124 results about "Projection mapping" patented technology

The invention, which relates to the technical field of image processing, discloses a city component automatic three-dimensional reconstruction method based on a sub-meter high-resolution remote sensing image, comprising the following steps: acquiring a sub-meter high-resolution remote sensing image of a target area, and establishing an initial ground-image projection mapping relation based on a rational polynomial coefficient model; acquiring a ground control point, correcting the mapping relation, and outputting a registration image; performing multi-view geometric matching on the registered image to generate dense earth surface point cloud, and inputting the dense earth surface point cloud into a semantic neural radiation field network to generate dense point cloud; performing cross-modal feature alignment and fusion on the dense point cloud and the vector data, and outputting a fused multi-source dense point cloud; according to the method, through fine correction of the multi-source remote sensing image, the problems of large registration error, more point cloud sparse noise and semantic deficiency are solved, and high-precision and automatic three-dimensional reconstruction of urban components is realized.

The invention discloses a power transmission line channel hidden danger distance measurement method based on point cloud image fusion. The method comprises the steps of data joint calibration, target recognition and modeling of a power transmission line channel scene, target segmentation and positioning of the power transmission line channel scene and judgment of the distance between a hidden danger target and a wire. The method has the advantages that the point cloud data and the image data are fused, the space coordinate conversion relation between the point cloud data and the image data is calibrated and solved, the point cloud data are projected to the image view angle to generate a new point cloud projection mapping graph, training recognition is conducted through the image data and the point cloud data, the recognition accuracy of the hidden danger target is improved, the SAM model serves as a basic model, and the recognition efficiency is improved. The point cloud projection mapping graph and the hidden danger target in the image data are used as samples, training is carried out, a hidden danger target segmentation model is generated, accurate contour information of the hidden danger target is extracted, and effective early warning information is provided for the safety of a power transmission line channel.

The invention discloses a naked-eye 3D display optimization method based on real-time eyeball tracking, and particularly relates to the technical field of naked-eye 3D display, and the method comprises the steps: capturing eyeball movement data in real time through a visual angle tracking sensor, and collecting illumination information in combination with an ambient light sensor; depth perception parameters are calculated based on pupil diameter variation and frequency, and a time sequence prediction type dynamic compensation coefficient is generated in combination with eyeball movement acceleration; acquiring an initial fixation point coordinate by using an improved spherical projection mapping model, and performing compensation coefficient correction to obtain a real-time coordinate; and finally, according to the real-time coordinates, dynamically adjusting the refractive index distribution of the nanostructure layer, the rotation angle of the polarizer, the focal length of the optical lens and other optical modulation parameters. The real-time distance is calculated through the binocular parallax algorithm, the depth mapping value is generated by combining the focal length of the camera and the baseline distance, the method can adapt to the illumination change and the user view angle, and the 3D display effect is optimized.

The invention discloses a multi-modal large model based on dynamic decision hybrid expert mapping, which comprises a visual coding group comprising a plurality of visual coding structures with different functions, a visual fusion and compression module, a dynamic decision hybrid expert mapping module and a large language model, the dynamic decision hybrid expert mapping module comprises a hybrid expert module and a shared expert module which respectively process the multi-scale image features to obtain high-discrimination features and comprehensive features, and a decision expert module which is used for carrying out importance estimation on the multi-scale image features; the decision connection module and the self-adaptive projection mapping layer are used for connecting the high-discrimination features and the comprehensive features according to importance estimated values, and the dynamic decision hybrid expert mapping module intelligently distributes parameters to different expert networks and dynamically adjusts weight configuration of experts. And the internal structure of the model can be automatically optimized according to different inputs, so that higher processing precision is achieved.

The invention relates to the technical field of bolt length detection data processing, in particular to a bolt length detection system, which is characterized in that a three-dimensional data acquisition module generates a high-density point cloud and extracts a thread curvature extreme point, and a dynamic model matching module screens an optimal lead angle parameter based on a Hausdorff distance; and constructing an axial expansion model carrying phase offset gradient features. The projection mapping correction module adopts a genetic algorithm to optimize a compensation coefficient matrix, and generates a dynamic weight correction parameter to suppress geometric distortion; and the phase error correction module combines Fourier spectrum decomposition and an exponential decay weight coefficient to construct a frequency domain transfer function to eliminate an axial accumulative error. According to the method, the problems of projection distortion and phase deviation caused by the lead angle are effectively solved through geometric-phase double-domain correction, dynamic parameter iteration and an error collaborative suppression mechanism, and the spatial positioning precision of the feature points at the tail end of the thread is improved.

