Summary of the invention
Problem to be solved by this invention
Even by existing method, can obtain correct reference point, but still need a large amount of calculating and processing time.Therefore, the target object that moves is photographed by two video cameras, and detect in real time three-dimensional shape method since the processing time grow the difficulty that becomes.
The present invention considers present actual conditions, has proposed the method about the correspondence of edge image reference point by the simple Mi Shixian of processing, and three-dimensional shape pick-up unit and the three-dimensional shape detection method that can handle in real time are provided.
The method of dealing with problems
In order to reach above-mentioned purpose of the present invention, three-dimensional shape pick-up unit of the present invention need possess, and from least two camera positions target object is photographed; Load onto at least 1 cover shooting instrument that has imaging apparatus and camera lens on each position; On the direction that connects at least two above-mentioned camera positions, load onto the instrument that the optical axis that makes above-mentioned shooting instrument is done minute movement; And many pieces of images that photograph when optical axis is moved carry out Flame Image Process, promptly utilize the method for edge extracting to generate separately edge image, more above-mentioned each edge image, the Flame Image Process instrument of the corresponding edge of detection target object.
The designated photograph position here can be at least one direction along horizontal direction, vertical direction, vergence direction, perhaps a plurality of direction configuration shooting instruments.
And, preferably have the Move tool that can move to video camera the designated photograph position.
Perhaps, at least two video cameras are pre-configured on each camera position of appointment.
And when using plural shooting instrument, each shooting instrument preferably takes place to make a video recording the identical minute movement and the same period simultaneously.
Moreover the instrument of optical axis fine motion can make above-mentioned camera lens do minute movement, also can make shooting instrument integral body do minute movement.
Moreover optical axis fine motion instrument can be made of among piezoelectric element, supermagnetic device, static gearing, oil hydraulic actuator, pneumatic actuator, electromagnet, optics gearing, marmem, the macromolecule gearing any one.
The minute movement of the optical axis fine motion instrument here can move along the normal direction of the optical axis of above-mentioned shooting instrument is parallel.
In addition, the minute movement of optical axis also can be to be the gyration at center with certain specified point.Here, the angle of revolution of gyration, for the amplitude that moves of the image of the target object that approaches infinity on imaging apparatus, with the width of 0-3 pixel for well.
And then, three-dimensional shape detection method of the present invention, need possess following process: at least two designated photograph positions, configuration has the process of at least 1 shooting instrument of imaging apparatus and camera lens to the target object photography, in photographic process, make the optical axis of above-mentioned shooting instrument carry out the process of minute movement simultaneously, optical axis is carried out Flame Image Process by the image that the shooting instrument of minute movement photographs, utilize the method for edge extracting to obtain the process of edge image separately, each above-mentioned edge image is compared, detect the testing process of corresponding edge.
The generative process of the edge image is here done standardization after can doing the difference processing of deep or light value or chroma value to the corresponding pixel of above-mentioned each image, and also can be done 2 values and handle.
Moreover the corresponding edge testing process can be a benchmark with the border width and the length of above-mentioned edge image, the edge of the edge image of corresponding above-mentioned plural designated photograph position.
And the process that the corresponding edge is detected also can be used the images match method.
And the process that the corresponding edge is detected can also be benchmark with the direction of visual lines of target object.In this occasion, also can then each edge image be compared the edge transition of the edge image that generates for the direction of visual lines with object-oriented object is the specified value width of benchmark.
The invention effect
In three-dimensional shape pick-up unit of the present invention and three-dimensional shape detection method, because generated the edge image that has range information, so need only note the edge length and the width of edge image, just can detect the corresponding edge easily, and, processing time shortens, and has the advantage that can handle in real time.
Human eyeball is unconsciously carrying out very small high-frequency vibration---tremor, and very little smooth motion---drift also has motion---three kinds of microvibrations of micro saccade of straight line.About these microvibrations what kind of influence visual recognition is had, a large amount of research has all been carried out in physiology and engineering science two aspects.Be considered to be used for detecting the motion of visual object distance.In these human eye movements, we especially pay attention to rectilinear motion, have developed the high-precision three-dimensional shape pick-up unit that the high speed of same action principle is arranged with rectilinear motion.
