CN104408762A - Method for obtaining object image information and three-dimensional model by using monocular unit and two-dimensional platform - Google Patents

Abstract

The invention relates to a method for acquiring object image information and a three-dimensional model by using a monocular and two-dimensional platform. The method places the object on an electric two-dimensional rotating platform, and controls WNSC400 to accurately adjust each optimal viewpoint according to the optimal viewpoint planning. The required rotation angle, using the PCL point cloud library to filter, segment, sample, and register the point cloud, can quickly reconstruct the 3D model of the unknown object. This method not only has fast reconstruction speed and good practicability, but also has simple operation, low reconstruction cost, easy operation and good reconstruction effect.

Description

Method for Obtaining Object Image Information and 3D Model Using Monocular and 2D Platform

technical field

The invention relates to the technical field of fast automatic three-dimensional reconstruction, in particular to an automatic reconstruction method for quickly acquiring a three-dimensional model of an unknown object by using monocular structured light and an electric two-dimensional rotating platform.

Background technique

Human beings' perception of the external world mainly relies on vision. According to research, 80% of the external information obtained by human beings through their own senses comes from vision. Vision itself is a process that can perceive light signals, and can transmit, process, store and understand visual information. With the development of science and technology and the progress of human beings, people began to use cameras to imitate human eyes to obtain environmental images, convert them into digital signals, and then process visual information through computers, which is computer vision. On the other hand, due to the needs of industrial production and other aspects, people continue to put forward new requirements for the measurement technology of object shape. The development of computer vision makes people gradually realize that this will cause a major change in measurement technology, especially non-contact measurement technology.

With the rapid development of science and technology and industrial production, many fields require fast and accurate measurement of the shape of three-dimensional objects. Moreover, with the development of computer technology and the emergence of digital imaging devices, lasers and other optoelectronic components, three-dimensional shape measurement technology has gradually matured and is widely used in different fields such as robot vision, automatic manufacturing and product quality control according to their respective characteristics. Three-dimensional measurement technology can be divided into two categories according to different measurement methods, namely contact measurement and non-contact measurement. The traditional three-dimensional measurement method is the contact method, such as a three-coordinate machine, which uses a probe that can be precisely positioned to touch the surface of the object and measure the spatial coordinates of the touched point. The probe scans the surface of the object once, and the spatial coordinates of each point on the surface of the object can be obtained. The contact measurement technology represented by the three-coordinate machine cannot be replaced by other methods in terms of accuracy, but some inherent defects of contact measurement also limit its own development. First, the probe will generate a certain degree of pressure when it is in contact with the measured object, which is enough to cause a small deformation on the surface of the soft object, thus affecting the measurement results; second, the ideal probe should be a surface area of zero. Points, and the actual surface area of the probe cannot be zero, so it is impossible to measure the subtle features of some complex surfaces: Third, the point-by-point measurement method limits the measurement speed and is not suitable for measuring objects with large surface areas. Fourth, the contact with the precision measured object will also cause its surface to wear, and at the same time, it will cause damage to the probe itself, which limits the number of measurements and accuracy. Therefore, the defect of contact measurement comes from three-dimensional measurement technology, which has become a research hotspot in various countries. It makes up for the shortcomings of contact measurement. Although the measurement accuracy cannot be compared with contact measurement, with the progress of computer and microelectronics technology , this issue is also being resolved. Vision-based non-contact 3D measurement is to recover the 3D shape of the object from the depth image obtained by the 3D vision sensor. This non-contact three-dimensional measurement technology has the advantages of high efficiency, high degree of automation, and low cost, and has broad application prospects in industrial production and real life.

Accurate and efficient 3D measurement methods are not only the basis of 3D reconstruction, but also necessary for fast and automatic 3D reconstruction. The existing three-dimensional model measurement methods are divided into contact measurement and non-contact measurement.

1. Contact measurement: Contact measurement mainly uses some mechanical manual measuring instruments or electronic measuring instruments composed of contact sensors, among which manual measurement is the main method. The tools for manual measurement mainly include: cloth tape ruler, steel tape measure, drawing pen, height measuring instrument, footprint device and foot calipers. Manual measurement is divided into direct method and indirect method in terms of measurement methods. The direct method is relatively simple. First, use a pen to mark the relevant measurement points on the characteristic parts of the feet, and use tools such as a cloth tape ruler and an altimeter to directly measure various relevant data of the feet. The indirect method uses a footprint tool to take a footprint and draw the outline of the foot, which is then measured and analyzed. The advantages of contact foot type measurement are less investment, simple operation steps, strong flexibility, easy to carry, and easy to measure at different measurement locations in a short time. The disadvantage is that the contact measurement method will generate pressure on the feet, cause deformation of the feet, cause measurement errors, take a long time to measure, and the measured parameters of the feet are limited, some more complex parameters or curves cannot be measured, low work efficiency and labor intensity Large, poor reproducibility, large errors between measurers, and at the same time, it is difficult to obtain the data information of the whole foot with this measurement method. Data collection cannot be performed on the cross-sectional shape of the foot either. Unable to conduct deeper research.