The invention provides an intelligent adaptive projection lamp and a projection control method, and the method comprises the steps: collecting multi-source environment data which comprises at least one of environment illumination data, projection surface characteristic data and user operation data; performing preprocessing, feature extraction and feature fusion on the multi-source environment data to generate an environment feature vector; performing projection mapping on the environment feature vector through a parameter prediction model based on rate distortion optimization to generate projection parameters; performing image deblurring processing and adaptive k-space sampling reconstruction on the projection parameters to generate an optimized projection image; collecting projection effect feedback data for the optimized projection image; and performing online updating optimization on the parameter prediction model according to the projection effect feedback data. According to the invention, intelligent adaptive adjustment of the projection parameters is realized, and the projection quality is improved.

The invention provides an interactive point cloud instance segmentation method based on a three-dimensional Gaussian scattering field, and the method comprises the steps: receiving click prompt data of a user, carrying out the calculation of interaction points and weights based on the click prompt data, and obtaining Gaussian features containing interaction prompt information; screening and converting the Gaussian features in the interaction point neighborhood to obtain point cloud batch data; performing foreground and background prediction on the point cloud batch data by using a point cloud segmentation network to obtain a three-dimensional instance label; performing projection mapping on the three-dimensional instance label through a Gaussian grating to obtain a two-dimensional segmentation mask; and performing morphological post-processing on the two-dimensional segmentation mask to obtain a segmentation result. According to the method, the cross-view consistency and the boundary precision of the point cloud segmentation result of the three-dimensional Gaussian scattering field instance are effectively improved, and the segmentation time efficiency is accelerated.

The invention discloses an intelligent interactive electrochromic projection glass control method and system, and the method comprises the steps: determining an interaction intention of a user, and calling a corresponding target projection parameter; querying a target transmittance value corresponding to the updated target projection parameter according to a preset transmittance projection mapping table; the light transmittance of the electrochromic glass is controlled to reach the target light transmittance value, and actual display parameters of the projection picture are collected; when the deviation between the actual display parameter and the updated target projection parameter exceeds a preset range, generating an RGB compensation coefficient according to the current environment illumination intensity value and the target transmittance value; and sending the RGB compensation coefficient to the projection unit to adjust the RGB three-color light output intensity of the projection light source. According to the method and the device, the environment illumination, the user interaction requirement and the voice instruction are sensed in real time, the glass transmittance is adaptively adjusted, and the RGB light source output is dynamically compensated, so that the consistent display effect of the projection picture is improved, and the conference cooperation difficulty is reduced.

The invention provides an oil-water well oil pipe thread profile measurement method based on three-dimensional point cloud adaptive slicing. The method comprises the following steps: step 1, obtaining an oil pipe thread three-dimensional point cloud image; 2, normalizing the 3D point cloud data; 3, extracting a global feature recognition point cloud image, carrying out point cloud positioning, and segmenting point cloud data; step 4, extracting the labeled detection area for denoising processing; step 5, segmenting the 3d point cloud image into a plurality of square blocks; step 6, performing projection mapping processing on the plurality of point cloud blocks from front view, top view and side view angles to obtain a pseudo picture; and step 7, carrying out pseudo image matching and combining with a loss function to judge whether the image is qualified. According to the oil-water well oil pipe thread profile measurement method based on the three-dimensional point cloud self-adaptive slicing, operation reworking caused by oil pipe sealing leakage is reduced, the working efficiency is improved, the crude oil yield loss is reduced, and it is avoided that the liquid supply capacity of an oil well is reduced due to leakage.

The invention relates to an AR projection interaction treadmill virtual scene generation method and device, equipment and a medium. The method comprises the following steps: acquiring motion inertia data and running belt motion parameters of a user on a running machine; respectively carrying out motion state extraction and gait phase extraction according to the motion inertia data and the running belt motion parameters to obtain a current motion state vector and gait phase information of the user; inputting the motion state vector and the gait phase information into a pre-trained motion state prediction model to obtain a preset predicted motion state of a plurality of frames in the future; based on the predicted motion state, performing layered scene generation through a scene generator to obtain a corresponding virtual scene; and performing projection mapping transformation processing on the virtual scene, and obtaining real-time AR projection data in combination with low-delay compensation. By adopting the method, a highly immersive and interactive virtual training environment can be provided in an AR projection mode.