Embodiment
Below be legend and explanation in order to realize optimal morphology of the present invention.Fig. 1 is the skeleton diagram of the camera part of three-dimensional shape pick-up unit of the present invention.As shown in the figure, video camera part 10 is made of camera lens 1 and imaging apparatus 2.Camera lens both can be a short-focus lens, also can be the zoom lens, both can be wide-angle lens, also can be narrow angle mirror head.These can be according to the application target of three-dimensional shape pick-up unit, with distance of target object etc., changes and selects.Imaging apparatus 2 is the components of photo-electric conversion of CCD or CMOS, by camera lens imaging on imaging apparatus, is image transformation electric signal by imaging apparatus.Then, by being connected on the image processing section 3 on the imaging apparatus 2, use the electric signal that obtains from imaging apparatus to generate image.
Characteristic of the present invention is that optical axis inching gear 4 is connected with camera part 10.By optical axis inching gear 4, make 10 pairs of optical axises of video camera part do the vibration of transverse direction, make optical axis that small laterally moving take place.Here do not have the special provision vibration frequency, for example can set by the movement velocity of considering target object in several Hz suitable vibration in the scope of tens Hz.Then, in the process of the optical axis of camera part 10 vibration, carry out many pieces of photographies by 10 pairs of target objects of camera part.Bottom line need be obtained 2 pieces of images that move middle two ends, the left and right sides, and this can reach by the photography time difference of adjusting optical axis inching gear 4 and camera part 10, also can use high-speed shutter to generate a large amount of images.A back in this way method when generating edge image, not only has the image at two ends, the left and right sides, can also generate the image of photography in the way, generates many pieces of images.In this link,, do suitable adjustment according to the processing power and the desired three-dimensional shape detection speed of image processing section 3.Image processing section 3 can be used personal computer, also can realize by the special circuit plate of DSP etc.
Optical axis inching gear 4, it can be any device that can make camera part 10 carry out microvibration at assigned direction, here do not do regulation, but provide concrete example: piezoelectric element, supermagnetic device, static gearing, oil hydraulic actuator, pneumatic actuator, electromagnet, optics gearing, marmem, macromolecule gearing any one etc. all feasible.
Optical axis inching gear 4 can be made and makes camera part 10 integral body do minute movement, also can make only to make camera lens 1 do minute movement.When making camera lens 1 do minute movement, can make the structure of camera lens as shown in Figure 2.Fig. 2 is for the front skeleton diagram of the example in the camera lens structure is described.As shown in the figure, be furnished with optical axis inching gear 4 around the camera lens, for example piezoelectric element more more specifically illustrates, disposes piezoelectric ceramics around camera lens, uses suitable material to be connected with camera lens 1.Illustrated piezoelectric ceramics is divided into 4 parts, by the control to each several part, can make camera lens at x, and y axle direction arbitrarily moves.But, if only need under the situation of move left and right, do not need piezoelectric ceramics is divided into 4 parts, it is just passable to be divided into two parts.Certainly, if the situation that camera part self is moved also can make to use the same method.
Referring again to Fig. 1, illustrate that along with the moving of camera part 10 distance of target object to the distance between the camera lens and camera lens to imaging apparatus reaches the relation between the variable quantity of imaging point on the imaging apparatus.As shown in the figure, target object 20 is d to the distance of camera lens 1, and camera lens 1 is h to the distance of imaging apparatus 2, and the distance that camera part moves is 1, the flat P of imaging point before camera part moves
1With move after imaging point P
2Between distance be s, represent with following formula:
s=h1/d ------(1)
Can obtain according to formula (1), the value of the more little s of value of d is big more, and the value of opposite d is big more, the rule that the value of s is more little.That is to say that camera lens is short more apart from d to target object, the mobile range s of imaging point is just big more.Therefore, according to the mobile range s of this imaging point, just can judge the distance of target object to a certain extent.That is to say that mobile range s has the range information of target object.Therefore, if synthetic several pieces of photographed images of camera part, relatively thicker after bigger many coincident of target object nearby of mobile range, the smaller distant object object edge of mobile range just seems thinner.
Fig. 3 is used for illustrating the generative process of the edge image of image processing part 3.So-called edge image is to aim at the mark on the image that object photographs the image at outstanding target object edge.Fig. 3 a does not use under the situation of optical axis inching gear, the edge image that obtains by prior art.For example, the tangible part of deep or light variation in the image handles the projecting edge part by the differential of deep or light difference.At same position target object is repeatedly photographed, edge image does not change, and the width at edge is identical, be without any the image of range information.
