deformation state data of the captured speckle corresponding to all adjacent original reference points within the preset neighborhood. Input the position data and deformation state data into a speckle deformation state prediction model that has been trained in advance by a deep learning algorithm. The model outputs the predicted deformation state data of the speckle corresponding to the missing point. Step 5: Combine the speckle deformation state predicted in Step 4 with the speckle deformation state data that has been successfully matched in Step 3 to form a deformed speckle dataset. Calculate the corresponding depth value for each speckle based on its displacement relative to its original reference position. Step 6: Based on the calculated depth values of all speckles, generate 3D point cloud data and reconstruct the surface to finally construct a 3D depth image of the target scene.
2. The method for three-dimensional depth imaging of a mobile terminal based on structured infrared light according to claim 1, characterized in that, In step 1, the preset spatial distribution structure is a pseudo-random distribution, and the number of spots in the infrared speckle array is between 10,000 and 40,000.
3. The method for three-dimensional depth imaging of a mobile terminal based on structured infrared light according to claim 1, characterized in that, The image acquisition device in step 2 is an infrared camera whose spectral response band matches the emission band of the infrared laser projection component. The standard reference speckle pattern is a speckle image acquired on a preset calibration plane.
4. The method for three-dimensional depth imaging of a mobile terminal based on structured infrared light according to claim 1, characterized in that, In step 3, the feature matching algorithm simultaneously performs missing point identification during the matching process. When matching the captured deformed speckle image with the standard reference speckle pattern point by point, a missing point is marked for reference points that meet any of the following conditions: No candidate speckles with a similarity exceeding a preset threshold were found in the deformed speckle image; The depth continuity check between the candidate speckle found in the deformed speckle image and other matched speckles in the neighborhood failed.
5. The method for three-dimensional depth imaging of a mobile terminal based on structured infrared light according to claim 1, characterized in that, After completing the initial matching, the feature matching algorithm in step 3 performs neighborhood consistency verification: for each successfully matched speckle, calculate the consistency of its displacement vector with other matching points in the preset neighborhood; if the displacement vector of a matching point deviates from the vector direction of more than a preset percentage of matching points in the neighborhood by more than a preset threshold or the difference in magnitude exceeds a preset threshold, then the point is determined to be a mismatched point, its matching result is revoked and it is remarked as a missing point.
6. The method for three-dimensional depth imaging of a mobile terminal based on structured infrared light according to claim 1, characterized in that, The construction process of the speckle deformation state prediction model in step 4 is as follows: The infrared speckle array is projected onto various standard geometric objects and complex surface targets by the infrared laser projection component. The relative distance and angle between the target object and the mobile terminal are controlled synchronously. The three-dimensional coordinates of the measured area on the surface of the target object are recorded synchronously by the laser tracker as the true value. At the same time, the corresponding deformed speckle image is acquired by the image acquisition device. The acquired deformed speckle images are denoised and normalized. The displacement vector of speckle in each image is extracted as label data. Based on the standard reference speckle pattern, the speckle that was not successfully acquired is marked as missing and the displacement state of its neighboring speckle is recorded. A prediction model based on a graph neural network is constructed, which abstracts speckles into graph nodes. The node features include their spatial coordinates and displacement vectors. The adjacency relationship is determined based on the topological connection of adjacent points in a preset spatial distribution structure. The network parameters are optimized through backpropagation until convergence, with the coordinates of the missing point and the displacement vectors of the adjacent points as inputs and the actual displacement vector of the missing point as the output target.
7. The method for three-dimensional depth imaging of a mobile terminal based on structured infrared light according to claim 6, characterized in that, The graph neural network includes a node embedding layer, a neighborhood aggregation layer, and a regression output layer. The node embedding layer encodes node coordinates into feature vectors through a fully connected layer. The neighborhood aggregation layer updates node features through an iterative message passing mechanism. The regression output layer performs displacement vector regression on the node features corresponding to the missing points, with an output dimension of 2, corresponding to the displacement on the x and y axes of the image plane.
8. The method for three-dimensional depth imaging of a mobile terminal based on structured infrared light according to claim 1, characterized in that, The depth calculation process in step 5 is based on triangulation. According to the baseline distance determined by the fixed spatial position relationship between the infrared laser projection component and the image acquisition device, and the optical focal length of the image acquisition device, the depth distance of the target surface position corresponding to the speckle feature point is obtained by calculating the parallax displacement of each speckle feature point relative to its standard reference position on the imaging plane.
9. The method for three-dimensional depth imaging of a mobile terminal based on structured infrared light according to claim 1, characterized in that, The three-dimensional graphics construction process in step 6 is as follows: based on the depth values of all the speckles obtained by calculation, combined with the intrinsic parameter matrix of the image acquisition device and the corresponding two-dimensional pixel coordinates, three-dimensional spatial point cloud data is generated by coordinate transformation calculation. Based on the depth value continuity constraint and spatial topological relationship, a continuous three-dimensional surface mesh model is generated by surface reconstruction algorithm for the generated three-dimensional point cloud data.
10. The method for three-dimensional depth imaging of a mobile terminal based on structured infrared light according to claim 9, characterized in that, The execution process of the surface reconstruction algorithm is as follows: Establish a spatial index structure for point cloud data, and perform fast neighborhood point search based on kd-tree or octree; Calculate the normal vector direction for each sampling point, and fit the local tangent plane using principal component analysis; The topologically continuous triangular mesh surface is generated using the moving cube algorithm or Poisson reconstruction.