A system including a drone having a projector to project an image from a projection origin. The drone also has a navigation unit to determine location information for the drone. A processor coupled to the drone includes a memory. Execution of programming by the processor configures the system to obtain a projection surface architecture for a projection surface. The projection surface architecture includes reference points that correspond to physical locations on the projection surface. Each reference point is associated with relationship data with respect to an architecture origin. The system also receives location information for the drone, adapts the relationship data responsive to change in the location information, adjusts the image using the adapted relationship data, and projects the adjusted image onto the projection surface.

An attraction system may include a projection mapped tangible show element with a first rigid body including a first shell coupled to a first tracker, a second rigid body including a second shell coupled to a second tracker, and a connection coupling the first rigid body and the second rigid body to one another. The connection may enable movement of the first rigid body relative to the second rigid body. The attraction system may also include an image sensor to generate sensor data indicative of a first position of the first rigid body and a second position of the second rigid body and a projector to project first image data on the first rigid body and second image data on the second rigid body.

The invention belongs to the technical field of spherical screen image processing, and particularly relates to an image processing system and method for suspension type spherical screen panoramic vision, and the system obtains the attribute and position information of a spherical screen and a plurality of projection devices through a spherical screen modeling module, constructs a spherical screen-projection integrated three-dimensional model, and precisely divides projection areas to correspond to the devices; the image processing module collects image sequences of different angles of the panoramic vision scene; the image splicing module uses feature point detection, a YOLO algorithm and a convolutional neural network to identify an overlapping region and discriminate a dynamic region, and obtains a dynamic pixel splicing line through a dynamic planning search algorithm; the image correction module constructs a projection mapping function, corrects geometric and optical distortion of the image, and optimizes a splicing effect through a splicing correction function; and finally, the multi-view output module projects the corrected image sequence to a corresponding area of the dome screen by using a plurality of projection devices, so that real-time dynamic splicing and monitoring are realized, and projection synchronization and image quality are ensured.

The invention provides an assembled building block model generation method and device, equipment and a storage medium, and the method comprises the steps: obtaining a source picture and a to-be-mapped model rendering graph, and generating an initial mapping rendering graph according to the source picture and the to-be-mapped model rendering graph through a deep learning diffusion model; taking the initial chartlet rendering image as a target chartlet rendering image, and carrying out matting processing on the target chartlet rendering image to obtain a matting bitmap; and taking the matting bitmap as a projection map, and carrying out projection mapping on the projection map according to a preset block structure of an assembled building block model so as to realize automatic mapping of the assembled building block model and obtain the assembled building block model after mapping is completed. According to the method, the deep learning technology and the image processing technology are combined, the full-automatic generation process from user input to the final chartlet model is achieved, the efficiency and quality of building block chartlet design are remarkably improved, and the problem that time and labor are consumed in a traditional manual design method is solved.

The invention provides an insulator defect identification method based on cross-modal data fusion. The insulator defect identification method comprises the following steps: acquiring and preprocessing an insulator visible light image, a thermal infrared image and point cloud data which are synchronously acquired; projecting all points in the point cloud to obtain a projection image and a projection mapping matrix; registering the visible light image and the thermal infrared image with the projection image to obtain a corresponding affine transformation matrix, and constructing a shielding matrix for shielding the background according to the projection image; performing feature extraction on the pre-processed visible light image, thermal infrared image and point cloud data to respectively obtain a texture feature map, a temperature feature map and a geometric feature map; fusing the texture feature map, the temperature feature map and the geometric feature map according to the shielding matrix and the affine transformation matrix to obtain cross-modal fusion features; and the cross-modal fusion features are used to discriminate and position insulator defects. According to the method, the problem of leak detection of insulator defects in different scenes by a traditional single-mode method can be solved.

