CN103236078A - Facet-based complex scene three-dimensional reconstruction method - Google Patents

Abstract

A facet-based complex scene three-dimensional reconstruction method is characterized by including: firstly, obtaining intrinsic and extrinsic parameters and distortion coefficients in a depth camera and a color camera by three-dimensional calibration; secondly, calculating three-dimensional information of a complex scene by the depth camera according to the principle of triangulation; thirdly, defining a point model as a facet, with each value as an attribute of the point model including the attributes such as normal vector and radius; fourthly, subjecting each facet to surface 'internal, external and intersection' relation test with another facet; fifthly, obtaining confidence neighborhood of the facets which are judged intersected, calculating boundaries of the most adjacent facets on the point model, and resampling the facets in the confidence neighborhood; sixthly, three-dimensionally drawing the complex scene according to results of Boolean operation in the step 5. The facet-based complex scene three-dimensional reconstruction method is highly timely and available for effectively extracting three-dimensional point cloud data of the complex scene.

Description

A kind of complex scene three-dimensional rebuilding method based on bin

Technical field

The present invention relates to computer vision field, be specifically related to the three-dimensional rebuilding method of complex scene.

Background technology

Technology to object or scene surface reconstruction mainly is divided into following several: conventional geometric formative method, 3-D scanning equipment obtain method and based on the three-dimensional reconstruction method of stereoscopic vision.The conventional geometric formative method need be measured the size of object, be difficult to irregular curved face object is carried out modeling, and the production process complexity has very high professional knowledge requirement to the modeling personnel, and modeling software mainly comprises AutoCAD, 3D-Max, Maya, OpenGL etc.3-D scanning equipment is divided into contact and contactless two kinds substantially, and it is simple that object is carried out method for reconstructing, can obtain accurate three-dimensional model.But these equipment prices are very expensive.Utilize binocular camera shooting head or degree of depth camera based on the three-dimensional reconstruction method of stereoscopic vision, the three-dimensional coordinate by principle of triangulation acquisition object or scene can obtain the higher reconstruction effect of precision, and this method economy is convenient, has obtained application more and more widely.

Scene rebuilding mainly is divided into traditional curved surface modeling mode and based on the modeling pattern of point model.Mainly comprise following three kinds based on the curved surface's modeling mode: grid surface adopts point, straight line and plane dough sheet to represent geometric model, can express the three-dimensional body of arbitrary topology and arbitrary shape.Parametric surface reconstruct is explicit surface reconstruction just, is the main method of describing geometric model always, the parametric surface method be simple and easy to, be easy to determine the position of putting on the curve and surface.The implicit surface method is a kind of curve match, is easy to represent the geometric model of topological structure complexity, need not discrete point cloud is carried out parametrization.Based on the modeling of the uncomfortable chalaza cloud of curved surface's modeling method: 1) polygonal mesh comprises the topology between the sampled point, stores a large amount of resource of these informational needs.2) based on the geometric expression mode of point model, the consistent topological relation of the Maintenance free overall situation.

Point model method for drafting based on bin is to regard a part as a directive thin discs, all disks cover mutually and constitute one closely the surface.Compare with traditional curved surface modeling mode, advantage based on the modeling method of bin is as follows: can finer and smoothlyer present three-dimensional detail, can draw complex-curved and three-dimensional scenic, storage is simple, do not have complicated overall topology information, need not safeguard the topological consistance of geometric jacquard patterning unit surface.

The existing defective that exists based on the scene modeling mode of bin is: real-time three-dimensional relatively poor, that can not extract complex scene is effectively tieed up cloud data.

Summary of the invention

The deficiency of relatively poor for the real-time that overcomes existing scene modeling mode in bin, as can not to extract complex scene effectively three-dimensional dimension cloud data, the present invention proposes a kind of scene modeling method based on bin, this method utilizes degree of depth camera can obtain the three-dimensional information of scene fast, have advantages such as scene rebuilding is effective, real-time, and can effectively extract the three-dimensional dimension cloud data of complex scene.

