CN104331894A - Robot unstacking method based on binocular stereoscopic vision - Google Patents

Abstract

The invention discloses a robot unstacking method based on binocular stereoscopic vision. The robot unstacking method includes the following steps of: simultaneously obtaining images within views of left and right eyes by using a binocular camera fixed right above a working area, and positioning a target by using a shape template matching method to obtain a three-dimensional coordinate under the camera coordinate system; converting from the camera coordinate system into a robot base coordinate system; arranging depth information and unstacking in a layered manner according to the detected target coordinate. According to the robot unstacking method based on binocular stereoscopic vision, target identification and three-dimensional positioning of product in an unstacking area are achieved directly through collecting binocular images, and thereby automatic unstacking task of the product is achieved, and the problems that manual unstacking is large in labor intensity and low in efficiency can be effectively overcome.

Description

A kind of robot de-stacking method based on binocular stereo vision

Technical field

The present invention relates to a kind of robot de-stacking method based on binocular stereo vision.

Background technology

In product storage system, generally adopt robotization palletizing system both at home and abroad at present, product directly can be stacked to pallet, then be stored up to packed storehouse by fork truck.But when needing product warehouse-out to transport outward, usually still by manually taking out product from pallet, and then carry out entrucking.Artificial de-stacking labour intensity is large, inefficiency.Along with the fast development of China's economy and the raising of people's living standard, also progressively there is the phenomenon of labor shortage in China in recent years, and particularly slightly, stupid flexibe work(labour) power is short especially.

Therefore be badly in need of a kind of robotization of configuration, intelligentized destacking apparatus and method and solve the contradiction of producing with transport.

Summary of the invention

The present invention is in order to solve the problem, propose a kind of robot de-stacking method based on binocular stereo vision, the method by binocular imaging collection, utilize binocular image to realize the target identification of de-stacking district product and three-dimensional localization, and then the robotization de-stacking task realized product, effectively can solve artificial de-stacking labour intensity large, inefficient problem.

To achieve these goals, the present invention adopts following technical scheme:

Based on a robot de-stacking method for binocular stereo vision, comprise the following steps:

(1) the utilization binocular camera be fixed on directly over workspace obtains the image in the right and left eyes visual field simultaneously, and the method utilizing shape template to mate carries out target localization, obtains the three-dimensional coordinate under camera coordinates;

(2) conversion between camera coordinates system and robot base coordinate sys-tem is carried out;

(3) according to the coordinates of targets detected, carry out the arrangement of depth information, carry out layering de-stacking.

In described step (1), its concrete steps comprise:

(1-1) obtain the image in the right and left eyes visual field by the binocular camera be fixed on directly over workspace simultaneously, and noise reduction process is carried out to it;

(1-2) respectively the right and left eyes image of binocular camera is carried out to the coupling of Shape-based interpolation template, obtain right and left eyes image pixel coordinates;

(1-3) by principle of triangulation, right and left eyes image pixel coordinates is converted to the three-dimensional coordinate under camera coordinates.

In described step (1-2), the concrete grammar of single eye images coupling is:

A) set up the shape template of product under off-line condition, complete the registration of template, template definition is point set p _i=(x _i, y _i) ^twith each corresponding direction vector d that point is concentrated _i=(t _i, u _i) ^t, i=1,2 ... the number of n, n to be direction vector be non-zero points, direction vector is calculated by Sobel gradient operator;

B), during online template matches, be each some q=(x, y) in image to be searched ^tcalculate a direction vector e _{x, y}=(v _{x, y}, w _{x, y}) ^t;

C) at certain specified point place of searching image, the similarity measurement between template and searching image is defined as their direction vector normalized some sum s:

$s=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\frac{d_i^{\′T}{e}_{q+p_i^{\′}}}{\left|\right|d_i^{\′}\left|\right|\left|\right|e_{q+p_i^{\′}}\left|\right|}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\frac{t_i^{\′}{v}_{x+x_i^{\′},{y+y}_i^{\′}}+u_{\text{i}}^{\′}{w}_{x+x_i^{\′},{y+y}_i^{\′}}}{\sqrt{(t_i^{\′2}+u_i^{\′2})\sqrt{v_{x+x_i^{\′},y+y_i^{\′}}^2+w_{x+x_i^{\′},y+y_i^{\′}}^2}}}---\left(1\right)$