The invention relates to the technical field of model training learning, in particular to a training method of an intention recognition model for understanding user requirements in an intext software scene, which comprises the following steps of: traversing paths to determine levels and establish variable-dimension projection mapping, generating semantic orthogonal conflict samples, obtaining variable-dimension features by utilizing matrix transformation, and obtaining an intention recognition model for understanding user requirements in an intext software scene. And quantizing the brother node overlapping degree to apply orthogonal constraint, updating the weight by minimizing the positive sample distance and maximizing the conflict sample distance, and constructing an intext software intention recognition optimization model. According to the method, a mapping mechanism of a hierarchical position and a variable-dimension projection matrix is established through software architecture topology, a semantic orthogonal logic conflict sample is constructed, brother node feature branch orthogonal constraint is introduced, and feature representation of parallel architecture branches is forcibly separated in a vector space; the logic conflict sample spacing is maximized, semantic overlapping of similar function nodes is eliminated, and a decision boundary with high distinction degree is constructed to accurately analyze a complex nested instruction.

To provide a golf control system in an indoor golf course capable of enhancing a feeling of presence and a feeling of immersion with a simple configuration.SOLUTION: A golf control system 30 of an indoor golf course 10 includes: projection control part for executing projection mapping in an area that a player can visually recognize of an indoor space; a storage part for storing content 40 of the projection mapping; and a sound control part for discriminating or outputting a patterned sound. The content 40 includes a moving image associated with the sound.SELECTED DRAWING: Figure 1

The invention provides a speed measurement method based on dynamic zooming of an unmanned aerial vehicle and an inspection system. The method comprises the following steps: acquiring a video stream acquired by the unmanned aerial vehicle in real time through a camera for dynamically adjusting focal length parameters; inputting the video stream into the detection model frame by frame; outputting a target detection frame set of the corresponding current frame; performing cross-frame association on the detection frames of the adjacent frames through a ByteTrack algorithm, and distributing a unique instance identifier for the same target; acquiring real-time parameters of the unmanned aerial vehicle, establishing a real pavement coordinate system and a pixel imaging coordinate system, and performing coordinate system projection mapping; and calculating the real speed of the target. The method has the beneficial effects that the coordinate mapping error is greatly reduced through dynamic parameter fusion and unmanned aerial vehicle displacement compensation, and particularly, the improvement is obvious in a pitch angle or zoom scene; through introduction of central point position transformation, image center offset caused by digital zooming is dynamically calibrated, and pixel-physical coordinate mapping distortion caused by pure image cutting or interpolation amplification is effectively eliminated.

The invention discloses a laser radar real-time semantic segmentation method and system based on image and point cloud multi-view fusion knowledge distillation, and the method comprises the steps: obtaining laser radar point cloud data and camera RGB image data, building a projection mapping relation between point cloud and image pixels, and enabling the mapping relation to be used for searching a common effective projection point; laser radar point cloud data is converted into a voxel grid view and a distance plane view, a main and auxiliary double-branch architecture is constructed in combination with camera RGB image data, hierarchical compression features are fused through three-level mutual attention gating, and multi-modal multi-view fusion features are generated; respectively distilling the multi-modal multi-view fusion features to a main branch and an auxiliary branch by adopting a bidirectional knowledge distillation strategy, and synchronously optimizing the dual-branch feature extraction capability; the auxiliary branch is unidirectionally distilled to the main branch, so that the performance of the main branch is further improved; and migrating the semantic segmentation capability on the point cloud multi-view and image-point cloud multi-modal fusion features to a point cloud voxel network, thereby realizing real-time high-precision semantic segmentation of a radar point cloud single mode. According to the method, the precision and the real-time performance of point cloud semantic segmentation in a dynamic scene are remarkably improved.

The invention discloses a three-dimensional projection mapping method and device, and particularly relates to the technical field of three-dimensional projection mapping. Three-dimensional data of the surface of an object are acquired in real time through multiple cameras and a sensor, a geometric model and texture information of the surface of the object are dynamically updated in combination with a deep learning algorithm, and the angle, focal length and illumination intensity of a projector are adjusted by adopting a self-adaptive algorithm according to real-time geometric data, illumination conditions and audience position changes, so that the image quality is improved. In addition, the real-time sensor feeds back the audience position and the environment illumination change, the system can dynamically adjust the projection content, it is ensured that the virtual image is accurately aligned with the surface of an object all the time, and the image quality is improved. Dislocation and distortion are avoided, and the stability and visual quality of the interactive display effect are remarkably improved.

