AU2002315499A1 - Overhead dimensioning system and method - Google Patents

Description

OVERHEAD DIMENSIONING SYSTEM AND METHOD

CROSS-REFERENCE TO RELATED APPLICATIONS 5 This application claims priority to U.S. Provisional Patent Application entitled "Overhead

Dimensioning System," Serial No. 60/302,509, filed June 29, 2001; the contents of which are incorporated herein by reference.

FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT 0 Not Applicable.

BACKGROUND OF THE INVENTION Technical Field

The present invention relates to a machine vision system for dimensioning large or palletized 5 freight, of one or more pieces.

Systems for visually dimensioning objects are generally well known. See for example, U.S. Patent Nos. 4,731,853; 5,193,120; 4,929,843; 5,280,542; and 5,555,090, and "Optical Three- Dimensional Sensing for Machine Vision," T.C. Strand, Optical Engineering, Vol.24 No. 1, pp.33- 40. Such systems scan the object and the surrounding surface with a laser, and detect the laser o reflected off of the scanned object, as well as off the surrounding surface, with a CCD camera. The detected laser image is analyzed to determine the dimensions of the object via triangulation.

Such systems for dimensioning objects have required a level of environmental structuring that has limited the potential application of automated dimensioning. In particular, such systems have been limited to substantially cuboidal obj ects and/or obj ects in known positions, and thus, have 5 been unable to tolerate objects having highly non-cuboidal shapes. Most often, limitations on the range of object size in relation to measurement resolution / accuracy have been imposed. In operation, these systems have been slow or awkward to use. Generally, the systems have been intolerant to variations in object reflectance, within or between objects, and/or ambient lighting. In order to reduce occlusion, such systems typically utilize movement, i.e., rotation, of the object and/or o the sensing device; or require optical components to be located at the level of the object rather than being positioned remotely overhead. And finally, the common dimensioning systems have required costly hardware. The present invention is provided to solve these and other problems. SUMMARY OF THE INVENTION

The present invention is a method for determining the dimensions of an item placed within a measurement space of a dimensioning system. The components of the dimensioning system may be mounted remotely overhead and are configured to minimize occlusion to recover the true dimensions 5 of the object. The method includes scanning a signal through a measurement space. The acquired images are optically filtered and differenced to isolate the signal. A first laser and a first camera are utilized to determine an approximate location and dimension of the item. A second laser and a second camera are utilized to determine an approximate location and dimension of the item. A first set of point cloud data is acquired wherein the first laser scans a first signal through the measurement o space and the first camera receives the reflections of the first signal. A second set of point cloud data is acquired wherein the second laser scans a second signal through the measurement space and the second camera receives the reflections of the second signal. A third set of point cloud data is acquired wherein the first laser scans tlie first signal through the measurement space and the second camera receives the reflections of the first signal. A fourth set of point cloud data is acquired 5 wherein the second laser scans the second signal through the measurement space and the first camera receives the reflections of the second signal. An image is constructed by merging the first, second, third and fourth sets of acquired point cloud data. A smallest rectangular prism, e.g., cuboid, rectangular parallelepiped; is determined to contain the constructed image.

A further aspect of the present invention includes utilizing an image point correction factor. o The image point correction factor is determined during calibration of the dimensioning system and includes a set of generated equations or look-up tables to correct lens distortion. The distortion corrections are utilized in cooperation with the constructed image to determine the cuboid.

Yet a further aspect of the present invention incorporates a rapid scanning technique wherein a first — coarse — scan quickly locates an object and once located, a second — fine — scan is 5 utilized to dimension the obj ect. Alternatively, an adaptive scanning technique is utilized using both coarse and fine scans to locate and dimension the object.

Yet another aspect of the present invention includes acquiring additional sets of point cloud data and constructing the image by merging all the sets of acquired cloud data.

One object of the present invention is directed to providing a dimensioning system wherein o the working component(s) of the system are mounted remotely, e.g., overhead, to allow unobstructed passage throughout the measurement space.

Another object of the present invention is to provide a system for dimensioning large or palletized freight, of one or more pieces.

Yet another object of the present invention is to provide a dimensioning system capable of 5 being installed within existing operational environments. In accordance with the present invention, the system can determine the dimensions of a rectangular prism having the smallest volume, but which would contain the freight.

In further accordance with the present invention, the system can determine the dimensions in varying levels of ambient light and varying object surface reflectance.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a block diagram of hardware in accordance with the invention;

FIG. 2 is a flow chart illustrating basic steps performed by the hardware of FIG. 1;

FIG. 3 is a more detailed flow chart of one of the steps of FIG. 2; FIG. 3 is a more detailed flow chart of another one of the steps of FIG. 2;

FIG. 4 is a more detailed flow chart of still another one of the steps of FIG. 2;

FIG. 5 is a more detailed flow chart of still another one of the steps of FIG. 2;

FIG. 6 is a more detailed flow chart of still another one of the steps of FIG. 2;

FIG. 7 is a block diagram of another embodiment of the present invention; FIG. 8 is an image of a box with a line of light from a projector;

FIG. 9 is a thresholded image of the box of FIG. 8;

FIGS. 10a and 10b are photographs showing one embodiment of the present invention;

FIG. 11 shows a perspective projection in which object points are projected through the image or view plane to a point known as the center of projection or focal point; FIG. 12 shows a schematic representation of the optical geometry used in the method of stereo triangulation;

FIG. 13 is a block diagram showing the primary image processing stages of one embodiment of the present invention;

FIG. 14 is a schematic side-view of one embodiment of the present invention; FIG. 15 is a schematic drawing showing geometric detail of the triangle formed by one embodiment of the present invention;

FIG. 16 is a schematic drawing of light from the box passing through the camera lens and impinging on the CCD camera detection surface;

FIG. 17 is a block diagram of an alternative embodiment of the present invention; FIG. 18 is a block diagram of an alternative embodiment of the present invention;

FIG. 19 is a block diagram of an alternative embodiment of the present invention;

FIG. 20 depicts an undistorted image of a square;

FIG. 21 depicts a simulated image affected by radial lens distortion;

FIG. 22 depicts a simulated image affected by radial lens distortion; FIG. 23 depicts an image on a coordinate frame; FIG. 24 is a distorted image of a square;

FIG. 25 is a distorted image of a square;

FIG. 26 is a distorted image of a square;

FIG. 27 depicts a screen of a graphical interface of the dimensioning system of the present 5 invention;

FIG. 28 depicts a screen of a graphical interface of the dimensioning system of the present invention;

FIG. 29 is a photograph of one embodiment of the hardware configuration of the present invention; o FIG. 30 is a schematic diagram showing the accumulated laser scan lines of a non-cuboid object;

FIG. 31 is a block diagram depicting one method of determining a minimum enclosing rectangle;

FIG. 32 is a block diagram depicting another method of determining a minimum enclosing 5 rectangle;

FIG. 33 is a front view of one embodiment of the dimensioning system of the present invention;

FIG. 34 is a front view of an alternative embodiment of the present invention shown in FIG. 33; 0 FIG. 35 is a top view of one embodiment of the dimensioning system of the present invention; and,

FIG. 36 is a top view of one embodiment of the dimensioning system of the present invention.

s DETAILED DESCRIPTION OF THE INVENTION

While this invention is susceptible of embodiment in many different forms, there is shown in the drawings and will herein be described in detail a preferred embodiment of the invention with the understanding that the present disclosure is to be considered as an exemplification of the principles of the invention and is not intended to limit the broad aspect of the invention to the embodiment 0 illustrated.

