CN104952075A - Laser scanning three-dimensional model-oriented multi-image automatic texture mapping method - Google Patents

Abstract

The invention discloses a multi-image automatic texture mapping method for laser scanning three-dimensional models, comprising: using a laser scanner to obtain a three-dimensional point cloud model of a three-dimensional scene, forming a corresponding triangular mesh model, and using a multi-image three-dimensional reconstruction system to solve the problem The 3D point cloud model of the 3D scene and the corresponding camera parameters, solve the rotation matrix, translation matrix, and scaling matrix between the triangular mesh model and the 3D point cloud model, and adjust the triangular mesh model to make it the most coincident with the multi-image 3D reconstruction model, Oriented to the adjusted triangular mesh model, the camera parameters are used to perform automatic texture mapping to obtain corresponding texture images. The multi-image automatic texture mapping method of the present invention greatly changes the current situation of poor texture mapping effects of traditional 3D models, and not only fully automates the entire texture mapping process and the resulting quality is better.

Description

Multi-image automatic texture mapping method for laser scanning 3D model

technical field

The invention relates to the field of computer vision, in particular to a multi-image-based three-dimensional reconstruction system and automatic texture mapping of a laser scanning model, in particular to a multi-image automatic texture mapping method for a laser scanning three-dimensional model.

Background technique

At present, the commonly used 3D reconstruction system based on multiple images uses the matching relationship between image feature vectors to solve the internal parameter matrix, external parameter matrix of the camera in the 3D scene and the coordinates of the world coordinate system of the 3D point.

Another commonly used method to obtain 3D data of a scene is to obtain a 3D model through laser scanning. In order to obtain as much detail as possible in the scene, it is necessary to use surface coloring instead of the point coloring method of the existing 3D model, that is, the pixels of each triangle patch on the triangle mesh model are taken from a triangle texture area on the texture image. Since the geometric structure of the laser scanning 3D model is highly accurate, the error of the camera parameters obtained by the multi-image 3D reconstruction method is very small, which can be calculated by combining the laser scanning 3D model and the 3D scene description (parameters of each camera) obtained based on the multi-image 3D reconstruction. A triangular mesh model with rich texture details is obtained.

The process of the 3D reconstruction system based on multiple images is to extract the feature vector of each image, match the feature vectors between the images, and use the matching result of the feature vector as the input of 3D reconstruction to obtain the camera parameters and 3D scene corresponding to each image The dense 3D point cloud, and eliminate the distortion of each image to obtain the image after distortion.

The input data required for multi-image automatic texture mapping are triangle mesh model, parameters of each camera and corresponding images.

In recent years, texture mapping has been carried out based on multi-image 3D reconstruction. Since the 3D model obtained based on multi-image 3D reconstruction has a large error in geometric structure, the final texture mapping result is not ideal. Laser scanning 3D model Introducing multiple image texture maps will make the final result richer in detail.

Contents of the invention

Aiming at the deficiencies of the prior art, the present invention provides a multi-image automatic texture mapping method for laser scanning three-dimensional models.

A multi-image automatic texture mapping method for laser scanning 3D models, comprising:

Step 1. Use a laser scanner to obtain a 3D point cloud model of a 3D scene, and form a corresponding triangular mesh model;

Step 2. Obtain several multi-directional and multi-angle images of the three-dimensional scene, and solve for each image corresponding to the camera internal parameter matrix, the camera external parameter matrix, and the dense three-dimensional point cloud of the three-dimensional scene;

The number of photos to be acquired must be at least 20. SLR cameras are usually used to take pictures of scenes in multiple directions and angles to obtain corresponding multi-directional and multi-angle images.

Step 3, matching the triangular mesh model with the dense 3D point cloud to obtain an aligned triangular mesh model;

Step 4. According to all the images, the camera internal parameter matrix and the external camera parameter matrix corresponding to each image, and the aligned triangular mesh model, obtain each triangle patch on the aligned triangular mesh model by using a perspective projection model method Corresponding texture regions in each image are spliced to obtain corresponding texture images.