Wherein, p ' _i=(x ' _i, y ' _i) ^t, d ' _i=(t ' _i, u ' _i) ^tbe respectively p _i, d _icoordinate after rigid transformation or similarity transformation conversion and direction vector; If s is greater than certain threshold value, then namely the region of this some correspondence is considered to a target image;

D) target search: adopt pyramid structure to implement by the thick matching strategy to essence on binocular camera right and left eyes image respectively, and with yardstick step delta s and anglec of rotation step delta θ discretize search volume;

E) according to precision needs, in steps d) target proximity that detects, adopt and carry out sub-pixel positioning based on the method edge point of Facet model, and then adopt ICP (Iterative Closest Point) algorithm to refine matching result, to obtain high-precision positioning precision.

The concrete grammar of described step (1-3) is:

According to the pixel coordinate (u of target in right and left eyes image _l, v _l) and (u _r, v _r), its three-dimensional coordinate under camera coordinates can be obtained by principle of triangulation;

$X=\frac{b}{d}C---\left(2\right)$

Wherein, X=[X Y Z] ^tfor target three-dimensional coordinate under camera coordinates system, d=u _l-u _r, b is parallax range, C=[u _lv f] ^t, camera used is here parallel optical axis binocular camera, and left and right camera parameter is identical, therefore has v _l=v _r=v, f _l=f _r=f.

The concrete steps of described step (2) comprising:

(2-1) calibrated reference is placed at random n position of random selecting in robot working space, guarantees that the position chosen will in the coverage of camera;

(2-2) three-dimensional coordinate (X of object of reference in camera coordinates system is calculated in each position by foregoing visual locating module _c, Y _c, Z _c);

(2-3) allow robot hand clamp an end sharp shaped material (as pencil), manipulation robot makes sharp shaped material end touch the center of object of reference, from the three-dimensional coordinate (X of the known object of reference of the reading of robot controller under robot coordinate system _r, Y _r, Z _r);

(2-4) because camera is fixed on the top of robot work region, therefore (X _c, Y _c, Z _c) and (X _r, Y _r, Z _r) there is following relation:

$\left\{\begin{array}{c}{X}_r=m_{11}{X}_c+m_{12}{Y}_c+m_{13}{Z}_c+P_x\\ Y_r=m_{21}{X}_c+m_{22}{Y}_c+m_{23}{Z}_c+P_y\\ Z_r=m_{31}{X}_c+m_{32}{Y}_c+m_{33}{Z}_c+P_z\end{array}\right.---\left(3\right)$

Note $R=\left[\begin{array}{ccc}{m}_{11}& m_{12}& m_{13}\\ m_{21}& m_{22}& m_{23}\\ m_{31}& m_{32}& m_{33}\end{array}\right],T=\left[\begin{array}{c}{P}_x\\ P_y\\ P_z\end{array}\right];$ R and T represents rotation matrix and translation matrix respectively; N the data of step (2-2) and (2-3) gained are brought into equation (3) 3n system of linear equations can be obtained;

(2-5) utilize least square method to solve system of linear equations and solve R and T.

The concrete grammar of described step (3) comprising:

(3-1) for n the object detected, first ascending arrangement is carried out to the depth information of each target, then carry out judging whether at one deck;

(3-2) to after object layering, sort according to a certain order to the object being in top layer, determine last de-stacking order, robot implements de-stacking to object;

(3-3) after robot completes the de-stacking task of this layer, return step (1), again drive binocular camera to carry out binocular image acquisition, carry out successively.

The detailed step of described step (3-1) is: for n the object { OBJ detected ₁, OBJ ₂..., OBJ _n, its coordinate is respectively (X ₁, Y ₁, Z ₁), (X ₂, Y ₂, Z ₂) ... (X _n, Y _n, Z _n), first ascending arrangement is carried out to the depth information of each target, then judge.If | Z _i-Z _j|≤λ H (i=1,2 ... n, j > i), then i-th is same layer object with jth, otherwise is that two objects are in different layers.Wherein H is object thickness, and λ is coefficient.

The concrete steps of described step (3-2) are: after object layering, sort according to a certain order to the object being in top layer, determine last de-stacking order: { OBJ ₁, OBJ ₂... OBJ _m, and its three-dimensional coordinate is passed to Robot Motion Controller, robot controller drive machines robot end actuator motions captures to above-mentioned position, and after providing off-position, can realize the de-stacking task to object.