One embodiment of a dimensioning system 10 of the present invention is illustrated in FIG.

1. The system 10 includes a CCD camera 12, sensitive to long wavelength visible light (680nm) having a lens 14 with an auto iris (iris controlled from computer and/or camera video signal).

Alternatively, rather than an auto iris, a software approach could be used, wherein the camera 5 integration time, i.e., shutter speed, is varied, which could permit faster operation. The system 10 further includes an infrared blocking filter 16 and a colored glass filter 18. The colored glass filter 18 passes deep red. The system 10 also includes a laser 22, having a 680 nm, 0.91 mW output, class II. The laser 22 produces a 'flat-top' line of light with a 60° fan angle. The laser 22 is powered by a 5 V DC power supply 24. A mirror 26 is incrementally rotated by a scanner 28 under the control of a motor drive 30. Specifically, the rotational position of the mirror 26 is proportional to the voltage input to the scanner 28. A personal computer 32, incorporating input/output cards (not shown), controls rotation of the mirror 26 and operation of the laser 22, as well as performs other calculations, discussed below. The personal computer 32 controls the laser 22 via a TTL signal 33. The laser 22 forms a plane of light, generally designated 34, upon an object 36 to be measured. The object 36 can be one or more pieces. The object 36 is located on a measurement space of a surface 38, which may be a pallet, or directly upon a floor surface.

The general steps performed by the system 10 are illustrated in FIG. 2. As discussed in greater detail below, the system 10 performs two scanning steps to dimension the object 36. The first scanning step is a coarse scan, wherein the mirror is incremented in relatively large increments, to coarsely determine the location of the start point and end point of the object 36. The second step is a fine scan, wherein the mirror 26 is incremented in relatively small increments near the start point and end point of the object 36, to precisely determine the location of the periphery of the object 36.

Preferably, in a Find step, the object 36 is scanned by the laser 22 in relatively coarse steps to determine whether an object 36 is present, and if so, the general location of the beginning and ending of the object 36. If an object is not present, the system 10 stops. However, if an object 36 is present, an Acquire step is performed, wherein the object 36 is re-scanned by the laser 22, but in relatively fine steps.

An alternative scanning technique — intelligent scanning — can significantly reduce the amount of time to dimension a single object. Intelligent scanning begins with a coarse scan at a location off-center of the measurement space wherein the object rests. The coarse scan continues in a first direction, e.g., forward, until an object is found or until it is determined that there is no object near the center of the measurement space. If an object is found, the coarse scan is continued in the first direction until an edge is found. The fine scan is then initiated in a second direction opposite to the first direction, e.g., backward, over tl e edge. The coarse scan is then resumed at the initial starting point in the second direction until the object's other edge is found, wherein the fine scan is initiated in the first direction upon location of a second edge. If the object is not found with the first scan signal, but the object edge is found with the subsequent coarse scan signal, the fine scan of the edge is immediately performed. Then, the coarse scan is resumed to find the other edge, wherein the fine scan is subsequently initiated. A Perspective step is then performed, which adjusts the length ("x") and width ("y") dimensions in view of the height ("z") dimension. This is because small objects close to the lens appear the same as large objects distant from the lens. A Cube function is then performed which determines the dimensions of a rectangular prism having the smallest volume about tl e object 36. 5 The Find step (coarse scan) is illustrated in greater detail in FIG. 3. In a first step, the system is initialized, and then a first image is obtained by the camera 12. The input voltage to the scanner 28 is increased by a coarse step, e.g., 0.4 V, which advances the mirror 26 a relatively large increment. The first image is electronically stored and a second image is obtained. In order to eliminate adverse effects of ambient light, the first image is subtracted from the second image, which o eliminates the ambient light factor, leaving only the laser portion. This gray level image is then utilized as a threshold to provide a binary image.

Since the color and reflectivity of objects being measured varies, the signal may overrun into adjacent pixels causing some measurement inaccuracies. Some of these inaccuracies may be addressed by a thresholding operation or by subsequent image filtering. Also, noise may be more 5 prevalent in light-colored, shiny objects. For instance, for light-colored, shiny objects, the laser signal reflection is bright; and conversely, for flat, dark-colored objects, the laser reflection signal is significantly smaller. Consequently, the optimum binary decision threshold to be used needs to be adaptive according to the reflectance/coloring of the object. It may also be necessary to adaptively alter either the camera aperture or camera integration time. Such "automatic" thresholding occurs 0 when an object is found during a scan and the gray-scale values of the points found in the image above a threshold are gathered. A statistical property value, e.g., mean, of these points is used to choose one of a predetermined set of threshold values, preferable a set of three. The three threshold values and the scan determination values are determined during a calibration phase of the system. To further increase the coarse scan speed, every fifth pixel of the threshold result is searched to 5 locate the highest pixel, and then the height of the highest pixel is determined. The present disclosure assumes the object has a minimum programmable height and may be located on a pallet of minimum programmable height, e.g., 8 cm high. Therefore, the object itself will always have a height greater than 8 cm. The system 10 can separate the object 36 from the pallet based upon its height. It is also possible for the system to automatically determine the height of the pallet. o The purpose of the Find function is to establish the position of the laser 22, measured in volts, both at the point at which the laser first, i.e. "start", and last , i.e. "stop," encounters the object 36. The lower box in FIG. 3 works as follows. At the start of the program, 'startflag' is initialized to 0. The input voltage to the scanner 28 is incrementally increased (by 0.4 V increments) within the loop. If an object greater than 8 cm in height is encountered while "startflag" equals zero, then 5 "start" is set to volts and "startflag" is set to one. By changing "startflag" to equal one, "start" will not be altered during subsequent passes through the loop. The second "if statement in the final block states that if height is greater than 8 cm then set "stop" to volts. Thus, for subsequent passes through the loop, "stop" may continually be reset, i.e., if height > 8 cm. Therefore, at the end of the laser scan "start" and "stop" are set to the points at which the laser 22 first and last encountered the object 36, respectively.

The Acquire function is illustrated in FIG. 4. This function is similar to the Find function, except that the mirror 26 is incremented in relatively small steps at the start point and end point of the object 36. Additionally, the height of every pixel — not every fifth pixel as in the Find function — is calculated and stored. Additionally, depending upon the quality of the lens (short focal length lens have greater peripheral distortion), peripheral correction can also be conducted. In a final step, data, e.g., a 3-dimensional cloud of points, having a height greater than 8 cm — to distinguish the object 26 from the pallet — is formed.