2. Non-contact measurement: Driven by the development of computer, optics and other disciplines. Optical non-contact measurement, as a measurement technology emerging in recent years, has received more and more attention. At present, optical non-contact measurement methods mainly include composition measurement, stereo vision measurement, laser measurement and so on.

(1) Structured light measurement: structured light 3D vision is based on the principle of optical triangulation. Such as a camera) to obtain images, through the geometric relationship of the system, using the triangulation principle to calculate the three-dimensional coordinates of the object. The structured light measurement method has the characteristics of simple calculation, small size, low price, easy installation and maintenance, and is widely used in actual 3D profile measurement. However, the measurement accuracy is limited by physical optics, and there is an occlusion problem. The measurement accuracy and speed are contradictory, and it is difficult to improve at the same time.

(2) Stereo vision measurement: In the computer vision system, two cameras with relatively fixed positions or one camera at two different positions are used to obtain two images of the same scene from different angles at the same time. The disparity in the image to obtain its three-dimensional coordinate value. The biggest feature of the stereo vision method is that the shooting speed is fast, and the shooting task can be completed in less than one second, which is suitable for occasions that require fast measurement. However, the stereo vision method has a large amount of data processing, a long processing time, and needs to match two images. When the gray level and surface shape of the object surface do not change much, the matching and measurement accuracy will be affected.

(3) Laser measurement: A linear visible laser beam positioned by a polygonal lens scans and measures the surface of the object through high-frequency scanning; applying the law of trigonometry, the laser beam is received by the laser receiver after being reflected on the surface of the object. Then the coordinates of the surface of the object are obtained through calculation. Laser scanning can accurately provide three-dimensional environmental information, data processing is simple, and it is less affected by the environment. But the cost is high, and there is a contradictory relationship between accuracy, ranging and scanning rate.

3D reconstruction technology has developed rapidly in recent years, mainly due to the development of science and technology, the reduction of 3D reconstruction costs, especially the maturity of 3D reconstruction technology, and the rapid development of structured light systems, which have greatly inspired researchers to apply 3D reconstruction technology. It is applied to various research fields such as medicine, entertainment, machinery, etc.

The system used in this patent includes a short-focus camera, projector, WNSC400, and corresponding software equipment. This reconstruction technology uses the method of combining Gray code and phase shifting, and performs three-dimensional measurement on the objects placed on WNSC400 according to the principle of triangulation, and analyzes the geometry and topology information of unknown objects from viewpoint 1, so as to obtain the three-dimensional information of other viewpoints successively , a complete 3D model is obtained by registration of several viewpoints.

Compared with manual measurement and other 3D scanning measurements, the advantages of this system are: (1) It can quickly perform 3D measurement on unknown objects; (2) It can automatically stitch the viewpoints and quickly complete the reconstruction of the finishing model; (3) The structure The optical system is cheap and the cost of use is low; (3) The operation is simple, and the user can quickly grasp its principle.

Contents of the invention

The invention relates to a method for acquiring object image information and a three-dimensional model by using a monocular and two-dimensional platform. The method not only has fast reconstruction speed and good practicability, but also has simple operation, low reconstruction cost, easy operation and good reconstruction effect.

In order to achieve the above object, the technical solution of the present invention is: a method for obtaining object image information and a three-dimensional model using a monocular and two-dimensional platform, comprising the following steps:

(1) Fast reconstruction: Firstly, the internal and external parameters of the camera and projector are calibrated, and according to the method of combining Gray code and phase shift, the image information is obtained by using the short focus in the monocular structured light, and the unknown object is reconstructed in three dimensions. The principle is to obtain the point cloud data of a certain viewpoint, remove the noise points, and obtain the 3D model under the viewpoint;

(2) View point planning: the user places the object to be reconstructed on the WNSC400 (electric two-dimensional rotating table), uses the camera to obtain the geometric information and topology information under the point of view, and compares the size of the visible area obtained under the left and right reference positions , take the position of the larger one as the position of the next viewpoint, and finally judge whether to complete the reconstruction through the planning self-termination criterion;