Beneficial effect of the present invention is:

(1) the robot de-stacking system based on binocular stereo vision of the present invention's proposition, its directly by binocular imaging collection, utilize binocular image to realize the target identification of de-stacking district product and three-dimensional localization, and then the robotization de-stacking task realized product, effectively can solve artificial de-stacking labour intensity large, inefficient problem;

(2) the robot de-stacking system based on binocular stereo vision of the present invention's proposition, adopts shape template matching process to carry out the Stereo matching of target detection and left images character pair point; This algorithm only make use of the gradient information on image, is not easily masked, clutter and non-linear illumination effect, compares traditional matching process relevant based on gray scale and has more practicality and robustness;

(3) the robot de-stacking system based on stereoscopic vision of the present invention's proposition, without the need to special product orientation device, can carry out de-stacking process to the product piling up neatly arbitrarily shape; Only need add new model data in software processing module can realize, to the de-stacking task of dissimilar product, realizing the flexibility of de-stacking operation, intellectuality, thus there is good popularizing application prospect.

Accompanying drawing explanation

Fig. 1 is that system hardware forms schematic diagram.

Fig. 2 is de-stacking system flowchart.

Fig. 3 is de-stacking algorithm schematic diagram.

Embodiment:

Below in conjunction with accompanying drawing and embodiment, the invention will be further described.

The present invention proposes a kind of robot de-stacking method based on binocular stereo vision.It calculates mechanism primarily of binocular camera, industrial robot and motion controller, image procossing and robot controlling and becomes, and structure as shown in Figure 1.Be provided with product de-stacking district and collecting region around industrial robot, binocular camera is fixed on directly over de-stacking district, is mainly used to sensing external environment, and the scene image data of acquisition is sent to computing machine carries out post-processed in real time; Industrial robot, as main topworks, is used for realizing the crawl of product or absorption (determining according to corresponding end effector) and puts it to the task of corresponding product collecting region.Software section mainly comprises image processing module and motion planning and robot control module, and the two operates in same computing machine, realizes thread communication to each other by message queue mode.The binocular image of Real-time Collection and registered template are carried out the template matches of Shape-based interpolation by image processing module, judge in binocular image, whether to occur corresponding product image, when all there is registered product image in and if only if binocular image, just calculate according to principle of triangulation the three-dimensional coordinate of product in camera coordinates system entering the binocular camera visual field.Then the transformational relation that the camera coordinates obtained according to off-line is tied to robot base coordinate sys-tem obtains its three-dimensional coordinate in robot coordinate system, and by thread communication, the three-dimensional coordinate of product in robot coordinate system is delivered to motion planning and robot control module, after motion-control module receives three-dimensional coordinate, operational order is sent according to predetermined de-stacking strategy, drive machines robot end actuator successively successively captures or draws product and place it in corresponding product-collecting district, realizes the de-stacking task capturing-place.

The workflow of de-stacking system as shown in Figure 2, can be divided into off-line to prepare and two stages of on-line implement.

One, the off-line preparatory stage

1. binocular camera is demarcated

First the inside and outside parameter of method to single camera such as Zhang Zhengyou method are adopted to demarcate; Then the relative tertiary location relation of optimized algorithm to two cameras is adopted to demarcate.

2. template registration

For the template image of product, first need the To Template setting up Shape-based interpolation.Template definition is point set p _i=(x _i, y _i) ^twith each corresponding direction vector d that point is concentrated _i=(t _i, u _i) ^t, i=1,2 ... the number of n, n to be direction vector the be point of non-zero, direction vector can be calculated by Sobel gradient operator.

3. hand and eye calibrating

Adopt picture on surface to be easy to the planished sheet of identification and location as calibrated reference, demarcating steps is as follows:

(1) object of reference is placed at random n position of random selecting in robot working space;

(2) three-dimensional coordinate (X of object of reference in camera coordinates system is calculated in each position by vision locating module _c, Y _c, Z _c);

(3) allow robot hand clamp an end sharp shaped material (as pencil), manipulation robot makes sharp shaped material end touch the center of object of reference, from the three-dimensional coordinate (X of the known object of reference of the reading of robot controller under robot coordinate system _r, Y _r, Z _r);

(4) because camera is fixed on the top of robot work region, therefore (X _c, Y _c, Z _c) and (X _r, Y _r, Z _r) there is following relation:

$\left\{\begin{array}{c}{X}_r=m_{11}{X}_c+m_{12}{Y}_c+m_{13}{Z}_c+P_x\\ Y_r=m_{21}{X}_c+m_{22}{Y}_c+m_{23}{Z}_c+P_y\\ Z_r=m_{31}{X}_c+m_{32}{Y}_c+m_{33}{Z}_c+P_z\end{array}\right.---\left(1\right)$

Note $R=\left[\begin{array}{ccc}{m}_{11}& m_{12}& m_{13}\\ m_{21}& m_{22}& m_{23}\\ m_{31}& m_{32}& m_{33}\end{array}\right],T=\left[\begin{array}{c}{P}_x\\ P_y\\ P_z\end{array}\right].$ R and T represents rotation and translation matrix respectively.The data of n the point by step (2) and (3) gained are brought into equation (1) 3n system of linear equations can be obtained.

(5) solve system of linear equations by least square method and can solve R and T.

Two, the on-line implement stage

1. start three-dimensional binocular camera to operative scenario synchronous acquisition image.

2. Image semantic classification: carry out low-pass filtering to input picture, rejects the noise in image to the full extent, then according to the calibration result of binocular camera, corrects the radial distortion of camera lens, and makes left images longitudinally alignment.

3. on left images, implementation goal detects respectively, and concrete steps are as follows:

(1) be each some q=(x, y) in searching image ^tcalculate a direction vector e _{x, y}=(v _{x, y}, w _{x, y}) ^t.

(2) at certain specified point place of searching image, the similarity s between calculation template and searching image:

$s=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\frac{d_i^{\′T}{e}_{q+p_i^{\′}}}{\left|\right|d_i^{\′}\left|\right|\left|\right|e_{q+p_i^{\′}}\left|\right|}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\frac{t_i^{\′}{v}_{x+x_i^{\′},{y+y}_i^{\′}}+u_{\text{i}}^{\′}{w}_{x+x_i^{\′},{y+y}_i^{\′}}}{\sqrt{(t_i^{\′2}+u_i^{\′2})\sqrt{v_{x+x_i^{\′},y+y_i^{\′}}^2+w_{x+x_i^{\′},y+y_i^{\′}}^2}}}---\left(2\right)$

Wherein, p ' _i=(x ' _i, y ' _i) ^t, d ' _i=(t ' _i, u ' _i) ^tbe respectively p _i, d _icoordinate after rigid transformation or similarity transformation conversion and direction vector.If s is greater than certain threshold value, then the region of this some correspondence can be considered to a target image.

(3) target search: adopt pyramid structure to implement by the thick matching strategy to essence on right and left eyes image respectively, and with yardstick step delta s and anglec of rotation step delta θ discretize search volume.

(4) according to precision needs, in the target proximity that step (3) detects, the method edge point based on Facet model can be adopted to carry out sub-pixel positioning, and then adopt ICP (Iterative Closest Point) algorithm to refine matching result, to obtain high-precision positioning precision.

4. draw the unique point coordinate (u of target in left images by above-mentioned steps _l, v _l) and (u _r, v _r), its three-dimensional space position can be obtained by principle of triangulation wherein X=[X Y Z] ^tfor target three-dimensional coordinate under camera coordinates system, d=u _l-u _r, b is parallax range, C=[u _lv f] ^t, here the left and right camera parameter of binocular camera used is identical and be parallel optical axis, therefore has v _l=v _r=v, f _l=f _r=f.

5. be its three-dimensional coordinate in robot base coordinate sys-tem target in the coordinate transformation of camera coordinates system by formula (1).

6. robot implements de-stacking operation: first carry out layering according to depth information to target, then capture the object being in top layer or draw operation, and puts it into corresponding product-collecting district.Concrete steps are as follows:

(1) judge with layer object: n the object OBJ that vision module is detected ₁, OBJ ₂..., OBJ _n, its coordinate is respectively (X ₁, Y ₁, Z ₁), (X ₂, Y ₂, Z ₂) ... (X _n, Y _n, Z _n), first ascending arrangement is carried out to the depth information of each target, then judge.If | Z _i-Z _j|≤λ H (i=1,2 ... n, j > i), then i-th is same layer object with jth, otherwise is that two objects are in different layers.Wherein H is object thickness, and λ is coefficient.

