METHOD OF SOLVING THE CORRESPONDENCE PROBLEM IN CONVERGENT STEREOPHOTOGRAMMETRY
REFERENCE TO RELATED APPLICATIONS
This application claims an invention which was disclosed in Provisional Application Number 60/458,074, filed March 27, 2003, entitled "METHOD OF PHOTOGRAMMETRY USING CENTRAL SCAN LINE FOR SOLVING CORRESPONDENCE PROBLEM" and Provisional Application Number 60/529,964, filed December 16, 2003, entitled "EXTRACTING SCALE AND CONVERGENCE OF
A SINGLE CAMERA 3-D ZOOM ADAPTER WITH STRUCTURED LIGHT". The benefit under 35 USC §119(e) of the United States provisional applications is hereby claimed, and the aforementioned applications are hereby incorporated herein by reference.
BACKGROUND OF THE INVENTION
FIELD OF THE INVENTION
The invention pertains to the field of stereographic photography. More particularly, the invention pertains to a stereographic method for extracting information from an image and finding matching points in multiple images.
DESCRIPTION OF RELATED ART
Stereographic photography is the method of producing images, which are apparently three-dimensional, by recording separate left- and right-eye images. The viewer reconstructs the 3-D image by viewing the two separate 2-D images simultaneously. Stereographic photography has been known since at least the mid- 19th century, when stereo viewers were a popular parlor accessory. Such stereo views have historically been created with two lenses on a single camera, spaced apart by approximately the inter-ocular distance of a human head.
Photogrammetry is the process of making measurements through the use of photographs. Stereo photogrammetry uses multiple images (usually left/right images from stereo pairs) to measure objects and positions in three dimensions. To do so, it is necessary to locate the same ("homologous") point in each of the multiple images. The problem of finding homologous image points in multiple images is known as the "problem of correspondence". As the camera or lens separation increases, so do the differences in the scene as recorded by each camera or lens, thus making it difficult to match corresponding points in the two images.
All of the multiple-lens or multiple-camera systems in the prior art have severe drawbacks, in the added complexity and cost of duplicating the complete camera system and the synchronization of the two separate images (this is especially a problem in film (non- video) applications). In addition, the use of two separate lenses (whether on one camera or two) introduces problems of synchronizing focus, view, and the images themselves.
hi order to minimize these drawbacks, the inventor of the present invention has developed methods and apparatus for using converging mirrors to enable multiple images to be obtained by a single lens, and has received the following U.S. patents: "SINGLE LENS APPARATUS FOR THREE-DIMENSIONAL IMAGING HAVING FOCUS- RELATED CONVERGENCE COMPENSATION", US Patent No. 5,532,777; "METHOD AND APPARATUS FOR THREE DIMENSIONAL MEASUREMENT AND
IMAGING HAVING FOCUS-RELATED CONVERGENCE COMPENSATION", US Patent No. 5,828,913; and "APPARATUS FOR THREE-DIMENSIONAL MEASUREMENT AND IMAGING HAVING FOCUS-RELATED CONVERGENCE COMPENSATION", US Patent No. 5,883,662. These patents are herein incorporated by reference.
SUMMARY OF THE INVENTION
One embodiment of the present invention is a method for solving the problem of correspondence associated with convergent 3D cameras.
The method includes the steps of aligning a central horizontal epipolar line parallel to the scan line of a charge-coupled device (CCD), choosing at least one scan line of the CCD that adheres to epipolar geometry, conduction of a disparity estimation of the scan line chosen, self calibration of the camera, and rectification of the images. In an alternative embodiment of the present invention, a laser pointer is aimed within the central epipolar plane and perpendicular to the base line to help aim the device at the object enclosed by the measurement volume, and allow the camera to adjust itself.
Another embodiment of the present invention is a method of extracting the settings of a 3D adapter with variable convergence from an image, using structured light, hi an alternative embodiment the magnification settings can be obtained from the image.
This method includes the steps of aligning a central horizontal epipolar line parallel to the scan line of the CCD; choosing at least one scan line of the CCD that adheres to epipolar geometry; illuminating the scene with a laser beam positioned within the central epipolar plane and halfway and perpendicular to the base line; obtaining an independent means of range estimation with an optical linear position sensor aimed under an angle of equal or less than 90 degrees to the laser beam thrown on the scene by the above described laser pointer; conducting a disparity estimation of the laser dot on the scan line chosen; and reconstructing the convergence from the range and the disparity.