The next step is the Perspective function and is illustrated in FIG. 5. In this function, the personal computer increments through the stored cloud of data points and converts the "x" and "y" values from pixels to centimeters. Based upon conventional equations, these converted values are then adjusted, based upon their respective "z" value and stored.

The next step is the Cube function, which determines the dimensions of a rectangular prism having the smallest volume about the object 36. The rectangular prism will always have a base parallel to the pallet, or other surface on which the object 36 rests. In a first step, the cloud of data points is rotated about the z-axis, to determine a rectangle having the minimum area, but which encloses all of the "x" and "y" coordinates. The cloud of data points continues to rotate a total of

180° to determine the smallest rectangle. This determines the length and width, e.g., breadth, of the rectangular prism. The system 10 then determines the largest "z" value, which is the height of the rectangular prism. Utilizing the plane of light, e.g., a laser line, provides advantages in terms of being resistant to the effects of changes in background lighting, or the presence of labels and other albedo patterns on the object 36. This ability may also be enhanced by placing a filter over the camera, which is opaque to all frequencies of light other than that of the laser. The scanning line can further enable detection of more complex morphologies, which is useful for objects other than cuboids. FIG. 7 depicts a dimensioning system 12 including a CCD camera 12 mounted above an object 36, e.g. cuboidal box, wherein a light strip (laser) projects diagonally onto the box to produce a "pulse" type of image. The trigonometry of the system in cooperation with the image can be analyzed to determine the dimensions of the box 36. The camera 12 is mounted at a predetermined distance from the box and captures the image by receiving a signal from the laser 22 shown in FIG. 8. This image was captured in the presence of a typical background, e.g., variable daylight. A filter in combination with a predetermined frequency of laser light can be utilized to effectively remove any detrimental background lighting effects. The filter is mounted on the camera and generally opaque while transmitting the laser frequency. Such a technique can provide a significant benefit of being able to operate in typical ambient lighting. The image depicted in FIG. 8 was thresholded and is shown in FIG. 9.

Small noise elements in the measuring field can cause large errors in the dimensioning process. The noise may be attributable to small debris objects within the field of view or specular reflections of the laser on the measuring surface. To remove visible noise from the image, median filtering can be applied to the image. Median filtering is considered appropriate when the aim is to reduce noise while preserving edges. Each pixel is set to the median of the pixel values in the neighborhood of the pixel, e.g., 4x4. During image measurement applications, edges are often more useful than regions. Therefore, the image can be subjected to further filtering that will result in an increased emphasis on the edges. FIG.9 more clearly shows the "pulse" referred to earlier. The height of the pulse can be used to determine the height of the object 36. For example, in FIG. 7, the height of the camera above the table is 160 cm, and the horizontal distance from the proj ector to the camera is 112 cm. The angle between the light strip and the floor, Θ, is 55°. Therefore,

H = d-tan(ι) Equation 1 where d is the apparent height of the pulse shown in FIG. 14 (referred to as the line separation), and H is the height of the box 36. The line separation can be determined by using the following procedure :

• calculate the center of mass (COM or centroid) for the image;

• calculate the average y-value for the pixels above the COM.;

• calculate the average y-value for the pixels below the COM; and,

• subtract the first y-value from the second to obtain the line of separation. The above procedure was employed in a MATLAB function and the line separation was found to be 146.3 cm. The length of a line on the floor was measured and compared to its length in the image in terms of pixels, and it was found that one pixel corresponds to 0.04774 cm. Consequently the line separation was found to be 6.98 cm. Utilizing this value, H is determined to be 9.98 cm. Since the measured value for H is 10.1 cm, the calculated box height has an accuracy of 98.8%.

The present invention is capable of incorporating several additional noise detectors, filters, and methods that can be implemented to find and eliminate noise during the dimensioning process. A further noise detection method computes a spatial histogram of a point cloud data image in the horizontal and vertical directions. Spatially comiected values in the histogram — or in the case of the readings along the vertical axis, values with minimal gapping are considered to be an object. Groups of spatially detached values in any of the histograms are determined to be noise or another object. If the total number of points in the secondary object is less than a predetermined threshold, then the points associated with that secondary object are considered to be noise and are removed from the point cloud data. 5 Further noise reduction can be accomplished by utilizing additional vertical and horizontal histograms of an array, or image. Multiple rotations can be incorporated at varying increments, e.g., 30°, 45°, etc., wherein the array is rotated in space — in the x and y planes.

Another noise detection method utilizes each column of each measurement image to identify a position of each disjoint point in the column. If more than one signal is found in each column, one o of the points can be assumed to be noise. When more than one signal is found in a given column, the height values of the multiple signals are compared with the height values of other signals in the surrounding spatial area. The signal point(s) that most closely matches those in the nearby area is considered as part of the object.

Yet another noise detection method sorts the heights of the points in the object cloud. The 5 spacing between the points is evaluated and points of similar height are grouped together. If any one group has a very small number of points, these points are eliminated from the object point cloud.

Another embodiment of the present invention for the determination of the height, length, and breadth of a cuboid, utilizes the method of stereopsis. This method can be used in conjunction with other methods described in the multiple camera configuration. The system comprises two identical o square pixel (11 x 11 mm) gray-scale cameras fitted with 8 mm (focal length) lenses. The cameras are positioned to view an object vertically from above, as shown in FIGS. 10a and 10b. The separation between the camera centers can vary in the range 4.5 cm to 58 cm; and still larger spacing can be attained, e.g., 6 ft, with the cameras angled inward. The camera optical axes are parallel and perpendicular to a baseline connecting the two cameras, and the lens optical centers are 5 approximately 116 cm above a level surface. The surface is preferably light-gray in color. Images of 768 x 576 pixels at 256 gray levels are acquired using an IMAQ 1408 framegrabber card. The object may be illuminated using two 500 W halogen lamps positioned near the cameras.

Generally, two classes of projection are considered in planar geometric projection — perspective, and parallel or orthographic projection. In the case of perspective projection, distant 0 objects appear smaller than those nearby, and is characterized by a point known as the center of projection. FIG. 11 shows a perspective projection in which object points are projected through the image or view plane to a point known as the center of projection or focal point. The location of the projected point on the image plane is given by: u = (fl(z+d))x v = (fl(z+d))y Equation 2 In parallel or orthographic projection, the lines of projected rays are assumed to be parallel, where the location of the projected point on the image plane is given by: u = x v =y Equation 3

Stereopsis, binocular stereo, and photogrammetry, all refer to a process of judging distance by observing feature differences between two or more images usually taken from different locations under similar lighting conditions. To interpret a stereo pair, it is necessary to recover a transformation between the two camera coordinate systems.

FIG. 12 shows a schematic representation of the optical geometry used in the method of stereo triangulation. The distance, or range, of an image feature from the view plane may be determined from the corresponding locations of any projected feature, e.g., the projected laser line, within the respective image planes of the two parallel cameras. Assuming the camera spacing (d) and camera focal lengths (/) to be fixed, the distance to the feature may be derived (using similar triangles) from, z = dfl(u r u_τ) Equation 4 wherein the term (uv u_r) is referred to as the image disparity.