Yet, use many pieces of images that the camera part of optical axis inching gear of the present invention the obtains edge image that obtains of the processing by image processing section 3 again, the theory of corresponding above-mentioned mobile range s is shown in Fig. 3 b, for nearby edge is thicker, the thinner image in edge at a distance.Therefore, we can say that the border width of this edge image has range information, the border width of target object that is positioned at same distance is identical, and on the contrary, the edge of different in width is positioned at different distances.In addition, Fig. 3 b is the edge image that camera part obtains during along optical axis horizontal direction move left and right, owing to be move left and right, in order to satisfy above-mentioned mobile width s, this image is the outstanding image of longitudinal edge.That is to say to have only the edge of longitudinal direction to contain range information, the edge of transverse direction does not contain range information.Therefore, if the marginal existence of complete level is arranged, width may not change, and after doing difference image and handling, is not expressed out fully.
The Flame Image Process of image processing section 3 both can be as mentioned above to the deep or light difference of each image, do differential and handle the back stack, also can get the difference of the deep or light value of each image or chroma value after, do standardization.The standardization here is meant, the negative value that occurs after the difference for each image is got its absolute value, or the processing that all values are moved etc. to positive dirction.Moreover, also can do 2 values processing and make it to oversimplify the image of obtaining.And, if necessary, also can add and emphasize to handle or the graph thinning processing.Above-mentioned generation edge image is handled and is comprised existing method and all disposal routes that will be developed from now on.
Formation when assigning to detect three-dimensional shape with the image pickup part of said structure is described as follows.Fig. 4 is the skeleton diagram for the structure that three-dimensional shape pick-up unit of the present invention is described.As shown in the figure, camera part 10 is the parallel placement in the left and right sides, is output as the input of image processing section 3 separately.Term in this instructions " sight line " is meant from camera part 10 to target object 20 connecting line, and term " optical axis " is meant, the optical centre axle of camera part 10.
As shown in Figure 4, be the camera part 10 that is placed on two assigned addresses, the example of target object 20 being photographed from different direction of visual lines.But the present invention is not limited thereto, can use independent camera part move left and right in camera positions equally, from two different position photographies.Implementation method is not in this case done concrete regulation, for example can carry camera part 10 in orbit, and camera part can be moved in orbit freely.Moreover camera positions are not confined to two positions.If the eyes formula can be used two camera positions, but, also can be placed in a plurality of positions to camera part if when thinking to detect the surface configuration of three-dimensional object accurately.That is to say, not only can be placed on side by side on the x direction of principal axis, also can be placed on side by side on the y direction of principal axis.Moreover, consider that from the simplification aspect of handling each imaging apparatus is placed at grade to well.
The optical axis inching gear 4 that is placed in above-mentioned camera part 10 is moved along two designated photograph position closures.This minute movement moves to well along the normal direction of the optical axis of camera part 10 is parallel.Not only can do parallel moving, also can do gyration.In the time of gyration, the centre of gyration is that point beyond the optical centre of camera lens is for well, as point on the optical axis or near the point the optical axis.That is to say, moving that the optical axis of camera part 10 changes got final product.Just as the mankind's eyeball reaches the effect of microvibration by gyration, therefore, when anthropomorphic robot's eye used three-dimensional shape pick-up unit of the present invention, the gearing that is used as common eye motion also can be used for as the optical axis inching gear.Also have, consider, the centre of gyration of each video camera is placed on the unified plane to well from the angle of the simplification handled.In addition, the angle of revolution of camera part, the degree that the mobile range s of target object on imaging apparatus that approaches infinity is adjusted to the 0-3 pixel is for well.Just, make the edge of the target object that approaches infinity, as far as possible carefully be (wide) well, perhaps fully not by imaging such as 2 pixels.
As mentioned above, " direction that connects along two designated photograph positions moves " mentioned in this instructions, not only be meant along the normal direction of camera part optical axis and do parallel motion, also comprise with each camera part or camera lens each other near or mutually away from direction on be the gyration at center with the specified point.
Fig. 5 illustrates the moving direction of optical axis inching gear 4 of the present invention, and the skeleton diagram that a plurality of camera part edge directions is separately moved is described, is the front skeleton diagram from the camera lens top view.Fig. 5 a is the example that has two camera part.The optical axis mobile device makes camera part 10a and 10b, along the direction left and right sides fine motion of ab.When 3 camera part, shown in Fig. 5 b, camera part is placed on the vertex of a triangle position, when by camera part 10a and 10b photography, along the fine motion of ab direction; Camera positions are changed, during by camera part 10a and 10c photography, along the fine motion of ac direction; Camera positions are changed once more, during by camera part 10b and 10c photography, along the fine motion of bc direction.As mentioned above, by making the camera part optical axis along a plurality of direction fine motions, the Edge Distance that detects a plurality of directions that can be more correct.And then, shown in Fig. 5 c, also can use 4 camera position 10a--10d, along fine motion separately up and down.
