CN103514627A - Adaptive dynamic programming matching method for colored structured light scanning system - Google Patents

Abstract

The invention is an adaptive dynamic programming matching method for a color-coded structured light scanning system, belongs to the technical field of three-dimensional structured light scanning and aims to solve the problem of matching projected coded stripes and stripes acquired by a camera. The invention provides a method of adaptive matching by using stripe neighborhood information to improve the accuracy and the precision of matching so as to achieve a better three-dimensional reconstruction effect. Key points of the technical scheme for solving the problem are as follows: special de Bruijn sequence stripe structured light is selected as a projected image, namely, at least two colored channels of adjacent stripe colors are different; and the weight is adjusted by making use of local color difference information and the distance between stripes so as to design an objective function. The invention mainly aims to improve the matching rate between coded stripes and acquired stripes and prepare for subsequent reconstruction of precisely-calculated three-dimensional coordinates and a smoother surface. The system can be used in a variety of colors of objects and can be used for scanning in a non-darkroom environment.

Description

A kind of self-adaptation dynamic programming matching process for color structure photo-scanning system

One, technical field:

The present invention relates to structural light three-dimensional scanning technique, relate in particular to the self-adaptation dynamic programming matching process for color coding structural light scanning system.

Two, background technology:

Three-dimensional reconstruction is the focus in computer vision research field, and this technology is widely used in the various fields such as reverse-engineering, man-machine interaction, Entertainment, has very strong practicality.How high resolving power, high real-time and the study hotspot that obtains the 3D data Shi Gai field of testee under physical environment.This technology can be applicable to the various fields such as reverse-engineering, man-machine interaction, Entertainment, has very strong practicality.How to obtain more accurately and real-time the target of the 3D data Shi Gai field personage research of object.

3D measuring method is divided into contact and contactless two classes, and Typical Representative is three dimensional coordinate measuring machine, and this metering system be take precision optical machinery as basis, and measuring accuracy can be up to micron order.But the method owing to can not measure soft object, cost is high, reconstruction speed is slow, efficiency is low, so the method for contact can not meet accurately, measure efficiently requirement.Comparatively speaking, non-contact measurement solves above-mentioned various defect, and continuity is measured in, maintenance fast with measuring speed, and the advantages such as easy digitizing become an important developing direction.

Contactless measurement is divided into again passive type and active two kinds.Wherein, passive type three-dimensional measurement, does not need extra light source [2], and illumination is provided by testee natural lighting condition around.Passive type commercial measurement ratio of precision is lower, and calculated amount is larger, cannot reach high-precision standard; And active method is to point to the pattern that testee projection has set, there is deformation because of the impact of testee shape in this pattern, and the deformation occurring by matched patterns obtains the three-dimensional information of object; Be characterized in that measuring accuracy is high, simple in structure, be easy to realize.

Contactless measurement research concentrates in the selection and measuring method of medium.Transfer Medium mainly contains laser, sound wave, electromagnetic field, infrared light etc., and wherein measuring method is comparatively fast developed because operating distance is large, measuring accuracy is high, can measure the advantages such as non-metallic objectsit is not a metal object.In numerous measuring methods, coded structured light with higher measuring accuracy, the outstanding advantage that is easy to realize becomes the focus of research.Structured light scanning calculates the degree of depth of object by the triangle relation of projector equipment, object and camera.

Structured light scanning technique is mainly divided into two large classes.The first kind is time encoding technology, by projecting a series of structured light, thereby obtain high accuracy three-dimensional cloud data, but these class methods can only scan static scene.Another kind of is space encoding technology, and the advantage of this class technology is to obtain three dimensional point cloud by the projection of primary structure light, and wherein the most frequently used is De Bruijn (DB) sequence.DB sequence is one group of pseudo-random sequence, has huge number of codewords and high linear complexity, is to have the most macrocyclic shift-register sequence.DB sequence is demarcated by two key elements (n, m), and it is by the m kind order encoding of tactic n kind eigenwert, and the sequence that wherein each substring length is m is only used once.The last color of passing through three stripeds of each coupling of process of reconstruction then finds unique position in de Bruijn sequence.Thereby can obtain each striped and the angle information of taking camera, thereby calculate object dimensional information, this sequence has extraordinary automatic correlation function.As body surface reflectivity, surround lighting etc., the decoding of structure light coding is had to very large interference.In order to address this problem, document " Fechteler, P., Eisert, P.:Adaptive color classification for structured light systems.Computer Vision, IET3 (2009) 49-59 " in proposed by the method for K-Means mean cluster adaptively by color classification, thereby can be under physical environment and without the kind of judging comparatively exactly color in color correction ground situation in advance.In the end in matching process, the mode that adopts probability model to combine with dynamic programming, accuracy rate is not satisfactory, has affected the precision of three-dimensional point cloud.

For the problems referred to above, the present invention proposes a kind of dynamic programming algorithm for improving grating matching accuracy rate.This method is by utilization fixedly color and the range information of striped in neighborhood window, and carrys out design object function by adjusting the weight of neighborhood stripe information, obtains better accuracy rate, up to more than 98%, thereby can obtain good reconstruction effect.

Three, summary of the invention:

The technical matters that the present invention solves is: the not high situation of accuracy rate for color coding structural light scanning system at coupling projection striped and shooting striped.The present invention proposes a kind of objective function that utilizes striped neighborhood information to design dynamic programming, thereby obtain good three-dimensional reconstruction effect.