In an alternative embodiment, the independent means for range estimation is a laser range finder using the 'time of flight' principle with its laser beam positioned within the central epipolar plane and positioned halfway and perpendicular to the base line.
In another embodiment, a second laser range finder positioned parallel to the first beam and halfway and perpendicular to the base line throws a laser dot on the scene so the scale can be extracted from the image of the scene.
BRIEF DESCRIPTION OF THE DRAWINGS
Figs. 1 A and IB show how correspondences are constrained to conjugate epipolar lines in a parallel camera setup and a convergent camera setup, respectively.
Fig. 2 shows the general concept of epipolar geometry.
Fig. 3 shows convergence extraction with linear PSD in a method of the present invention.
Fig. 4 shows convergence extraction with a laser range finder in a method of the present invention.
Fig. 5 shows a structured light set up to extract convergence and scale in a method of the present invention.
DETAILED DESCRIPTION OF THE INVENTION
The present invention relies on the fundamentals of epipolar geometry to solve the stereo correspondence problem of finding homologous image points in multiple images. Epipolar geometry is a property specific to stereo cameras (see Fig 1 A).
Epipolar geometry is a geometric relationship between two perspective cameras.
The epipole is the point of intersection of the line joining the optical centres with the image plane. This is the image in one camera of the optical centre of the other camera. The epipolar plane is a plane defined by a 3D point and the optical centres. It is the plane defined by an image point and the optical centres. The epipolar line is a straight line of intersection between the epipolar plane and the image plane. It is the image in one camera of a ray through the optical centre and image point in the other camera. Epipolar lines all intersect at the epipole.
Regarding correspondences between images, a point in one image creates a line in the other image on which its corresponding point lies. Therefore, the search for correspondences is reduced from a region to a line. This creates an epipolar constraint, which arises because, for image points corresponding to the same 3D point, there is coplanarity between the image points, 3D point and the optical centres.
Figure 2 shows a stereo (pinhole) camera. The optical centers (centers of the lenses of actual cameras, OLFL and OLFR) are the origins of the left and right lens reference frames. The line through these two optical centers is called the baseline (21). Any plane
(20) in 3-D space that contains the baseline is called an epipolar plane. All scene points in such a plane are projected on a line in each of the images (24), (25). These lines are the epipolar lines. A pair of epipolar lines (26), (27) that share the same epipolar plane are
called conjugate epipolar lines. If two points from the image pair correspond, they must lie on conjugate epipolar lines. This is called the epipolar constraint. It reduces the set of possible correspondence candidates for a point in the left image from all points in the right image to only those on the conjugated epipolar line in the right image. For pinhole cameras, the epipolar lines are straight. Due to lens distortion, the epipolar lines may become curved.
Epipolar geometry is also disrupted as soon as the two cameras deviate from a stereo configuration whereby the optical axes of both cameras no longer are parallel as shown in Fig IB. If you change the convergence of both cameras over one and only one common axes of rotation there will always be one plane for which the epipolar constraint is valid. This epipolar plane (the "central epipolar plane") (30) goes through the baseline and is normal to the common axes of rotation over which the cameras converge.
This concept is illustrated in Figure IB. The images shown are made with a setup using convergent left and right views of the kind used in the patents cited above. Charge - coupled devices, image sensors that separate the spectrum of color into red, green, and blue, or a video camera using another technology, are preferably utilized in the present invention. In one example, CCDs are used in conjunction with each of the lenses (24) (25). hi this example, the CCD is an area CCD, which is square or rectangular in shape and can capture an entire image at once. Alternatively, another type of single video camera is used in conjunction with the single camera 3D adapter to obtain a left and right view of the scene (24) (25). Herein, the terms CCD and video camera are used interchangeably to mean an image forming device.
The present invention aligns a central horizontal epipolar line (10) parallel to the central scan line of the video camera, so that at least one of the scan lines from the video camera adheres to epipolar geometry. The scan lines preferably adhere to epipolar geometry for all convergence settings. The epipolar line (10) is preferably perpendicular to the single axis of rotation for convergence adjustment. Therefore, if at least one of the scan lines from the video camera has scene points from each of the images on one line, a pair of such scan lines would share the same epipolar plane, and if the two points from the image pair were to correspond, they would lie on conjugate epipolar lines.