Three-dimensional shape pick-up unit of the present invention, since can move optical axis along the closure between camera positions, can certainly be as Fig. 5 b and Fig. 5 c a plurality of direction fine motions along combination.
Certainly, also can make optical axis separately do fine motion along the direction beyond the closure of the camera part of Fig. 5.By making the optical axis fine motion along a plurality of directions, the motion of the smooth motion beyond can the oculomotor rectilinear motion of simulating human and the correspondence of dither, thus edge image is supplied.
When using a plurality of camera part, each camera part is photographed simultaneously to well.That is to say that the minute movement that makes each camera part is identical moves, make exactly its simultaneously, equidirectional, with the minute movement of carrying out of amplitude.Photography is also carried out simultaneously.Therefore, also can obtain good edge image for the target object in the motion.
The following describes the method that each edge image that uses said method to obtain carries out the reference point correspondence.General two camera part of passing through are photographed when asking the corresponding point of image, use the images match method usually.The present invention certainly uses the images match method.Also can edge image be oversimplified carry out high speed processing by following method.
Use two camera part that optical axis inching gear of the present invention is housed in order to illustrate, each edge image that obtains when target object is photographed, Fig. 6 are that the skeleton diagram that how to obtain edge image is described.Make by the optical axis inching gear that the camera part left and right sides is little to shake, obtained the edge image that vertical direction is reinforced,, obtained the result that the edge that has only longitudinal direction as shown in the figure is extracted its further simplification.As shown in the figure, the edge of edge image of the present invention contains range information, and the object edge width of same distance is also identical.Therefore, be benchmark with the length and the width at these edges, make the corresponding possibility that becomes in edge on the edge image that obtains from two camera part.That is to say, left side edge edge of image Ra, Rb, Rc and Fig. 6 b right side edge edge of image La, Lb, Lc with regard to Fig. 6 a, when the edge in the decision right side edge image corresponding with left side edge edge of image Rb, from the edge image on right side, finding out with edge Rb has the edge of same widths and length to get final product.Like this, just detected the edge Lb of equal length and width.Therefore, be reference point with the upper left summit of edge Rb, Lb, make the corresponding possibility that becomes of each several part of left and right edges image.
And complex-shaped when target object, under the dissimilar situation of edge length width of edge image, use the images match method, in the image of the left and right sides, find identical pattern to carry out correspondence.
Then, use the corresponding edge image of crossing of reference point to obtain parallax, three-dimensional shape is detected becomes possibility.
As mentioned above, can make obtaining of edge image become very simple, only use the length of edge image and width to be benchmark, just can carry out the reference point correspondence, the corresponding desired processing time of work is significantly subdued with the reference point correspondence by the present invention.And, if obtain edge image, can also detect trickleer three-dimensional shape from essential plural camera positions.Moreover, generate stereo-picture after, also can be with by the level and smooth surface configuration of the target object of surrounded by edges, by the parsing of deep or light value, as the surface three-dimensional information stickup of target object on stereo-picture.
In addition, from about camera part between the nearer target object of central shaft, can pass through said method corresponding edge image.But, when target object when the central shaft of left and right sides camera part is far away, the identical edge and the distance of each camera part are very different, even the corresponding edge, change has also taken place in border width.About above-mentioned situation, can be that benchmark detects the corresponding edge with direction of visual lines to target object.Such as, judge direction of visual lines by edge calculation to the distance at edge image center, thereby from the right side edge image, find the corresponding edge that has equal length with the left side direction of visual lines.Moreover, to be benchmark, can make the edge transition on the edge image become normal width to the width on the direction of visual lines of target object.That is to say that the width (normal width) after the edge on the direction of visual lines that has obtained departing from moves on the optical axis direction makes and do not consider direction of visual lines, simple comparison edge carries out correspondence and becomes possibility.
Three-dimensional shape pick-up unit of the present invention and three-dimensional shape detection method are not limited to above-mentioned example, and only otherwise break away from aim of the present invention, various changes all are allowed to.