(3) Point cloud registration: reconstruct the 3D point cloud from viewpoint 1-n, transform the point cloud data into the coordinate system of viewpoint 1, use the PCL point cloud library to read the point cloud, and perform ICP (Iterative Closest Point Algorithm) iteration , when the iteration tends to be stable and the registration ends, the point cloud in the overlapping area is removed, the reconstructed point cloud data is analyzed, and the complete 3D model of the object is obtained;

In an embodiment of the present invention, the step (1) includes the following steps:

(1.1) Calibration of internal and external parameters: According to Zhang Zhengyou’s calibration method, use the matlab calibration toolbox to calibrate the internal parameter matrix of the projector and short-focus camera, and then use the checkerboard to calibrate the R (rotation) and T of the short-focus and projector (translation) matrix, so as to calibrate the internal and external parameters;

(1.2) According to the method of combining Gray code and phase shift, the short focus in monocular structured light is used to reconstruct the unknown object in 3D and obtain its point cloud data.

(1.3) Optimizing the surface quality of the model: denoising is carried out through a smoothing algorithm, the measured 3D data is optimized, and the accuracy of the 3D data is improved.

In one embodiment of the present invention, the step (2) uses a WNSC400 electric two-dimensional rotary table, including the following steps:

(2.1) WNSC400 initialization: set the initial parameters of the electric two-dimensional rotary table, calculate its pulse equivalent, and the pulse amount required for a certain angle of rotation;

(2.2) Extract the point cloud information under viewpoint 1: place the object on WNSC400, reconstruct the 3D point cloud information, convert the reconstructed point cloud into pcd format, and use the PCL point cloud library to perform direct filtering on the point cloud, Separate the useful minimum point cloud under viewpoint 1, then use the filter to filter, remove the discrete points, and finally divide the filtered point cloud into several types of point cloud data through Euclidean clustering, and set the threshold , to get the useful point cloud data under viewpoint 1 with the largest amount of point cloud;

(2.3) Determine the next optimal viewpoint (NVB): From the point cloud data obtained in 2.2, analyze the geometry and topology information of the previous viewpoint to obtain the largest viewing area as the next optimal viewpoint;

(2.4) Reconstruct the overall model: obtain each optimal viewpoint, reconstruct the 3D point cloud of each viewpoint, use the angle criterion as the planning self-termination criterion, and align the data obtained after n times of planning to the initial viewpoint coordinate system (workbench The angle formed between the center and the boundary of the model obtained at the previous viewpoint and another boundary of the model obtained at the current viewpoint is defined as the boundary angle), which satisfies the angle criterion (that is, the sum of the boundary angles is greater than ), it is considered that the three-dimensional reconstruction of the object surface is completed and the overall model is obtained.

Compared with the prior art, the beneficial effect of the present invention is that the monocular structured light system and the WNSC400 are applied to fast acquisition of three-dimensional models of unknown objects and automatic reconstruction. It is easy to realize, only one monocular structured light system and one WNSC400 are needed to complete the three-dimensional measurement of the object; the data collection efficiency is high; this patent adopts the optical coding technology, and uses continuous lighting, which does not require special photosensitive materials, only ordinary CMOS photosensitive chip, so the cost can be greatly reduced; combined with traditional 3D reconstruction technology, PCL technology is used in point cloud processing, which can quickly and automatically become the splicing of each viewpoint, and the reconstruction efficiency is high; the measurement efficiency is high, and it has a strong Practicality and broad application prospects.

Description of drawings

Fig. 1 is a model diagram of a certain character model in viewpoint 1 in an embodiment of the present invention.

Fig. 2 is a model diagram in the embodiment of the present invention, taking a character model as an example under viewpoint 2.

Fig. 3 is a model diagram under viewpoint 3 taking a character model as an example in the embodiment of the present invention.

Fig. 4 is a model diagram of a certain character model in viewpoint 4 in the embodiment of the present invention.

Fig. 5 is an encapsulation diagram of an overall model taking a character model as an example in the embodiment of the present invention.