The technical solution adopted for the present invention to solve the technical problems is:

A kind of complex scene three-dimensional rebuilding method based on bin, this method adopt degree of depth camera collection data, and described complex scene three-dimensional rebuilding method may further comprise the steps:

1), by three-dimensional scaling method, obtain inside and outside parameter and the distortion factor of degree of depth camera and colour imagery shot;

2), the three-dimensional information by degree of depth camera calculation of complex scene, according to principle of triangulation, following formula is arranged:

$D=\frac{b*f}{d}---\left(1\right)$

Wherein, D is depth information, unit: rice, described degree of depth camera comprises " infrared camera " and " infrared projection machine ", " infrared camera " and " infrared projection machine " is horizontal positioned, b is the length (unit: rice) of the horizontal base line between " infrared camera " and " infrared projection machine ", and f is that (unit: pixel), d is two parallax (units: pixel) between the camera for the focal length of " infrared camera "; A certain point coordinate is (X in the camera coordinate system _C, Y _C, Z _C) can have principle of triangulation to calculate acquisition;

3), point model is defined as bin, each value is an attribute of point model, and described point model comprises attributes such as normal vector and radius, and the some cloud is the set to the surface sampling of unknown curved surface, i.e. Q={q _i∈ R ³, the normal vector on the curved surface is defined as N={n _i∈ R ³, || n _i||=1}, wherein i ∈ 1,, m} sets up the KD-Tree data directory structure of acceleration search according to a cloud quantity;

4), each bin is done " inner, outside, intersect " relation test with another bin surface, specific as follows:

If α and β are the bin set of surface points model, S _αAnd S _βBe the point model surface that α and β cover, V _α, V _βBe S _α, S _βThe three dimensions entity that surrounds, ψ are the bin of some discrete points, and complex-curved point model is carried out Boolean calculation:

α∪β=<ψ|ψ∈α&&β _out&&ψ∈β&&α _out> （5）

α∩β=<ψ|ψ∈α&&β _in&&ψ∈β&&α _in> （6）

β-α=<ψ|ψ∈β&&α _out&&ψ∈α&&β _in> （7）

α-β=<ψ|ψ∈α&&β _out&&ψ∈β&&α _in> （8）

Wherein, α _In, α _Out, β _In, β _OutRefer to that respectively ψ is at point model S _αAnd S _βInboard and the outside;

5), to differentiating the bin for intersecting, that obtains bin earlier puts the letter neighborhood, the border of the most contiguous bin on the calculation level model, and in putting the letter neighborhood bin being resampled;

6), according to the Boolean calculation result of step 5), carry out the 3 D rendering of complex scene.

Advantage of the present invention is: can carry out collection and the processing of cloud data fast; Can finish the calculating of the inside and outside parameter of degree of depth camera fast; Can effectively extract the three-dimensional dimension cloud data of complex scene; Can finish the three-dimensional reconstruction of complex scene fast.

Description of drawings

Fig. 1 is complex scene three-dimensional reconstruction equipment synoptic diagram of the present invention.

Fig. 2 is complex scene three-dimensional rebuilding method process flow diagram of the present invention.

Fig. 3 is bin synoptic diagram of the present invention.

Embodiment

Below in conjunction with accompanying drawing the present invention is further described.

With reference to Fig. 1～Fig. 3, a kind of complex scene three-dimensional rebuilding method based on bin, this method adopts degree of depth camera collection data, and described complex scene three-dimensional rebuilding method may further comprise the steps:

1), by three-dimensional scaling method, obtain inside and outside parameter and the distortion factor of degree of depth camera and colour imagery shot;

2), the three-dimensional information by degree of depth camera calculation of complex scene, according to principle of triangulation, following formula is arranged:

$D=\frac{b*f}{d}---\left(1\right)$

Wherein, D is depth information, unit: rice, described degree of depth camera comprises " infrared camera " and " infrared projection machine ", " infrared camera " and " infrared projection machine " is horizontal positioned, b is the length of the horizontal base line between " infrared camera " and " infrared projection machine ", unit: rice, f is the focal length of " infrared camera ", unit: pixel, d is two parallaxes between the camera, unit: pixel; A certain point coordinate is (X in the camera coordinate system _C, Y _C, Z _C) can have principle of triangulation to calculate acquisition.