In color coding structural light scanning system, as shown in Figure 1,102 projector projects 103 structured lights, structured light adopts 6 kinds of colors, and every Seed Sequences length is 4; Every kind of color is all heavy shade, i.e. red, green, blue, bluish-green, fuchsin, Huang.Put on number to these 6 kinds of colors, and add a restrictive condition, it is different that the RGB passage of adjacent two colors has two passages at least.With 6 kinds of colors of graphical presentation above can adjacent principle as chart 2.The number of sequence striped is 162 like this, between each strip encoding, by black interval, has just formed 103 structured lights of projection.Structured light sequence chart as shown in Figure 3.When ray cast is to 104 body surfaces, through body surface reflection, by 101 cameras, catch and obtain colour picture.Striped color is judged through means clustering algorithm, the distribution by the colouring information on stripe centerline at rgb space is expressed as cluster straight line by red, green, blue, bluish-green, fuchsin, yellow six kinds of colors in three dimensions, and straight-line equation is: o _c+ xr _c, c ∈ { r, g, b, c, m, g} wherein.Calculate each o'clock to the distance of six straight lines, find distance minimum, be the classification of certain color.And using this minor increment and relational expression to the distance sum of six straight lines as one of parameter of subsequent dynamic planning.

Formula is as follows:

$p_{i,\mathrm{color}}=\frac{{[d(p_i,g_{\mathrm{color}})+\mathrm{\ϵ}]}^{-1}}{{\mathrm{\Σ}}_{c=1}^6{[d(p_i,g_c)+\mathrm{\ϵ}]}^{-1}}---\left(1\right)$

Wherein, ε is for fear of makeing mistakes and the just minimum constant of affix except zero.Utilize this parameter and neighborhood design object function.

First near 3 stripeds that we choose the striped of Shi center are neighborhood window, choose the first from left right side two, adjacent 4 stripeds are one group of research object, this is only in complete sequence, to occur once because the DB sequence of projection is every four adjacent striped combinations, so can guarantee that the score of four adjacent sequences that are combined into is the highest, thereby easily match.

$\mathrm{Score}\left(i\right)={\mathrm{\Σ}}_{j=1}^4{w}_p(i,j)\ln w_d(i,j)\frac{P_{i,c}}{P_{i,\stackrel{\‾}{c}}}---\left(2\right)$

Wherein, w _d(i, j) represents the weight of adjacent stripes color distortion, w _p(i, j) represents the distance difference of adjacent stripes.Account form is as follows:

$w_d(i,j)=\exp (-\frac{P_{j,c}}{P_{i,c}})---\left(3\right)$

$w_d(i,j)=k\·\frac{\mathrm{\γ}}{|j-i|}$

(4)

$k=\left\{\begin{array}{c}1(j\∈\mathrm{match})\\ 0(j\∉\mathrm{match})\end{array}\right.$

Utilize this objective function to set up accumulation matrix, and follow the tracks of maximal value with Metzler matrix.Arthmetic statement is as Fig. 2.Metzler matrix is convenient solves optimal value with back-track algorithm.By the method, obtaining grating matching rate reaches more than 99%.

Four, accompanying drawing explanation:

Fig. 1 is color coding structural light scanning system Organization Chart

Fig. 2 is the false code description figure of dynamic programming algorithm

Fig. 3 is specific embodiment of the invention process

Five, specific implementation method:

The following describes specific embodiment of the invention process, as shown in Figure 2:

Each Streak parameters of step 200 is obtained: after unsupervised K-Means means clustering algorithm, obtained the parameter of each striped by formula (1).The probable value that the color of each striped is judged into certain color like this can be used as the parameter value of next step dynamic programming.

Step 201 is calculated the color distortion of neighborhood striped: utilize the parameter information that obtains each striped above, by formula (3), calculating neighborhood striped can affect the weight foundation of center striped on the color distortion of center striped as it.

Step 202 is calculated the distance difference of neighborhood striped: by formula (4), calculated the distance difference of neighborhood striped and center striped, as it, can affect another weight foundation of center striped.

Step 203 design object function: utilize the weight information that difference that above-mentioned two steps calculate can be on the impact of center striped as this adjacent stripes, design object function, i.e. formula (2).

Step 204 is set up Metzler matrix: the present invention utilizes Metzler matrix to record the optimal value that the objective function in dynamic programming process can obtain, and uses simple numerical method, records contingent three kinds of situations.This method can facilitate follow-up trace-back process.Algorithm as shown in Figure 2.

The bottom-up optimum solution of obtaining of step 205: this step is the final step in dynamic programming conventional algorithm.Bottom-up recall find common subsequence, thereby find optimum solution.

Claims (1)

1. for a self-adaptation dynamic programming matching process for color structure photo-scanning system, its feature comprises following step:

A. each Streak parameters is obtained: after unsupervised K-Means means clustering algorithm, by each striped to cluster centre linear distance ratio the parameter as each striped.The probable value that the color of each striped is judged into certain color like this can be used as the parameter value of next step dynamic programming.

B. calculate the color distortion of neighborhood striped: utilize the parameter information that obtains each striped above, by neighborhood striped, on the color distortion of center striped, as it, can affect the weight foundation of center striped.

C. calculate the distance difference of neighborhood striped: utilize the parameter information obtaining in the first step, by neighborhood striped, on the distance difference of center striped, as it, can affect another weight foundation of center striped.

D. design object function: utilize the weight information that difference that above-mentioned two steps calculate can be on the impact of center striped as this adjacent stripes, design object function.

E. set up Metzler matrix: the present invention utilizes Metzler matrix to record the optimal value that the objective function in dynamic programming process can obtain, and uses simple numerical method, record contingent three kinds of situations.

Images