Detailed ways

A measurement method for quickly acquiring image information and a three-dimensional model of an unknown object by using monocular structured light and an electric two-dimensional rotating platform, comprising the following steps:

(1) Fast reconstruction: Firstly, the internal and external parameters of the camera and projector are calibrated, and according to the method of combining Gray code and phase shift, the short-focus in the monocular structured light has the function of quickly obtaining image information, and three-dimensional reconstruction of unknown objects , obtain the point cloud data of a certain point of view by the principle of triangulation, remove the noise points, and obtain the 3D model under the point of view, which specifically includes the following steps:

(1.1) Calibration of internal and external parameters: According to Zhang Zhengyou’s calibration method, use the matlab calibration toolbox to calibrate the internal parameter matrix of the projector and short-focus camera, and then use the checkerboard to calibrate the R (rotation) and T of the short-focus and projector (translation) matrix, so as to calibrate the internal and external parameters;

(1.2) According to the method of combining Gray code and phase shift, the short focus of binocular structured light is used to reconstruct the unknown object in 3D, and its point cloud data is obtained.

(1.3) Optimizing the surface quality of the model: denoising is carried out through the smoothing algorithm, the measured three-dimensional data is optimized, and the accuracy of the three-dimensional data is improved.

(2) View point planning: the user places the object to be reconstructed on the WNSC400 (electric two-dimensional rotating table), uses the camera to obtain the geometric information and topology information under the point of view, and compares the size of the visible area obtained under the left and right reference positions , take the position of the larger one as the position of the next viewpoint, and finally determine whether the reconstruction is completed by planning the self-termination criterion, which specifically includes the following steps:

(2.1) WNSC400 initialization: set the initial parameters of the electric two-dimensional rotary table, calculate its pulse equivalent, and the pulse amount required for a certain angle of rotation;

(2.2) Extract the point cloud information under viewpoint 1: place the object on WNSC400, reconstruct the 3D point cloud information, convert the reconstructed point cloud into pcd format, and use the PCL point cloud library to perform direct filtering on the point cloud, Separate the useful minimum point cloud under viewpoint 1, then use the filter to filter, remove the discrete points, and finally divide the filtered point cloud into several types of point cloud data through Euclidean clustering, and set the threshold , to get the useful point cloud data under viewpoint 1 with the largest amount of point cloud;

(2.3) Determine the next optimal viewpoint (NVB): From the point cloud data obtained in 2.2, analyze the geometry and topology information of the previous viewpoint to obtain the largest viewing area as the next optimal viewpoint;

(2.4) Reconstruct the overall model: obtain each optimal viewpoint, reconstruct the 3D point cloud of each viewpoint, use the angle criterion as the planning self-termination criterion, and align the data obtained after n times of planning to the initial viewpoint coordinate system (workbench The angle formed between the center and the boundary of the model obtained at the previous viewpoint and another boundary of the model obtained at the current viewpoint is defined as the boundary angle), which satisfies the angle criterion (that is, the sum of the boundary angles is greater than ), it is considered that the three-dimensional reconstruction of the object surface is completed and the overall model is obtained.

(3) Point cloud registration: reconstruct the 3D point cloud from viewpoint 1-n, transform the point cloud data into the coordinate system of viewpoint 1, use the PCL point cloud library to read the point cloud, and perform ICP (Iterative Closest Point Algorithm) iteration , when the iteration tends to be stable and the registration ends, the point cloud in the overlapping area is removed, the reconstructed point cloud data is analyzed, and the complete 3D model of the object is obtained;

The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

1. As shown in Figure 1, put the binocular structured light system and WNSC400 in a suitable position, and place the object to be reconstructed (take the character model as an example) on top of WNSC400.

2. Calibrate the camera and projector separately to obtain their internal parameters: focal length, principal point coordinates, and distortion parameters. At the same time, calibrate the external parameters of the camera and projector to obtain their mutual R (rotation), T (translation) matrix.

3. Put the object to be reconstructed on the WNSC400, and use the principle of triangulation according to the method of combining Gray code and phase shift to quickly complete the reconstruction with short focus.

4. For the point cloud obtained in 3, denoising is carried out through a smoothing algorithm to optimize the measured 3D data and improve the accuracy of the 3D data.

5. Analyze the preprocessed point cloud to obtain its geometric information and topological information. By comparing the size of the visible area obtained under the left and right reference positions, the position of the larger one is taken as the position of the next viewpoint. According to the planning self-termination criterion , until all viewpoints have been reconstructed.

5.1 Convert the obtained point cloud information into pcd format, use PCL to perform through filtering, and separate the useful minimum point cloud under viewpoint 1, then use the filter to filter, remove discrete points, and finally pass Euclidean aggregation The class divides the filtered point cloud into several types of point cloud data, sets the threshold value, and obtains the useful point cloud data under viewpoint 1 for the one with the largest amount of point cloud (as shown in Figure 1, the point cloud data under viewpoint 1) ;

5.2 From the point cloud data obtained in 5.1, analyze the geometry and topology information of the previous viewpoint to obtain the largest viewing area as the next optimal viewpoint;

5.3 For each optimal viewpoint obtained from the calculation in 5.2, adjust the WNSC400 to reconstruct the 3D point cloud of each viewpoint, and use the angle criterion as the planning self-terminating criterion to reconstruct the overall model.