The 3D point cloud model of the 3D scene of the present invention is obtained by scanning the scene with a laser scanner, and is not calculated by conventional multi-image 3D reconstruction. For the 3D point cloud model obtained by scanning with a laser scanner, the 3D point cloud model is removed during processing The colors are then triangulated to form a corresponding triangular mesh model.

Preferably, the triangular mesh model is in PLY format. Compared with other file formats representing 3D models, PLY can store triangular mesh models in a list of elements (3D vertices, texture coordinates, triangular patches) more simply,

The technology is mature and easy to implement. In the step 2, the multi-image three-dimensional reconstruction method is used to solve the camera internal parameter matrix, camera external parameter matrix and dense three-dimensional point cloud corresponding to each image.

By matching, the two models of the laser scanning 3D model (that is, the triangular mesh model) and the dense 3D point cloud obtained from the multi-image 3D reconstruction are most coincident, ensuring that the camera parameters obtained from the multi-image 3D reconstruction are valid under the laser scanning 3D model. As preferably, the step 3 is specifically as follows:

Calculate the displacement matrix between the triangular mesh model and the dense 3D point cloud according to the geometric similarity between the triangular mesh model and the dense 3D point cloud (geometric features such as contours), the displacement matrix includes Rotation matrix, translation matrix, scaling matrix and distortion parameters;

Taking the dense three-dimensional point cloud as a reference, adjusting the direction, position, and size of the triangular mesh model according to the displacement matrix to obtain an aligned triangular mesh model.

In the present invention, matching needs to be performed in the same world coordinate system.

In the present invention, when selecting the final texture region from all corresponding initial texture regions, the principle is to ensure that the final texture regions corresponding to all triangular patches come from the same image as much as possible. That is, select the one that covers the most pixels among the many selectable images, and select as many adjacent triangular patches as possible on the geometric topology to correspond to the same image to ensure the optimization of the texture mapping effect.

Each triangular patch corresponds to a texture region, when the texture regions corresponding to all the triangular patches are spliced into a texture image, there will be splicing gaps in the texture image obtained, in order to improve the quality of the texture image obtained, as a preference, the present invention also includes Perform chromatic aberration weakening processing on the stitched texture image.

In the present invention, the finally obtained texture images can be in three file formats: OBJ, MTL, and PNG, that is, the result of multi-image automatic texture mapping for laser scanning three-dimensional models.

Compared with the prior art, the multi-image automatic texture mapping method for the laser scanning 3D model of the present invention greatly improves the geometric accuracy of the triangular mesh model and the detail richness of the model after texture mapping, and is easy to implement.

Detailed ways

The present invention will be described in detail below in conjunction with specific examples. The following examples will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form.

The multi-image automatic texture mapping method for the laser scanning three-dimensional model of this embodiment includes:

Step 1. Use a laser scanner to scan the scene to obtain a 3D point cloud model, and obtain a triangular mesh model in PLY format after processing.

Step 2. Use a single-lens reflex camera to take pictures of the scene in multiple directions and angles, and use the multi-image 3D reconstruction method to solve the camera internal parameter matrix, camera external parameter matrix, and dense 3D point cloud of the scene corresponding to each image.

Step 3. Matching the laser scanning 3D model and the dense 3D point cloud obtained by multi-image 3D reconstruction, taking the dense 3D point cloud obtained by multi-image 3D reconstruction as a benchmark, and calculating Obtain the rotation matrix, translation matrix, and scaling matrix between the models, and adjust the direction, position, and size of the laser scanning 3D model so that the dense 3D point cloud obtained from the laser scanning 3D model and multi-image 3D reconstruction is the most coincident, and an aligned triangular mesh model is obtained.

The two models of the dense 3D point cloud obtained by matching are the most coincident, ensuring that the camera parameters obtained from the multi-image 3D reconstruction are valid under the laser scanning 3D model.