The invention provides a condition image generation method based on point rasterization and a related device, and the method comprises the steps: carrying out the coloring processing of a target point cloud set according to an environment image corresponding to a target moment; performing foreground and background separation processing on the target point cloud set after coloring processing to determine a dynamic foreground region and a static background region in the target point cloud set; complementing the incomplete area in combination with a historical point cloud set in the time window; projecting the complemented target point cloud set to an image coordinate system of the vehicle by using a perspective projection mapping relation to obtain a first projection image; and rendering the grating area in the first projection image to obtain a point rasterization condition image. According to the method, frame-by-frame analysis and image plane mapping are carried out on original laser radar point cloud data, point rasterization processing is combined, a geometric condition image with pixel-level precision is constructed, and structural prior support is provided for subsequent new view synthesis and controllable video generation.

The invention provides a three-dimensional semantic scene reconstruction method and device, electronic equipment and a storage medium, and belongs to the technical field of artificial intelligence. The method comprises the following steps: determining three-dimensional coordinates of RGB image pixel points in a world coordinate system; extracting two-dimensional image features of pixel points in the RGB image to obtain a two-dimensional feature image corresponding to the RGB image; based on the three-dimensional coordinates of the pixel points in the RGB image under the world coordinate system, mapping the pixel points in the RGB image to the three-dimensional occupied grid sub-model under the world coordinate system to obtain a target three-dimensional occupied grid sub-model; determining a grid feature vector of a three-dimensional grid in the target three-dimensional occupied grid sub-model based on the target three-dimensional occupied grid sub-model, the two-dimensional feature image corresponding to the RGB image and the projection mapping matrix; and calling the three-dimensional semantic prediction sub-model, processing the grid feature vector of the three-dimensional grid in the target three-dimensional occupation grid sub-model, and obtaining a good reconstruction result of the target indoor scene.

The invention discloses a strip mine unmanned terrain editable digital twinning method and system, and belongs to the technical field of digital twinning systems. The method comprises the following steps: constructing a three-dimensional terrain contour with topological information, and constructing a three-dimensional terrain with topological information based on the three-dimensional terrain contour; when the three-dimensional terrain changes, the latest three-dimensional terrain is downloaded, the latest boundary collected by the current vehicle is collected, a three-dimensional track route is drawn and updated, and the updated three-dimensional terrain is uploaded to a database; downloading two-dimensional graph data, and projecting and mapping a two-dimensional graph on a three-dimensional terrain; acquiring real-time coordinate data of the operation equipment, and adsorbing the real-time coordinate data on the three-dimensional terrain after projection mapping for driving; according to the method, the terrain can be dynamically edited, the editing flexibility higher than that of the traditional technology is provided, and the safety of mine field operation and the effectiveness of management are guaranteed.

The invention relates to the technical field of strain measurement, in particular to a three-dimensional full-field strain measurement method and system based on multi-view collaboration, and the method comprises the steps: constructing a sparse topology skeleton model of a measured target under a global coordinate system, and building a bidirectional projection mapping relation between the global situation and each local situation; the method comprises the following steps: synchronously acquiring a time sequence texture feature image of a measured target, calculating a luminosity geometric coupling confidence tensor of each pixel point and extracting an initial texture feature field under each view angle; utilizing a skeleton model to guide the initial texture feature field to perform limited local matching, and resolving to obtain a local independent displacement field and residual energy distribution of each view angle; performing multi-view weighted fusion based on a bidirectional projection mapping relation, and reconstructing a global three-dimensional displacement field; and performing differential calculation on the global three-dimensional displacement field to obtain a deformation gradient tensor and generate three-dimensional full-field strain distribution. According to the method, gradient mutation caused by view field splicing errors can be effectively inhibited, and physically real and full-field continuous three-dimensional strain distribution is obtained.

The present invention discloses a method for hardware acceleration of immersive projection mapping based on a CACHE architecture, and belongs to the field of image processing. The method includes six parts: target graphic scanning, source image coordinate calculation, on-chip buffering, coordinate conversion, interpolation processing, and image processing. A group-associated CACHE structure is used to optimize the problem of low efficiency of accessing SDR / DDR SDRAM due to discontinuous reading of source image pixels during graphic image projection transformation. Through a flexible CACHE structure and replacement strategy, it is ensured that better efficiency is achieved at different deformation angles. At the same time, by caching the source image coordinates to the on-chip cache, the main processing modules can be executed in parallel, further improving the processing speed. This method can achieve real-time processing of three-dimensional transformation of images with a resolution above full HD, avoiding the waste of storage bandwidth caused by writing to the memory area again.