From Equation 4, it can be readily observed that:

• the distance (z) is inversely proportional to the disparity; the distance to near objects can therefore be measured accurately, while the distance to far off objects cannot;

• the disparity is directly proportional to the separation of the cameras, d; hence, given a fixed error in determining the disparity, the accuracy of z (depth) determination increases with increasing d; and,

• the disparity is proportional to the lens focal length,/; this is because image magnification increases with an increase in focal length.

From the above, it is clear that the greater the camera separation (d), the greater the disparity, and the better the accuracy in the determination of z. However, as the separation of the cameras increases, the two images become less similar. This is sometimes known as wide-angle stereo, i.e., there is likely to be less overlap between the two fields of view. For example, some objects imaged by one camera may not be visible to the other. This leads to a breakdown in tlie method. Also, it is more difficult to establish correspondence between image points in wide-angle stereo. The difficulty in applying stereo triangulation arises in reliably determining the corresponding features within the two separate images. The key to an automated stereo system is a method for determining which point in one image corresponds to a given point in another image.

Utilizing an invariant moment analysis method for the determining an object's length and breadth, the ratio of the object's principal axes may be derived. If the object is assumed to be a cuboid, then the length and breadth (in addition to the location of the centroid, and the orientation of -l i¬

the principal axis) can be determined in units of pixels. To express these dimensions in real world units, e.g., cm, it is necessary to calibrate the system. That is, to establish the size of an image pixel in world units. For an obj ect at a fixed distance, this may readily be done by first acquiring an image of a similar object of known size. However, in the current application, the distance to the top of the 5 cuboid obj ect is a variable, which is dependent upon the obj ect' s height. Thus, two cuboid obj ects of equal length and breadth, but differing height, can appear to differ in all three dimensions. It is therefore necessary to introduce a calibration factor in terms of the variable z: calibrated dimension = pixel dimension * (pixel size * range (z) I lens focal length (/))

Since the fixed position of the cameras is known, the object height may be determined using 0 Equation 4. To achieve this, it is necessary to solve the correspondence problem, i.e., to find an object feature, or more specifically an object point, that is visible in both camera images. This pair of image points is sometimes known as a conjugate pair. Several techniques have been reported in the scientific literature for undertaking this task, including correlation methods, gray-level matching, and edge-based methods. One solution is to utilize the projected laser in each view to form the 5 conjugate pair.

As shown in FIG. 13, the primary image processing stages are: acquisition, i.e., the capture of stereo gray level images; pre-processing, i.e., convolution filtering to improve edge definition, etc.; blob, e.g., object, segmentation, i.e., using a fixed or adaptive threshold; and feature extraction, i.e., determination of principal dimensions. o The feature extraction stage includes the determination of obj ect height in world coordinates, e.g., cm; length and breadth in image coordinates, e.g., pixels; and length and breadth in calibrated world coordinates, e.g., cm.

To further understand the present invention, the results and analysis of a method utilizing scanning laser light and vision system techniques for determining the height of a cuboidal object is 5 presented. It is to be understood that the present invention is not to be limited to these results and analysis. A geometrical analysis was performed to allow for parallax and perspective effects. The technique produced accurate height values. For boxes placed directly under the camera, errors in the measurements were less than the variation in height across the width of the box. For example, an 18 cm height box was moved by 50 cm in the x and y directions, and the corresponding height value o was 17.9 cm. Therefore, for this analysis, maximum errors in height determination were less than

+/-1%.

The system comprised a laser apparatus having a Class II laser diode (635nm) with a cylindrical lens producing a plane of light with full divergence angle of 60° and a precision scanner with mounted mirror utilizing drive electronics tuned to mirror. The orientation and location of the 5 scanner and mirror can be adjusted as required for the application. Also included in the system were instrumentation and control apparatus including an input/output card, framegrabber card, cabling, and connections. The software included Lab VIEW 61 withNI-DAQ software (used for controlling the mirror) and IMAQ software (for image acquisition and analysis). Additional equipment comprised: 512x512 gray-scale camera (pixels 11 micron x 11 micron), HP Vectra PC, and cuboidal boxes of various dimensions. The measurement surface on which the boxes were placed was painted matte black.

The configuration of the system is shown in FIG. 14 wherein Hm = 190 cm, He = 139 cm, and Lo = 116 cm. Three possible locations for a box are shown in FIG. 14. A geometrical analysis was performed for the two general cases shown, i.e., placement of the box 36 at position 1, and at position 2. FIG. 14 is a schematic side-view of the experimental arrangement of the mirror, camera, and box (shown at three possible locations). Many of the angles and dimensions that need to be found for the determination of box height are shown in FIG. 14. FIG. 15 shows detail of the triangle formed by the camera and the point at which the laser light impinges on the box and on the surface. For this triangle, the Sine Rule states, d/sin(D) = i/sin(I) d = i-sin(D)/sin(I) Since the sum of the internal angles for a triangle is 180°,

I = 180 - D - (A + E) Also, from the Theorem of Pythagoras, i = ((LI)² + (He)²)^{0 5} d = ((LI)² + (Hc)²)°-⁵)sin(D)/sin(180 - D - (A + E)) It can also be seen from FIG. 15 that, cos(A) = hel/d hel = d-cos(A) Therefore, hel = ((LI)² + (Hc)²)⁰'⁵)sin(D)-cos(A)/sin(l 80 - D - A - E) Equation 5

Equation 5 can be used when the horizontal distance from the mirror to the box is less than Lo. Similarly, for a box positioned at position 2, he2 = ((L3)² + (Hc)²)°-⁵)sin(H)cos(C)/sin(l 80 - H - C + G)) Equation 6 Equation 6 can be used when the horizontal distance from the mirror to the box is greater than Lo.

Equations 5 and 6 can therefore be used to determine the height of a box, assuming that the laser light can be seen as it impinges on the top of the box and on the surface. This would be seen at the camera as two lines of light. Further, due to the uncertainty as to the color and texture of the surface that is utilized in the warehouse environment, it is desirable if tl e height of the box could be determined without the need to detect the laser light as it impinges on the adjacent floor surface of the measurement space. Black rubber matting has a tendency to reflect a minimal proportion of the incident light, so that good 5 imaging of the line may not be possible. It is further desirable if the height of the object could be determined purely from analysis of the line of laser light visible on the top of the obj ect. This can be achieved due to the high level of accuracy and repeatability attainable from the scanner that is used for positioning the mirror. The rotational position of the mirror is proportional to the voltage supplied to the scanner's drive electronics. Lab VIEW software is utilized to supply a number of l o voltages and the corresponding position of the laser line on the table can be measured. Trigonometry is used to relate this to the angle of the mirror, A. Solving the resulting simultaneous equations allows for the angle of the mirror to be calibrated in terms of applied voltage using, for example, the following equation:

A = 1.964(V) + 17.94 Equation 7

15 where V is the applied volts.

