CN-121999124-A - Three-dimensional reconstruction method, system, terminal and storage medium in night low-illumination scene

CN121999124ACN 121999124 ACN121999124 ACN 121999124ACN-121999124-A

Abstract

The invention relates to the field of data processing and discloses a three-dimensional reconstruction method, a system, a terminal and a storage medium under a night low-illumination scene, wherein the method comprises the steps of constructing a three-dimensional geometric texture model, obtaining a night low-illumination simulation scene based on the texture model, simulating a low-altitude acquisition track of an unmanned aerial vehicle in the simulation scene, and acquiring night real data and daytime real data in real city real scenes; the method comprises the steps of constructing a target reconstruction data set based on a texture model, a simulation scene, a low-altitude acquisition track of an unmanned aerial vehicle, night real data and daytime real data, constructing an enhancement network model, training the enhancement network model by using the target reconstruction data set and a low-illumination image data set to obtain a target enhancement network model, acquiring a night low-illumination image to be processed, inputting the night low-illumination image into the target enhancement network model to obtain an enhanced image, and reconstructing a 3D point cloud based on the enhanced image. The invention can efficiently realize three-dimensional reconstruction application under urban night scenes.

Inventors

TIAN QIN
Ling Qiuyuan
Yu Wenlv
ZHANG CHENGYUE
ZHANG XIAOYI
LIANG ZIFENG
CAO HONGBIN
ZHONG XIAORONG
WU HAIMING

Assignees

深圳大学
深圳市规划和自然资源数据管理中心(深圳市空间地理信息中心)

Dates

Publication Date: 20260508
Application Date: 20251205

Claims (10)