A method that searches for disparity in the central epipolar plane starts its searching at the CCD scan line closest to the central epipolar plane, but must expand symmetrically within adjustable limits its searching to adjacent CCD scan lines until it solves the correspondence for homologous points in the left and right view of the scene.
With convergent three-dimensional imagers one can resurrect epipolar geometry for points in 3-D space that do not lay in the above defined central epipolar plane, by rectifying left and right views. Rectification is only possible when both views are calibrated. U.S. Patent No. 5,532,777 describes an adapter that generates a left and right view with variable convergence from a single lens, in which the convergence is coupled mechanically to the zoom or focus setting of the lens. U.S. Patent No. 5,828,913 describes a method to dynamically calibrate such a system by selecting one calibration map matched for a particular convergence setting from multiple calibration maps for multiple convergence settings. Mechanical solutions are very expensive to make when they have to adhere to the tight tolerances required by the solution proposed in the preceding patents.
In the co-pending patent application "ACQUISITION OF 3-D SCENES WITH A
SINGLE HAND HELD CAMERA", U.S. Patent Application number 09/595,402, herein incorporated by reference, a method for self calibration was proposed that took advantage of the fact that the mechanism used in adapters defined in U.S. Patent No. 5,532,777 has a predictable effect on the disparity among homologous points. However the method described is slow in part because it relies on two-dimensional correspondence solving.
The present invention does not rely on a mechanical coupling to select the correct calibration map. Instead, it extracts from the scene the convergence of the system by scanning for the disparity between left and right views of homologous points, or a red aiming dot illuminated by a laser beam thrown on the scene in the central epipolar plane and halfway, perpendicular to the baseline, the line that connects the centers of left and right view. Because such points lay on conjugate epipolar lines, the correspondence problem reduces to a one dimensional search problem which is simpler and therefore faster. Referring to Fig. 3, once the correspondence for left and right view of the aiming dot (or homologous points) is known, the disparity can be constructed as follows:
By using the independently obtained range, using a linear position sensor (31) of the laser dot projected on the scene, the convergence can be derived by comparing the disparity with the observed range. Next you select a calibration map matching the current convergence, calibrate the 3-D imager and rectify the left and right views. Once both views are rectified all points in left and right views lay on conjugate epipolar lines and solving the correspondence in left and right view for homologous points is relatively simple and therefore fast.
As shown in Fig. 4, in an alternative embodiment, the range of the aim dot is obtained with a laser range finder (40) that works according to the 'time of flight' principle.
Some design requirements of the 3D imager to make this scheme work are:
Change of convergence for both views must be over one, and only one, common axis of rotation.
The CCD scan lines must be exactly parallel to the central epipolar plane (30) of the imager. The CCD scan lines are also preferably perpendicular to the common axis of rotation for the convergence adjustment.
To aid in detecting the disparity of homologous points in low contrast scenes, a laser beam aimed at the scene and positioned in the central epipolar plane preferably half way and perpendicular to the baseline is used. A laser range finder (40), whether it is one using a linear optical position sensor (31) or a time of flight sensor, needs its laser beam aimed exactly the way it is defined for the laser beam used for aiming as described in the previous paragraph.
hi a third embodiment shown in Fig. 5, both convergence and scale are extracted from the image. The system works as follows:
Convergence extraction:
With a laser range finder (LRF) the convergence can be reconstructed using the range c, the known baseline z, and the directions of the line of sight for the dot for left [XL, j and right view [X , VR]. Once the convergence is known the system can be calibrated
with a static calibration map matching the current convergence angle and the image can be rectified.
Scale extraction:
Any time the lens zoom setting is changed, a second laser beam of a laser range finder positioned at an offset d and parallel to the LRF illuminates the target with a dot.
Because the system is rectified it is possible to obtain a non-scaled length estimate of distance f. The offset d is known and the ranges c, b allow a direct scaled estimate off. Comparing the estimated and direct measurements leads to a scaling correction so that the rectified system can measure known units, which depends on the focal length of the lens used with the system.
Accordingly, it is to be understood that the embodiments of the invention herein described are merely illustrative of the application of the principles of the invention. Reference herein to details of the illustrated embodiments is not intended to limit the scope of the claims, which themselves recite those features regarded as essential to the invention.