For a given voltage applied to the scanner, it is possible to predict the position of the laser line on the floor surface. This position is quantified in terms of the y-pixel coordinates of the centroid of the line, as viewed at the camera. The camera was arranged such that y-coordinate values increased as the line moved to the left side, as shown in FIG. 14. This pixel value does not 20 vary linearly with the angle of the mirror, A, however it may be expected to be proportional to tan(A). Therefore, the mirror can be positioned to various angles and noting the corresponding pixel values. Solving the simultaneous equations yields the following: pixel y-value = -1020.43(tan(A)) + 883.32 Equation 8

Most of the values needed in Equation 5 to calculate the height are available wherein LI can 25 be found from the geometry shown in FIG. 14, the equation is:

Ll = Lo ~ Hm(tan(A)) The determination of the angle D, which is the angle subtended at the camera lens by the pixel y-value of the laser line that impinges on the floor surface (determined using Equations 7 and 8 for a given voltage, V, and the pixel y-value of the laser line that impinges on the top of the box 3 o (found from analysis of the image).

The angle D can be determined through analysis of the paths of light that pass through the camera lens and impinge upon the charge coupled array. This is shown in FIG. 16 for detection of the height of the box at position 1. From FIG. 16, q = yi - yθ where y 1 is the predicted y pixel value for the laser light which impinged on the floor surface, and yO is the y value of the central pixel, i.e., 256. Also, q + r = y2 - yO where y2 is the y-pixel value for the line on the top of the box. As explained above, yO, y 1 , and y2 can be found, and therefore q and r can be determined, p is the focal length of the lens, e.g., p = 8 mm. Therefore, t can be found from the Theorem of Pythagoras, the Cosine Rule states, cos(D) = (t² + s² - r²)/2ts Equation 9

The above formula provides for determining the angle D. This can then be combined with the other derived parameters, and substituted into Equation 5 to give the height of the box, hel .

In one example, He, the height of the camera above table, is 139 cm. Hm is the height of the scanner mirror above the table and is 190 cm. Lo is 116 cm and is the orthogonal distance from the scanner mirror to the table. A voltage of 4.9 V applied to the scanner driver provides the mirror an angle A of 27.56°. E was determined to be 6.9°, and LI to be 16.83 cm. A box was placed on the surface and the measured value for y2 (the y pixel of the centroid of the line on top of the box) was found to be 389.8. The value for y 1 (the predicted y value for the centroid of the line on the floor) was 350.71. The value for yO, the center pixel in the camera's field of view, is 256. q = yi - yθ = 350.71 - 256 = 94.7 pixels thus, q = 1.04 mm (1 pixel has a side length of 11 microns) q + r = y2 - yO = 389.8 - 256 = 133.8 pixels = 1.4718 mm

Therefore r = 0.43 mm. p, q, and r can be used to find t and s: t = (p² + q²)⁰⁵ (p is the focal length of the lens, e.g., 8 mm.) thus, t = 8.067 mm s = (p² + (q + r)²)⁰-⁵ thus, s = 8.134 mm

Entering these values into Equation 9 yields a value for angle D of 3.005°. By substituting this value into Equation 5, along with the other values given above, the value of hel was determined to be 10.67 cm. The measured height of the box was found to be 10.6 cm.

An accuracy check of the laser line method for height measurements of a box at a significantly different point in the field of view reveals that a change in the position of the box in the camera's field of view has any significant effect on the accuracy with which the height can be determined using the scanning laser line technique. Again, using the 8mm lens, a box was placed at a displacement of 40 cm in both x and y directions from the camera. The line of light impinged on the top of the box when a voltage of 3.9 V was applied to the scanner driver. Calculations showed that A = 25.6°, L1 = 24.97 cm, D = 5.3797°, and E= 10.18°. From these values, hel was determined to be 17.9 cm. This compares with a height value from direct measurement with a rule of 18 cm; giving an error of 0.55%.

The line scanning technique described here offers a number of advantages in addition to high accuracy height measurement. For example, image analysis is simplified since at any given time the image captured is simply that of the line section which is impinging on the top of the box, and the orientation of this line relative to the camera does not change. A combination of such images (for different mirror angles) can be used to determine the length, width, and location of the box, as described earlier. Due to the large amount of information provided during the scan, the technique also offers potential for quantification of the morphology of more complex shaped objects. Various techniques can be implemented to reduce the scanning time and amount of memory typically required in dimensioning systems. Some of these techniques include a quick scan of each image to determine if any object, i.e., line segment, is present. If not, then that image would be immediately discarded. Also, coarse scanning of a plane of light could be utilized for position detection, followed by finer scanning for determination of the four sides of the object. The measurement density required will depend upon the resolution required from the system. For example, if the smallest object that the system needs to detect is a cube of side length 30 cm, then it would not be necessary to scan the line across the floor in intervals of less than approximately 25 cm. If further reductions in time are required, conventional image processing could be combined with line scanning. The direct image processing might quickly give the centroid of the plan view of the box, (along with its length and width). The laser line would be directed to the image centroid, and then scanned until an image of the line on top of the box was attained. Processing of this one image would then give tlie box height. Such a system may generally be expected to allow determination of the box parameters in very short time, e.g., less than one second.

Perhaps one of the more formidable difficulties to be overcome in vision system box measurement is associated with thresholding and field of view. By means of adjusting the camera aperture or integration time, and application of a suitable threshold, it is possible to obtain images consisting of only the laser line as it passed over the top of the object. However, when the intensity of the background light increases, other features become visible, such as reflections of daylight from the floor, and from plastic tape present on the boxes. These effects can be avoided by utilizing an infrared laser with a filter placed on the camera lens so that only the laser light would be visible to the CCD array.

The active nature of the structured lighting approach has significant advantages over more passive lighting techniques, particularly given possible complexity in object shape, and the already 5 relatively unstructured nature of the environment, i.e., difficulty in controlling ambient lighting, and variation in object position and size. Shadowing problems may be alleviated by moving the laser closer to the camera (with some reduction in accuracy) or simultaneously scanning from opposing directions. FIG. 17 depicts this configuration, although deep recesses will remain difficult to recover. o Alternatively, as shown in FIG. 18 , stereo triangulation in cooperation with a scanning laser mounted near the camera(s) can be utilized to determine range. This configuration reduces the problem of shadows while again taking advantage of structured lighting to simplify the image analysis. It might be possible to determine object position, length, and width by initially using a single uniformly illuminated image together with method of moments, and then actively directing 5 the laser to, and locally scanning across, the object to recover the height profile using the triangulation method. Such a system is a hybrid of both passive (relatively unstructured) and active (structured) illumination, attainable perhaps by utilizing a dual image threshold.

Alternatively, when capable of segmenting the object by tlπesholding, determining the height of a more complex obj ect is simplified by utilizing a second camera viewing the obj ect horizontally. o One such configuration of a two-camera system is shown in FIG.27. The second camera is mounted at approximately 60° from the horizontal. This type of configuration may require a type of tomographic approach, or a form of stereo, to find the dimensions of the object.