1. The three-dimensional reconstruction method under the night low-illumination scene is characterized by comprising the following steps of: Constructing a three-dimensional geometric texture model according to measured data, obtaining a city night low-illumination simulation scene based on the three-dimensional geometric texture model, simulating a low-altitude acquisition track of an unmanned aerial vehicle in the city night low-illumination simulation scene, and acquiring night three-dimensional reconstruction real data and daytime three-dimensional reconstruction real data in a real city real scene; Constructing a target reconstruction data set based on the three-dimensional geometric texture model, the urban night low-illumination simulation scene, the unmanned aerial vehicle low-altitude acquisition track, the night three-dimensional reconstruction real data and the daytime three-dimensional reconstruction real data; Constructing an enhanced network model, and performing supervision training on the enhanced network model by using the target reconstruction data set and the low-illumination image enhancement field public data set to obtain a target enhanced network model; and acquiring a night low-illumination image to be processed, inputting the night low-illumination image into the target enhancement network model for enhancement processing, obtaining an enhanced image, and reconstructing a 3D point cloud based on the enhanced image.
2. The method for three-dimensional reconstruction under a night low-light scene according to claim 1, wherein the constructing a three-dimensional geometric texture model according to measured data, obtaining a city night low-light simulation scene based on the three-dimensional geometric texture model, and simulating a low-altitude acquisition track of the unmanned aerial vehicle in the city night low-light simulation scene, comprises the following steps: Acquiring a daytime image of a real world scene through aerial photography, acquiring camera parameters and GPS information in the aerial photography process, and carrying out three-dimensional reconstruction on the daytime image, the camera parameters and the GPS information to obtain a three-dimensional geometric texture model; In a virtual engine, adding a high-simulation light effect for the three-dimensional geometric texture model to obtain a city night low-illumination simulation scene, wherein the high-simulation light effect comprises different light sources combined with the real city night scene layout and different brightness distribution arranged according to light intervals; Setting motion tracks of a rendering camera in the urban night low-illumination simulation scene, adjusting the key point position, the flying height and the viewing angle of each motion track, adjusting the parameters of the rendering camera to obtain adjusted parameters, and setting a high-altitude overview path and a low-altitude detail path; And simulating a low-altitude acquisition track of the unmanned aerial vehicle according to the motion track, the key point position, the flying height, the visual angle, the adjusted parameters, the high-altitude overview path and the low-altitude detail path.
3. The method for three-dimensional reconstruction in a night low-light scene according to claim 1, wherein the step of collecting the night three-dimensional reconstruction real data and the daytime three-dimensional reconstruction real data in the real city reality comprises the following steps: Designing an acquisition area, wherein the acquisition area comprises a common scene and a high-challenge scene of a three-dimensional reconstruction algorithm; Planning an acquisition track, and covering a target area by adopting a network-type low-altitude flight path until the coverage rate of the path reaches a preset threshold value; controlling illumination conditions, and respectively acquiring daytime normal illumination and nighttime low illumination of the unmanned aerial vehicle under the same pose view angles in different time periods of the same day; and acquiring night three-dimensional reconstruction real data and daytime three-dimensional reconstruction real data under the acquisition region, the acquisition track and the illumination condition.
4. The method according to claim 1, wherein the target enhanced network model comprises a frequency decomposition module, a high frequency information flow module, a low frequency information flow module, and a fusion reconstruction module.
5. The method for three-dimensional reconstruction in a night low illumination scene according to claim 4, wherein the step of inputting the night low illumination image into the target enhancement network model for enhancement processing to obtain an enhanced image specifically comprises: inputting the night low-illumination image into the frequency decomposition module for decomposition to obtain high-frequency data characteristics and low-frequency data characteristics; inputting the high-frequency data characteristics into the high-frequency information flow module for texture detail enhancement and denoising processing to obtain first output data; Inputting the low-frequency data characteristics into the low-frequency information flow module to perform brightness enhancement and color enhancement to obtain second output data; And inputting the first output data and the second output data into the fusion reconstruction module for fusion reconstruction to obtain an enhanced image.
6. The method according to claim 5, wherein the high frequency information stream module comprises five residual convolution layer sub-modules, the residual convolution layer sub-modules comprising a convolution layer and a correction linear unit; the step of inputting the high frequency data characteristics into the high frequency information flow module for texture detail enhancement and denoising processing to obtain first output data specifically comprises the following steps: inputting the high-frequency data characteristics into the convolution layer and the correction linear unit of the residual convolution layer sub-module to perform texture detail enhancement and edge enhancement to obtain high-frequency enhancement data; and removing noise from the high-frequency enhanced data to obtain first output data.
7. The method of three-dimensional reconstruction in a night low illumination scene as set forth in claim 6, wherein the low frequency information stream module comprises an encoder, a consistency constraint module, and a decoder; The step of inputting the low frequency data characteristic into the low frequency information flow module for brightness enhancement and color enhancement to obtain second output data specifically comprises the following steps: Inputting the low-frequency data characteristics into the encoder of the low-frequency information flow module to gradually downsample to generate a multi-level low-frequency characteristic diagram; marking the two-dimensional image pixels of the low-frequency feature map to obtain a marked feature map, and mapping the marked feature map into a sequence with a fixed image block size; Selecting a local area with consistent dimensions from the sequence as a target image block, inputting the target image block into the consistency constraint module of the low-frequency information flow module for multi-head attention calculation, and splicing calculation results to obtain a constraint feature map; And inputting the constraint characteristic diagram to the decoder of the low-frequency information flow module for step-by-step up sampling to obtain enhanced second output data.
8. A three-dimensional reconstruction system in a night low-light scene, the three-dimensional reconstruction system in the night low-light scene comprising: The data simulation module is used for constructing a three-dimensional geometric texture model according to the measured data, obtaining a city night low-illumination simulation scene based on the three-dimensional geometric texture model, simulating a low-altitude acquisition track of the unmanned aerial vehicle in the city night low-illumination simulation scene, and acquiring night three-dimensional reconstruction real data and daytime three-dimensional reconstruction real data in a real city real scene; the data set construction module is used for constructing a target reconstruction data set based on the three-dimensional geometric texture model, the urban night low-illumination simulation scene, the unmanned aerial vehicle low-altitude acquisition track, the night three-dimensional reconstruction real data and the daytime three-dimensional reconstruction real data; The enhancement model construction module is used for constructing an enhancement network model, and performing supervision training on the enhancement network model by using the target reconstruction data set and the low-illumination image enhancement field public data set to obtain a target enhancement network model; the three-dimensional reconstruction application module is used for acquiring a night low-illumination image to be processed, inputting the night low-illumination image into the target enhancement network model for enhancement processing, obtaining an enhanced image, and reconstructing a 3D point cloud based on the enhanced image.
9. A terminal comprising a memory, a processor and a three-dimensional reconstruction program stored on the memory and operable on the processor in a night low-light scene, the three-dimensional reconstruction program in the night low-light scene when executed by the processor implementing the steps of the three-dimensional reconstruction method in the night low-light scene as claimed in any one of claims 1-7.
10. A computer readable storage medium storing a three-dimensional reconstruction program in a night low-light scene, which when executed by a processor, implements the steps of the three-dimensional reconstruction method in a night low-light scene as claimed in any one of claims 1-7.