3), point model is defined as bin, each value is an attribute of point model, and this model has comprised attributes such as normal vector and radius.The point cloud is the set to the surface sampling of unknown curved surface, sets up the KD-Tree data directory structure of acceleration search according to a cloud quantity;

4), each bin is done " inner, outside, intersect " relation test with another bin surface, specific as follows:

If α and β are the bin set of surface points model, S _αAnd S _βBe the point model surface that α and β cover, V _α, V _βBe S _α, S _βThe three dimensions entity that surrounds, ψ are the bin of some discrete points, and complex-curved point model is carried out Boolean calculation:

α∪β=<ψ|ψ∈α&&β _out&&ψ∈β&&α _out> （5）

α∩β=<ψ|ψ∈α&&β _in&&ψ∈β&&α _in> （6）

β-α=<ψ|ψ∈β&&α _out&&ψ∈α&&β _in> （7）

α-β=<ψ|ψ∈α&&β _out&&ψ∈β&&α _in> （8）

Wherein, α _In, α _Out, β _In, β _OutRefer to that respectively ψ is at point model S _αAnd S _βInboard and the outside;

5), to differentiating the bin for intersecting, that obtains bin earlier puts the letter neighborhood, the border of the most contiguous bin on the calculation level model, and in putting the letter neighborhood bin being resampled;

6), according to the Boolean calculation result of step 5), carry out the 3 D rendering of complex scene.

In the present embodiment, as shown in Figure 1, use 8*8 gridiron pattern plane as scaling board, each tessellated length of side is 27mm.Be placed on about 1 meter of degree of depth camera and three-dimensional camera front, guarantee that the angle point in the gridiron pattern is captured by camera.By three-dimensional scaling method, obtain inside and outside parameter and the distortion factor of degree of depth camera and colour imagery shot.Degree of depth camera calculates in visual range by principle of triangulation, the coordinate of each point under the camera coordinate system in the scene.As shown in Figure 3, each point in the scene is represented with bin, and to each bin, done " inner, outside, the crossing " relation test with another bin surface.To differentiating the bin for intersecting, obtain the letter field of putting of bin earlier, the border of the most contiguous bin on the calculation level model, and in putting the letter field, bin is resampled.According to the Boolean calculation result, carry out the 3 D rendering of complex scene at last.In the present embodiment, by 4 different scenes are carried out three-dimensional reconstruction, experimental result such as table 1.

4 complex scene curve reestablishing complexities

Table 1.

Claims (1)

1. complex scene three-dimensional rebuilding method based on bin, this method adopt degree of depth camera collection data, and described complex scene three-dimensional rebuilding method may further comprise the steps:

1), by three-dimensional scaling method, obtain inside and outside parameter and the distortion factor of degree of depth camera and colour imagery shot;

2), the three-dimensional information by degree of depth camera calculation of complex scene, according to principle of triangulation, following formula is arranged:

$D=\frac{b*f}{d}---\left(1\right)$

Wherein, D is depth information, unit: rice, and described degree of depth camera comprises " infrared camera " and " infrared projection machine ", described " infrared camera " and " infrared projection machine " be horizontal positioned all; B is the length of the horizontal base line between " infrared camera " and " infrared projection machine ", unit: rice, and f is the focal length of " infrared camera ", unit: pixel, d is two parallaxes between the camera, unit: pixel; A certain point coordinate is (X in the camera coordinate system _C, Y _C, Z _C) can have principle of triangulation to calculate acquisition;

3), point model is defined as bin, each value is an attribute of point model, described point model comprises attributes such as normal vector and radius, and the some cloud is the set to the surface sampling of unknown curved surface, sets up the KD-Tree data directory structure of acceleration search according to a cloud quantity;

4), each bin is done " inner, outside, intersect " relation test with another bin surface, specific as follows:

If α and β are the bin set of surface points model, S _αAnd S _βBe the point model surface that α and β cover, V _α, V _βBe S _α, S _βThe three dimensions entity that surrounds, ψ are the bin of some discrete points, and complex-curved point model is carried out Boolean calculation:

α∪β=<ψ|ψ∈α&&β _out&&ψ∈β&&α _out> （5）

α∩β=<ψ|ψ∈α&&β _in&&ψ∈β&&α _in> （6）

β-α=<ψ|ψ∈β&&α _out&&ψ∈α&&β _in> （7）

α-β=<ψ|ψ∈α&&β _out&&ψ∈β&&α _in> （8）

Wherein, α _In, α _Out, β _In, β _OutRefer to that respectively ψ is at point model S _αAnd S _βInboard and the outside;

5), to differentiating the bin for intersecting, that obtains bin earlier puts the letter neighborhood, the border of the most contiguous bin on the calculation level model, and in putting the letter neighborhood bin being resampled;

6), according to the Boolean calculation result of step 5), carry out the 3 D rendering of complex scene.

Images