6. Reconstruct the 3D point cloud from viewpoint 1-4 (Fig. 1-4), transform the point cloud data into the coordinate system of viewpoint 1, use PCL to read, and perform ICP iteration. When the iteration tends to be stable, the registration ends , remove the point cloud in the overlapping area, analyze the reconstructed point cloud data, and obtain the 3D measurement model of the person (as shown in Figure 5).

The above are the preferred embodiments of the present invention, and all changes made according to the technical solution of the present invention, when the functional effect produced does not exceed the scope of the technical solution of the present invention, all belong to the protection scope of the present invention. the

Claims (3)

1. utilize monocular and two-dimensional stage to obtain a method for object image information and three-dimensional model, it is characterized in that, comprise the following steps:

(1) Fast Reconstruction: the inside and outside parameter first calibrating camera and projector, according to the method that Gray code is combined with phase shift, short Jiao in monocular structured light is utilized to obtain image information, three-dimensional reconstruction is carried out to unknown object, the cloud data of a certain viewpoint is obtained by principle of triangulation, pick impurity point, obtain the three-dimensional model under this viewpoint;

(2) viewpoint planning: object to be reconstructed is placed on 2-d rotating platform by user, camera is utilized to obtain geological information under this viewpoint and topology information, the size of effective area is obtained by comparing lower of reference position, left and right, get the position of large person position as next viewpoint, finally by planning self termination criterion, judge whether reconstruction;

(3) registration point cloud: the three-dimensional point cloud that viewpoint 1-n is rebuild out, under cloud data being transformed into viewpoint 1 coordinate system, PCL point cloud storehouse is utilized to read some cloud, carry out ICP iterative closest point algorithms iteration, when iteration tends towards stability, registration terminates, and removes overlapping region point cloud, analyze the cloud data reconstructed, obtain the complete three-dimensional model of object.

2. the method utilizing monocular and two-dimensional stage to obtain object image information and three-dimensional model according to claim 1, it is characterized in that, described step (1) comprises the following steps:

(1.1) demarcation of inside and outside parameter: according to Zhang Zhengyou scaling method, utilize matlab calibration tool case, calibrate the Intrinsic Matrix of projector and short burnt camera, the R(that recycling gridiron pattern calibrates short Jiao and projector rotates), T(translation) matrix, thus demarcation obtains inside and outside parameter;

(1.2) according to the method that Gray code is combined with phase shift, utilize short Jiao in monocular structured light to carry out three-dimensional reconstruction to unknown object fast, obtain its cloud data;

(1.3) Optimized model surface quality: carried out denoising by smoothing algorithm, optimizes the three-dimensional data measured, and improves the precision of three-dimensional data.

3. the method utilizing monocular and two-dimensional stage to obtain object image information and three-dimensional model according to claim 1, it is characterized in that, described step (2) adopts the electronic 2-d rotating platform of WNSC400, comprises the following steps:

(2.1) WNSC400 initialization: the initial parameter setting electronic 2-d rotating platform, calculates its pulse equivalency, and rotate to an angle required pulsed quantity;

(2.2) the some cloud information under viewpoint 1 is extracted: be placed in by object on WNSC400, reconstruct three-dimensional point cloud information, by the some cloud reconstructed, be converted into pcd form, utilize PCL point cloud storehouse to carry out straight-through filtering to a cloud, isolate the useful smallest point cloud under viewpoint 1, connect down and re-use wave filter and carry out filtering, remove discrete point, finally by Euclidean cluster, the some cloud after is after filtering partitioned into a few class cloud data, arrange threshold values, obtaining the large person of a cloud amount is useful cloud data under viewpoint 1;

(2.3) determine next best view (NVB): the cloud data obtained by step (2.2), analyze geometry and the topology information of last viewpoint, to obtain maximum effective area for next best view;

(2.4) block mold is rebuild: obtain each best view, rebuild the three-dimensional point cloud of each viewpoint, utilize angle criterion as planning self termination criterion, by n time planning after obtain alignment of data under initial viewpoint coordinate system, meet angle criterion, then think to the three-dimensional reconstruction of body surface, obtained block mold.