Description

Three-dimensional reconstruction method, system, terminal and storage medium in night low-illumination scene Technical Field The present invention relates to the field of data processing, and in particular, to a three-dimensional reconstruction method, system, terminal, and computer readable storage medium in a night low-light scene. Background Two-dimensional visual tasks mainly acquire information from a single image, three-dimensional space information is hardly effectively utilized, and three-dimensional information can truly reflect states of objects and environments and is closer to a human perception mode. The three-dimensional reconstruction is a key technology in three-dimensional vision, aims to establish a 3D model expressing an objective world through a computer, can reconstruct object surface profile information in a real environment through two-dimensional multi-view image data of an object, and has an irreplaceable effect in the fields of smart city map building, disaster emergency response and the like as a key technology foundation of digital twin and space intelligence. However, the current technical system is mainly applied to daytime scenes, the existing three-dimensional reconstruction frame is insufficient in adaptability to nighttime scenes, the problem of urban complex low illumination is difficult to effectively treat, and the lack of the urban nighttime three-dimensional reconstruction reference data set further causes the lack of reliable data support for algorithm training and evaluation, so that the construction of an all-weather operation and maintenance system of a city is severely restricted. Accordingly, the prior art is still in need of improvement and development. Disclosure of Invention The invention mainly aims to provide a three-dimensional reconstruction method, a system, a terminal and a computer readable storage medium in a night low-illumination scene, and aims to solve the problems that in the prior art, the lack of a city night three-dimensional reconstruction reference data set further causes the lack of reliable data support for algorithm training and evaluation and the difficulty of three-dimensional reconstruction in the city night low-illumination scene is high. In order to achieve the above object, the present invention provides a three-dimensional reconstruction method in a night low-illuminance scene, the three-dimensional reconstruction method in the night low-illuminance scene comprising the steps of: Constructing a three-dimensional geometric texture model according to measured data, obtaining a city night low-illumination simulation scene based on the three-dimensional geometric texture model, simulating a low-altitude acquisition track of an unmanned aerial vehicle in the city night low-illumination simulation scene, and acquiring night three-dimensional reconstruction real data and daytime three-dimensional reconstruction real data in a real city real scene; Constructing a target reconstruction data set based on the three-dimensional geometric texture model, the urban night low-illumination simulation scene, the unmanned aerial vehicle low-altitude acquisition track, the night three-dimensional reconstruction real data and the daytime three-dimensional reconstruction real data; Constructing an enhanced network model, and performing supervision training on the enhanced network model by using the target reconstruction data set and the low-illumination image enhancement field public data set to obtain a target enhanced network model; and acquiring a night low-illumination image to be processed, inputting the night low-illumination image into the target enhancement network model for enhancement processing, obtaining an enhanced image, and reconstructing a 3D point cloud based on the enhanced image. Optionally, the method for reconstructing a three-dimensional scene under night low-illuminance includes constructing a three-dimensional geometric texture model according to measured data, obtaining a city night low-illuminance simulation scene based on the three-dimensional geometric texture model, and simulating a low-altitude acquisition track of the unmanned aerial vehicle in the city night low-illuminance simulation scene, wherein the method specifically includes: Acquiring a daytime image of a real world scene through aerial photography, acquiring camera parameters and GPS information in the aerial photography process, and carrying out three-dimensional reconstruction on the daytime image, the camera parameters and the GPS information to obtain a three-dimensional geometric texture model; In a virtual engine, adding a high-simulation light effect for the three-dimensional geometric texture model to obtain a city night low-illumination simulation scene, wherein the high-simulation light effect comprises different light sources combined with the real city night scene layout and different brightness distribution arranged according to light intervals; Setti