Step 4. Conversion of 3D scene expression methods, extracting the most concise 3D scene expression methods (camera focal length, rotation matrix, translation matrix, distortion parameters) that can be recognized by multi-image automatic texture mapping from the results obtained by the multi-image 3D reconstruction method to represent the 3D scene.

Step 5, multi-image automatic texture mapping, according to the perspective projection model and the 3D scene expression method extracted in step 4, obtain and align the texture area corresponding to each triangular patch on the 3D model and splicing the texture areas corresponding to all triangular patches Get the corresponding texture image as the mapping result.

In this embodiment, the corresponding texture area is obtained for each triangular patch by the following method:

The texture area corresponding to the triangular surface in each image is obtained by using a perspective projection model method as the initial texture area, and then one of the corresponding initial texture areas is selected as the final texture area.

When selecting the final texture region from all corresponding initial texture regions, it is a principle to ensure that the final texture regions corresponding to all triangular patches come from the same image as much as possible.

On the geometric topology, select as many adjacent triangular patches as possible to correspond to the same image while ensuring the optimization of the texture mapping effect.

The above-mentioned specific embodiments have described the technical solutions and beneficial effects of the present invention in detail. It should be understood that the above-mentioned are only the most preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, supplements and equivalent replacements made within the scope shall be included in the protection scope of the present invention.

Claims (7)

1. A multi-image automatic texture mapping method for laser scanning three-dimensional models, characterized in that, comprising: Step 1. Use a laser scanner to obtain a 3D point cloud model of a 3D scene, and form a corresponding triangular mesh model; Step 2. Obtain several multi-directional and multi-angle images of the three-dimensional scene, and solve for each image corresponding to the camera internal parameter matrix, the camera external parameter matrix, and the dense three-dimensional point cloud of the three-dimensional scene; Step 3, matching the triangular mesh model with the dense 3D point cloud to obtain an aligned triangular mesh model; Step 4. According to all the images, the camera internal parameter matrix and the external camera parameter matrix corresponding to each image, and the aligned triangular mesh model, obtain each triangle patch on the aligned triangular mesh model by using a perspective projection model method Corresponding texture regions in each image are spliced to obtain corresponding texture images. 2. the multi-image automatic texture mapping method for laser scanning three-dimensional model as claimed in claim 1, is characterized in that, described triangular mesh model is PLY format. 3. the multi-image automatic texture mapping method for laser scanning three-dimensional model as claimed in claim 1, is characterized in that, in described step 2, utilize multi-image three-dimensional reconstruction method to solve and obtain the camera internal parameter matrix corresponding to each image, camera External parameter matrix, and dense 3D point cloud. 4. the multi-image automatic texture mapping method for laser scanning three-dimensional model as claimed in claim 1, is characterized in that, described step 3 is specifically as follows: According to the geometric similarity between the triangular mesh model and the dense 3D point cloud, the displacement matrix between the triangular mesh model and the dense 3D point cloud is calculated, and the displacement matrix includes a rotation matrix, a translation matrix and a scaling matrix. matrix; Taking the dense three-dimensional point cloud as a reference, adjusting the direction, position, and size of the triangular mesh model according to the displacement matrix to obtain an aligned triangular mesh model. 5. the multi-image automatic texture mapping method facing laser scanning three-dimensional model as claimed in claim 1, is characterized in that, in described step 4, obtain corresponding texture area for each triangular surface by the following method: (4-1) Obtaining the texture area corresponding to the triangular patch in each image based on the perspective projection model method as the initial texture area; (4-2) Select one from all corresponding initial texture regions as the final texture region. 6. the multi-image automatic texture mapping method facing laser scanning three-dimensional model as claimed in claim 5, it is characterized in that, when selecting the final texture region from all corresponding initial texture regions, to ensure that the final texture regions corresponding to all triangle faces are as far as possible Probably derived from the same image as a principle. 7. The multi-image automatic texture mapping method for laser scanning three-dimensional models according to any one of claims 1 to 6, further comprising performing color difference weakening processing on the stitched texture images.