Another aspect of the present invention involves a simple method for correcting the location of image points when subject to radial lens distortion. The approach requires only two calibration 5 images to establish the necessary distortion coefficients.

Given the desire to achieve a relatively compact overall working volume of the dimensioning system 10, it may be preferable to view large objects at relatively close proximity. A wide angle of view may be achieved by using a lens of short focal length, e.g., less than 12 mm, however, this is at the cost of some image distortion, sometimes known as "barrel distortion." Radial lens distortion 0 can be approximated mathematically; however, as related by Scliluns and Koschan, it becomes difficult to reliably model the distortion given inevitable variations in lens quality. An ideal model of lens distortion leads to an infinite number of distortion coefficients.

FIG. 20 depicts an undistorted image of a square and FIG. 21 depicts an image subject to considerable radial lens distortion in which the corners of the distorted image are observed to be projected towards the center of the image. Notice also that the distortion can be reversed, in which case the corners of the image are now projected away from the image center, as shown FIG. 22.

A reasonable approximation of the lens distortion may be obtained by considering only two coefficients, C_t and C₂. Consider a coordinate frame located at the center of the image shown in FIG. 23. Let and y be the distorted image coordinates, and x_u and y_u be the undistorted image coordinates, for which: x_u = x_d(l+Cj(x_d ²+y_d ²) +C₂(x_d ²+y_d ²)²) and y_u = y_d(l+C₁(x_d ²+y_d ²) +C₂(x_d ²+y_d ²)²) The distortion coefficients, d and C₂, can be determined by suitable calibration. If Ci or C₂ are positive, then the image is projected in towards the center, and conversely if negative, the image is projected out away from the mage center.

To calculate Cι and C₂, distorted images of two objects of differing size are utilized. The objects are positioned at the center of the field of view. Given that the image distortion tends to increase towards the edge of the image, one of the objects is chosen to be quite large, in relation to the field of view. The distorted images of a square of 100 pixel and 150 pixel are shown in FIGS.24 and 25, respectively. Preferably, the objects are square-shape so that the corner features might readily be identified. The coordinate location of the top left corner of each square is measured, relative to the center of each image, and found to be (-45, 45) and (-60, 60), respectively, where the corresponding undistorted coordinates are (-50, 50) and (-75, 75), respectively. (Image size 200 x 200 pixels with coordinate frame located at image center.) Thus, -50 = -45(1 + 4050Cι + 16.4x10⁶C₂) and,

-75 = -60(1 + 7200Q +51.84xl0⁶C₂) Solving these simultaneous equations yields Ci = 1.8xlO^"5 and C₂= 2.3xl0^"9. Further, x_u = x_d(l+1.8xl0^"5(x_d ²+y_d ²) +2.3x10^"9(x_d ²+y_d ²)²) and, y_u = yd(l+1.8xl0-⁵(x_d ²+y_d ²) +2.3x10-⁹(x_d ²+y_d ²)²) For a distorted image of a square of 180 pixels shown in FIG. 26, the measured x-coordinate of the upper left corner was found to be -67 pixels. (Image size 200 x 200 pixels, with coordinate frame located at image center.) This gave a calculated undistorted location of -90.25 pixels, which compares favorably with the actual undistorted location of -90 pixels.

This relatively simple approach provides a useful mechanism for the correction of the location of image points when subj ect to significant radial lens distortion. The distortion coefficients can be determined during calibration of the dimensioning system and stored in a look-up table for access during the dimensioning process. Alternatively, using aspherical lenses may also reduce the effects of "barrel" distortion.

Another alternative to correcting for the lens distortion is to create equations or look-up tables to compensate for the distortion. The laser signal is scanned over the entire measuring region in very fine increments. At each position of the laser, through mathematical modeling using the known angle of the laser, relative camera and laser positions, and ideal lens properties, the theoretical placement of the signal on the sensor array can be determined. Images are gathered at each laser position by the camera. A comparison is made between the theoretical value the pixel 5 should have, and the actual value detected during the measurement. From the resulting data, a look-up table can be generated that indicates pixel correction values for each pixel.

An alternative method of removing distortions requires scanning the measurement in relatively small, predetermined increments. The x-coordinate field is segmented into multiple segments, e.g., 10. A mean y-coordinate value is determined fro each segment and each scan. 0 Creating sets of (x, y) data where the x value represents the voltage increment of the laser, and the y-value represents the spatial y-position of the laser in the image, polynomial line-fitting routines are used to create equations that describe a baseline voltage-laser relationship for the image. This baseline measurement effectively provides information that, when compared with expected values, is used to remove distortions. 5 A graphical interface for a cuboidal and non-cuboidal obj ect dimensioning system is depicted in FIGS. 27 and 28, respectively. The cuboidal system also incorporates a second screen (not shown) that displays the scanning laser line as it traverses across the object. The graphical window for the non-cuboidal system also displays the scanning laser line, as well as an image representing the entire scanned object surface, with a superimposed minimum enclosing box. o FIG. 29 is a photograph of one embodiment of the dimensioning system hardware. A frame constructed of aluminum supports the laser scanning unit and a camera. The laser control electronics and computer system, including I/O and frame grabber cards, are shown near the left side of the photograph.

Operation of the object measuring system is based upon the concepts and methods described. 5 For the non-cuboidal system, the height of the object is continuously determined during the laser scan of the object and then, on completion of the scan, the object's length and width are determined. In total, a cloud of 442,368 three-dimensional data points are typically acquired during a single scan. By calculating the object's height during the scan, it is possible to selectively remove low-lying points — often representing a pallet — from the point cloud data. The dimensioning system o incorporates a short focal length lens (6.5mm) to allow obj ects ranging in size from 12 in.³to 96Hx

72 L x 72 W in. to be measured using a system height of only approximately 186 inches. The camera utilizes a narrow band interference filter to eliminate ambient light.

The system 10 was implemented by employing a program written using National

Instrument' s CVI software, a C-based programming language that incorporates specialized functions 5 for data and image acquisition and processing. In determining the dimensions of a cubodial object, the dimensioning system utilizes a saw-tooth waveform generator (with a suitable I/O card) to produce an analog voltage. At specified intervals, the voltage is sent to the scanner electronics and used to drive the mirror to a known position. A "framegrabber" is then used to grab an image using the CCD camera attached to the system. The capture of an image while the mirror (and therefore line) is momentarily stationary, reduces and/or eliminates any possible errors caused by movement. The image is subtracted from a previous image, and then thresholded to produce a binary image. The height of each point is calculated using the previously described methods. The points of all scans are combined into a new image "cloud."

During determination of the dimensions of a non-cuboidal object, the dimensioning system continually calculates the height of all three-dimensional pixel points during the laser sweep of the measuring volume. This allows any background objects, such as a pallet or any markings on the floor, etc., to be removed from the cubing task. For example, the system may delete all pixels below 6 cm in height. As shown schematically in FIG.30, the remaining pixels are accumulated to form a three-dimensional cloud of data points representing the surface of the scanned object(s). Object maximum and average height are calculated during the laser sweep. Object length and width are calculated by fitting a "minimum enclosing rectangle" to a plan view of the data point cloud, as shown in FIG. 31.