(2) to after object layering, the object being in top layer is sorted according to a certain order, determine last de-stacking order: { OBJ ₁, OBJ ₂... OBJ _m), and its three-dimensional coordinate is passed to Robot Motion Controller, robot controller drive machines robot end actuator motions captures to above-mentioned position, and after providing off-position, can realize the de-stacking task to object.

(3) after robot completes the de-stacking task of this layer, again drive binocular camera to carry out binocular image acquisition, obtain the three-dimensional coordinate of camera fields of view internal object thing.Drive machines robot end moves to assigned address again, completes this de-stacking, the like complete de-stacking task to whole workspace target.

Thus, based in the industrial robot de-stacking system of binocular camera, vision processing module and motion-control module work coordinated with each other, vision processing module carries out target identification and three-dimensional localization based on the product of the binocular image collected to current de-stacking district; And motion-control module sends corresponding operational order based on this positioning result, the product in current de-stacking district is carried to corresponding product-collecting district by drive machines robot end actuator; So move in circles.Show through experimental verification, the robot de-stacking method based on binocular stereo vision that the present invention proposes can surely, standard, realize the product sorting process in de-stacking district soon.

By reference to the accompanying drawings the specific embodiment of the present invention is described although above-mentioned; but not limiting the scope of the invention; one of ordinary skill in the art should be understood that; on the basis of technical scheme of the present invention, those skilled in the art do not need to pay various amendment or distortion that creative work can make still within protection scope of the present invention.

Claims (8)

1., based on a robot de-stacking method for binocular stereo vision, it is characterized in that: comprise the following steps:

(1) the utilization binocular camera be fixed on directly over workspace obtains the image in the right and left eyes visual field simultaneously, and the method utilizing shape template to mate carries out target localization, obtains the three-dimensional coordinate under camera coordinates;

(2) conversion between camera coordinates system and robot base coordinate sys-tem is carried out;

(3) according to the coordinates of targets detected, carry out the arrangement of depth information, the object being in top layer is sorted, determine that last de-stacking order carries out layering de-stacking.

2. a kind of robot de-stacking method based on binocular stereo vision as claimed in claim 1, is characterized in that: in described step (1), its concrete steps comprise:

(1-1) obtain the image in the right and left eyes visual field by the binocular camera be fixed on directly over workspace simultaneously, and noise reduction process is carried out to it;

(1-2) respectively the right and left eyes image of binocular camera is carried out to the coupling of Shape-based interpolation template, obtain right and left eyes image pixel coordinates;

(1-3) by principle of triangulation, right and left eyes image pixel coordinates is converted to the three-dimensional coordinate under camera coordinates.

3. a kind of robot de-stacking method based on binocular stereo vision as claimed in claim 2, is characterized in that: in described step (1-2), and the concrete grammar of single eye images coupling is:

A) set up the shape template of product under off-line condition, complete the registration of template, template definition is point set p _i=(x _i, y _i) ^twith each corresponding direction vector d that point is concentrated _i=(t _i, u _i) ^t, i=1,2 ... the number of n, n to be direction vector be non-zero points, direction vector is calculated by Sobel gradient operator;

B), during online template matches, be each some q=(x, y) in image to be searched ^tcalculate a direction vector e _{x, y}=(v _{x, y}, w _{x, y}) ^t;

C) at certain specified point place of searching image, the similarity measurement between template and searching image is defined as their direction vector normalized some sum s:

$s=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\frac{d_i^{\′T}{e}_{q+p_i^{\′}}}{\left|\right|d_i^{\′}\left|\right|\left|\right|e_{q+p_i^{\′}}\left|\right|}=\frac{1}{n}\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\frac{t_i^{\′}{v}_{x+x_i^{\′},y+y_i^{\′}}+u_i^{\′}{w}_{x+x_i^{\′},y+y_i^{\′}}}{\sqrt{(t_i^{\′2}+u_i^{\′2})\sqrt{v_{x+x_i^{\′},y+y_i^{\′}}^2+w_{x+x_i^{\′},y+y_i^{\′}}^2}}}---\left(1\right)$

Wherein, p' _i=(x' _i, y' _i) ^t, d' _i=(t' _i, u' _i) ^tbe respectively p _i, d _icoordinate after rigid transformation or similarity transformation conversion and direction vector; If s is greater than certain threshold value, then namely the region of this some correspondence is considered to a target image;

D) target search: adopt pyramid structure to implement by the thick matching strategy to essence on binocular camera right and left eyes image respectively, and with yardstick step delta s and anglec of rotation step delta θ discretize search volume;

E) according to precision needs, in steps d) target proximity that detects, adopt and carry out sub-pixel positioning based on the method edge point of Facet model, and then adopt ICP (Iterative Closest Point) algorithm to refine matching result, to obtain high-precision positioning precision.