The application provides a speed measurement method and an inspection system based on dynamic zoom of a UAV, comprising: acquiring a video stream collected by a camera of the UAV in real time by dynamically adjusting a focal length parameter; inputting the video stream into a detection model frame by frame; outputting a target detection frame set corresponding to a current frame; performing cross-frame association on detection frames of adjacent frames by a ByteTrack algorithm to assign a unique instance identifier to the same target; acquiring real-time parameters of the UAV, establishing a real road surface coordinate system and a pixel imaging coordinate system, and performing coordinate system projection mapping; and calculating a real speed of the target. The application has the beneficial effects that dynamic parameter fusion and UAV displacement compensation greatly reduce coordinate mapping errors, especially in a pitch angle or zoom scene, and the improvement is obvious; by introducing a center point position transformation, the application dynamically calibrates image center deviation caused by digital zoom, effectively eliminating pixel-physical coordinate mapping distortion caused by simple image cropping or interpolation enlargement.

The invention discloses a closed data space model test self-adaption method based on double-stage feature whitening, and belongs to the field of model self-adaption, and the method comprises the steps: inputting each enhanced view in an enhanced data set composed of a first enhanced view and a second enhanced view into a data processing model, obtaining an intermediate layer activation feature and an output result respectively corresponding to each enhanced view; determining a first whitening feature corresponding to each intermediate layer activation feature, and determining a second whitening feature corresponding to each first whitening feature after projection mapping based on all the first whitening features; and performing parameter updating on a target data layer in the data processing model based on joint model loss constructed by the output result and the second whitening feature. According to the method, a two-stage feature whitening mechanism is introduced, so that the model can obtain better adaptive ability in a closed data space, and the generalization performance and reasoning stability of the model in a complex test environment are improved.

The present application provides a kind of conditional image generation method and related device based on point rasterization, according to the environment image corresponding to target time, target point cloud set is colored processing;The target point cloud set after coloring processing is carried out front and background separation processing, to determine the dynamic foreground region and static background region therein;Combining the historical point cloud set in time window, the incomplete area is completed;Using perspective projection mapping relationship, the target point cloud set after completion is projected to the image coordinate system of vehicle, to obtain first projection image;The raster area in first projection image is rendered, to obtain the conditional image of point rasterization.It is analyzed frame by frame to original laser radar point cloud data and image plane mapping, and combined with point rasterization processing, constructs the geometric conditional image with pixel level precision, provides structure priori support for subsequent new view synthesis and controllable video generation.

The invention relates to the field of medical image processing, and discloses a method and device for fusing a DR image and an ultrasonic image, computer equipment and a storage medium, and the method comprises the steps: obtaining a DR image containing a DR detection array position; the position of the DR detection array is calibrated through optical positioning equipment; acquiring an ultrasonic image containing position data of a probe detection array; the position data of the probe detection array are calibrated through optical positioning equipment; determining a projection mapping relation between the DR image and the ultrasonic image according to the position data of the DR detection array and the position data of the probe detection array; and fusing the DR image and the ultrasonic image according to the projection mapping relation to obtain DR ultrasonic bimodal fusion data. According to the method, the precision and efficiency of DR image and ultrasonic image fusion are effectively improved, and reliable bimodal visualization support is provided for medical image diagnosis.

The invention discloses a high-density color point cloud generation method based on projection mapping and depth prediction, and belongs to the technical field of three-dimensional reconstruction and computer vision, and the method comprises the following steps: 1, sensor calibration: carrying out the internal reference calibration of a camera through a Zhang's calibration method, and obtaining an internal reference matrix K, determining an external parameter matrix T of the camera and the laser radar through joint calibration; in the invention, first, high-precision data fusion: through accurate sensor calibration and coordinate transformation, high-precision mapping of point cloud data and image pixels is realized, and the precision of data fusion is significantly improved; 2, high-efficiency calculation: through optimizing a depth prediction algorithm and setting an iteration convergence criterion, the method has relatively high calculation efficiency when processing large-scale data; and 3, accurate depth prediction: by introducing a gradient regularization item, depth information can be accurately predicted when a complex scene is processed, and the precision of three-dimensional reconstruction is improved.