Determination of the minimum enclosing rectangle is acquired by using the earlier described techniques, see FIG. 6, in which the enclosing rectangle is effectively rotated through a series of angular increments, e.g., 3°, until the smallest — interms ofarea — enclosing rectangle is found. If the smallest dimension, i.e., object width, and the dimension perpendicular to this, i.e., object length, are found. Although the enclosing rectangle will have the smallest width — rectangle A in FIG. 32 — it may not have the smallest area. Alternatively, the solution may be to find the enclosing rectangle with the smallest area — rectangle B in FIG. 32. The system 10 of the present invention is able to accurately determine the height of any object. This is due to the geometrical analysis and calculations that were performed to take into account the effects of parallax and perspective.

In one embodiment of the present system, height data is continuously calculated and used to find the maximum and average height values during the laser scanning cycle. The maximum height is sensitive to disturbance from noisy outliers and may cause a reduction in measurement accuracy.

Alternatively, the point cloud data can be accumulated and stored during the laser scan and then subsequently analyzed. A further advantage allows three-dimensional cloud of data points to be displayed with the minimum-enclosing cube superimposed, offering better visualization of the cubing task. Outliers and noise can be more readily deleted from the body of acquired data, possibly using global methods such as erosion and dilation. The duration of the scan could further be reduced by only considering pixels at or behind the advancing scan line, i.e., floor level. Time taken for the analysis of the data itself could also be improved by only considering object edge or boundary pixels during the cubing of the point cloud data.

In general terms, the more distant the lasers and cameras from the object, the greater the tendency toward orthographic project. While this helps to reduce occlusion, the laser signal will tend to be reduced in intensity, and the system accuracy reduced. Similarly, positioning the laser and camera units in close proximity will also tend to reduce occlusion, but at the cost of a reduction in system accuracy. These issues can be addressed by utilizing appropriate subsystem configurations of the present invention. FIG. 30 depicts an alternate embodiment of the dimensioning system's hardware configuration adopted for the reduced laser and camera occlusion — sometimes referred to as shadowing. This arrangement represents a compromise in terms of minimizing camera and laser occlusion while simultaneously offering reasonably optimized dimensional recovery accuracy when combined with a look-up table calibration approach previously described. By locating the laser units outside the cameras, the lasers tend towards an ideal collimated source, helping to minimize possible occlusions off to one side, i.e., away from the axis of the horizontal mounting rail.

In terms of hardware, there are now two sub-systems, i.e., there are two cameras and two lasers. However, from an operational standpoint, there are actually four sub-systems available. Table 1 lists the hardware components of the four operational sub-systems.

Table 1

Together, the four sub-systems offer differing operational characteristics that the controlling software may call upon during a given measurement cycle. For example, sub-systems 1 A and 2A behave as the existing overhead dimensioning system, but with differing fields of view. When operating together, for an object positioned centrally below, they are able to reduce the problem of laser and camera occlusion. The accuracy of sub-systems 1A and 2A can be improved across the field of view by the addition of the look-up table calibration approach. Alternatively, sub-systems IB and 2B have a much greater baseline separation and are thus able to offer significantly improved accuracy of height determination, although at the cost of increased laser occlusion.

It can be observed that the determination of the object's maximum height does not suffer from the problem of occlusion, therefore sub-systems IB and 2B are able to provide increased accuracy for this purpose. On the other hand, sub-systems 1 A and 2A have the ability to recover occluded areas and thereby improve accuracy in the determination of the object's length and breadth.

Thus, the sub-systems offer a hybrid approach to the dimensioning system.

Generally, objects to be dimensioned are nominally placed on a floor mark, i.e., measurement space, located centrally between the two cameras. The central placement reduces occlusion issues,

5 although objects located between and at the periphery of both camera fields of view can be disadvantageous due to radial lens distortion — with any registration errors being more significant.

The dimensioning process begins by scanning laser 1 rapidly through the measurement space.

During the rapid scan, cameras 1 and 2 determine the approximate location and extent of the object.

Laser 1 is scanned over the object and cameras 1 and 2 (sub-systems 1A and 2B) acquire point o cloud data simultaneously. Laser 2 is scanned over the object and cameras 1 and 2 (sub-systems IB and 2A) acquire point cloud data simultaneously. The point cloud data acquired by the sub-systems is merged and fit in a cuboid. It is to be understood that the acquisition of point cloud data can be attained by multiplexing these steps to gain a speed advantage. Furthermore, it may also be possible to apply a transformation when merging the cloud data to accommodate any mis-registration. s To combat accuracy errors, e.g., distortion or mis-registration, arising from objects placed between the cameras near the periphery and between of the two fields of view, the configuration shown in FIG. 33 can be arranged as shown in FIG. 34 wherein the cameras are pointed toward the central object location. In this configuration, it is necessary to perform transformations upon the acquired point cloud data to map data acquired in the local coordinate frames to a common world o coordinate frame. However, to provide the same combined field of view (with reduced occlusion) as obtained with parallel optical axes, the camera spacing should be increased. Also, to avoid a reduction in accuracy caused by a reduction in camera-to-laser separation, the lasers can be further separated, although this may result in a fall-off of reflected intensity. Because both cameras have a view of much of the object, a stereo vision approach can be incorporated. 5 Another embodiment of the present invention shown in FIG. 35 can significantly reduce occlusion by utilizing a four camera-laser system set-up configured in 90° increments. A less costly configuration is shown in FIG. 36 and incorporates a three camera-laser system set-up arranged in 120° increments and the local data mapped to a world coordinate frame.

While the specific embodiment has been illustrated and described, numerous modifications o come to mind without significantly departing from the spirit of the invention and the scope of protection is only limited by the scope of the accompanying Claims.

Claims (67)

CLAIMS We Claim:

1. A method for determining the dimensions of an item, or group of items, 5 placed within a measurement space, the method comprising the steps of: determining the approximate location and extent of an item; acquiring a first set of point cloud data by utilizing a first laser to transmit a first signal over the item and utilizing a first camera to receive the reflection of the first signal; constructing a three-dimensional array that defines the item from the first o set of acquired point cloud data; and, determining a rectangular prism to contain the constructed array, the rectangular prism having a height, length, and breadth.

2. The method of Claim 1 further comprising: s utilizing a second laser and a second camera to determine an approximate location and dimension of the item; acquiring a second set of point cloud data by utilizing the second laser to transmit a second signal over the item and utilizing the second camera to receive the reflection of the second signal; and, o constructing a three-dimensional array that defines the item by merging the first and second sets of acquired cloud data.

3. The method of Claim 2 further comprising: acquiring a third set of point cloud data by utilizing the second camera to 5 receive the reflection of the first signal; and, constructing the three-dimensional array that defines the item by merging the first, second, and third sets of acquired cloud data.