4. a kind of robot de-stacking method based on binocular stereo vision as claimed in claim 2, is characterized in that: the concrete grammar of described step (1-3) is:

According to the pixel coordinate (u of target in right and left eyes image _l, v _l) and (u _r, v _r), its three-dimensional coordinate under camera coordinates can be obtained by principle of triangulation;

$X=\frac{b}{c}C---\left(2\right)$

Wherein, X=[X Y Z] ^tfor target three-dimensional coordinate under camera coordinates system, d=u _l-u _r, b is parallax range, C=[u _lv f] ^t, camera used is here parallel optical axis binocular camera, and left and right camera parameter is identical, therefore has v _l=v _r=v, f _l=f _r=f.

5. a kind of robot de-stacking method based on binocular stereo vision as claimed in claim 1, is characterized in that: described step

(2) concrete steps comprise:

(2-1) calibrated reference is placed at random n position of random selecting in robot working space, guarantees that the position chosen will in the coverage of camera;

(2-2) three-dimensional coordinate (X of object of reference in camera coordinates system is calculated in each position by foregoing visual locating module _c, Y _c, Z _c);

(2-3) allow robot hand clamp an end sharp shaped material, manipulation robot makes sharp shaped material end touch the center of object of reference, from the three-dimensional coordinate (X of the known object of reference of the reading of robot controller under robot coordinate system _r, Y _r, Z _r);

(2-4) because camera is fixed on the top of robot work region, therefore (X _c, Y _c, Z _c) and (X _r, Y _r, Z _r) there is following relation:

$\left\{\begin{array}{c}{X}_r=m_{11}{X}_c+m_{12}{Y}_c+m_{13}{Z}_c+P_x\\ Y_r=m_{21}{X}_c+m_{22}{Y}_c+m_{23}{Z}_c+P_y\\ Z_r=m_{31}{X}_c+m_{32}{Y}_c+m_{33}{Z}_c+P_z\end{array}\right.---\left(3\right)$

Note $R=\left[\begin{array}{ccc}{m}_{11}& m_{12}& m_{13}\\ m_{21}& m_{22}& m_{23}\\ m_{31}& m_{32}& m_{33}\end{array}\right],T=\left[\begin{array}{c}{P}_x\\ P_y\\ P_z\end{array}\right];$ R and T represents rotation matrix and translation matrix respectively; N the data of step (2-2) and (2-3) gained bring into equation (3) 3n system of linear equations;

(2-5) utilize least square method to solve system of linear equations and solve R and T.

6. a kind of robot de-stacking method based on binocular stereo vision as claimed in claim 1, is characterized in that: the concrete grammar of described step (3) comprising:

(3-1) for n the object detected, first ascending arrangement is carried out to the depth information of each target, then carry out judging whether at one deck;

(3-2) to after object layering, sort to the object being in top layer, determine last de-stacking order, robot implements de-stacking to object;

(3-3) after robot completes the de-stacking task of this layer, return step (1), again drive binocular camera to carry out binocular image acquisition, carry out successively.

7. a kind of robot de-stacking method based on binocular stereo vision as claimed in claim 6, is characterized in that: the detailed step of described step (3-1) is: for n the object { OBJ detected ₁, OBJ ₂..., OBJ _n, its coordinate is respectively (X ₁, Y ₁, Z ₁), (X ₂, Y ₂, Z ₂) ... (X _n, Y _n, Z _n), first ascending arrangement is carried out to the depth information of each target, then judge; If | Z _i-Z _j|≤λ H (i=1,2 ... n, j > i), then i-th is same layer object with jth, otherwise is that two objects are in different layers.Wherein H is object thickness, and λ is coefficient.

8. a kind of robot de-stacking method based on binocular stereo vision as claimed in claim 6, it is characterized in that: the concrete steps of described step (3-2) are: after object layering, the object being in top layer is sorted according to a certain order, determines last de-stacking order: { OBJ ₁, OBJ ₂... OBJ _m, and its three-dimensional coordinate is passed to Robot Motion Controller, robot controller drive machines robot end actuator motions captures to above-mentioned position, and after providing off-position, can realize the de-stacking task to object.