4. The method of Claim 3 further comprising: o acquiring a fourth set of point cloud data by utilizing the first camera to receive the reflection of the second signal; and, constructing the three-dimensional array that defines the item by merging the first, second, third, and fourth sets of acquired cloud data.

5 5. The method of Claim 1 further comprising: compensating for lens distortion of the constructed three-dimensional array.

6. The method of Claim 5 wherein compensating for lens distortion comprises: utilizing an pixel point correction value in cooperation with the acquired first set of point cloud data to adjust the location of each pixel point affected by radial lens distortion, wherein the pixel point correction value being determined during calibration of the dimensioning system. 0

7. The method of Claim 6 further comprising: providing a pixel value for a pixel within the measurement space; acquiring a scanned pixel value by utilizing the first laser to transmit the first signal over the measurement area and utilizing the first camera to receive the reflection off 5 the pixel of the first signal; comparing the pixel value with the scanned pixel value; and, generating a pixel correction value in response to the comparison.

8. The method of Claim 7 further comprising: o storing the pixel correction value in a calibration look-up table, wherein the pixel correction value can be utilized during construction of the three-dimensional array.

9. The method of Claim 7 further comprising: utilizing the pixel correction value to generate an equation for correcting 5 distortions.

10. The method of Claim 1 further comprising: reducing noise from the image.

0 11. The method of Claim 10 wherein reducing noise form the image utilizes image subtraction.

12. The method of Claim 10 wherein reducing noise comprises: acquiring a first array that represents the item by utilizing the first laser to transmit the first signal over the measurement area and utilizing the first camera to receive tlie reflections off the measurement are of the first signal; acquiring a second array that represents the item by utilizing the first laser to transmit the first signal over the measurement area and utilizing the first camera to receive the reflections off the measurement are of the first signal; subtracting the second array from the first array; and, utilizing the gray-level image as a threshold value for providing a binary image. 0

13. The method of Claim 10 wherein reducing noise comprises: determining a median pixel value for a predetermined area surrounding a pixel; and, setting each pixel to its respective median pixel value. 5

14. The method of Claim 10 wherein reducing noise comprises: computing a spatial histogram of the point cloud data in a vertical direction; computing a spatial histogram of the point cloud data in a horizontal 0 direction; grouping points having a spatially detached value; comparing the amount of points in a grouping against a predetermined value; identifying each grouping having a lesser amount of points than the 5 predetermined value; and, removing each identified group.

15. The method of Claim 14 wherein reducing noise comprises : computing a vertical spatial histogram from rotation of the point cloud o data in an x-plane; and, computing a horizontal spatial histogram from rotation of the point cloud data in a y-plane.

16. The method of Claim 10 wherein reducing noise comprises: 5 identifying points in a point cloud, each point having a height; grouping the points by the height of each point; comparing the amount of points in each group against a predetermined value; identifying each grouping having a lesser amount of points than the 5 predetermined value; and, removing each identified group.

17. The method of Claim 10 wherein reducing noise comprises: identifying a position of each disjoint point in a measurement array; i o comparing a height value of each disj oint point against a height value of a surrounding signal; and, removing each disjoint point not matching the height value of the surrounding signal.

is 18. The method of Claim 1 further comprising: utilizing a point threshold in cooperation with the image during construction of the array.

19. The method of Claim 18 further comprising: 2 o identifying a gray-scale value for each acquired point; utilizing each identified point to determine a statistical property of the gray-scale value; and, defining the point threshold in response to the determined statistical property of the gray-scale value.

25

20. The method of Claim 18 further comprising: providing a group of point threshold values from which to select the point threshold, the group of point threshold values being determined in cooperation with calibration of the dimensioning system.

30

21. The method of Claim 1 further comprising: transforming the constructed array to a global coordinate system.

22. The method of Claim 1 further comprising: determining the dimensions of the rectangular prism by rotating a co-ordinate frame about the centroid of the constructed array through a plurality of angular increments; and, measuring a distance from the centroid to the edge of the item for each 5 angular increment.

23. The method of Claim 22 further comprising: storing each measurement; identifying a length measurement and a breadth measurement; and, 0 selecting a single length measurement and a single breadth measurement, wherein the selected measurements, in combination with the determined height of the item, compose the dimensions of the rectangular prism having the smallest volume, but which would contain the item.

5 24. The method of Claim 1 wherein acquiring a first set of point cloud data comprises: coarsely transmitting the first signal in a first direction at an off-center location within the measurement space; identifying a first edge of the item; o finely transmitting the first signal in a second direction over the first edge, the second direction being opposite the first direction; coarsely transmitting the first signal in the second direction at the off-center location within the measurement space; identifying a second edge of the item; and, 5 finely transmitting the first signal in the first direction over the second edge.

25. A system for determining the dimensions of an item, or a group of items, set within a measurement space, the system comprising: 0 a first laser being capable of transmitting a first signal through the measurement space, the first laser having a coarse transmission mode and a fine transmission mode; a first camera being capable of receiving the first signal; a first set of point cloud data being acquired by utilizing the first laser to transmit a first signal over an item and utilizing the first camera to receive the reflections of the first signal; an array generator for constructing an array from the first set of acquired cloud data; and, a rectangular prism generator for constructing a rectangular prism in response to the dimensions of the constructed item.

26. The system of Claim 25 further comprising: a lens distortion compensator for compensating for lens distortion of the constructed image.

27. The lens distortion compensator of Claim 26 further comprising: an image point correction factor being determined during calibration of the system, the image point correction factor being utilized in cooperation with the acquired first set of point cloud data to adjust the location of each image point affected by radial lens distortion.

28. The system of Claim 25 further comprising: a noise filter.

29. The noise filter of Claim 28 further comprising: a median pixel value being determined by an area surrounding a pixel; and, a designator for setting each pixel to its respective median pixel value.

30. The noise filter of Claim 28 further comprising: a vertical spatial histogram of the first set of acquired data from rotation of the point cloud data in a vertical direction; a horizontal spatial histogram of the first set of acquired data from rotation of the point cloud data in a horizontal direction; and, a grouper for grouping points having a spatially detached value, wherein each group having a lesser a lesser amount of points than a predetermined value is removed from the array.

31. The noise filter of Claim 28 further comprising: an identifier for identifying points in a point cloud, each point having a height; a grouper for grouping the points by the height of each point; and, a comparator for comparing the amount of points in each group against a 5 predetermined value, wherein each grouping having a lesser amount of points than a predetermined value is removed.

32. The noise filter of Claim 28 further comprising: o an identifier for identifying a position of each disjoint point in a measurement image; and, a comparator for comparing a height value of each disjoint point against a height value of a surrounding signal, wherein each disjoint point not matching the height value of the s surrounding signal is removed.

33. The dimensioning system of Claim 25 further comprising: a point threshold being determined by: identifying a gray-scale value for each point found in an image; o utilizing each identified point to determine a statistical property of the gray-scale value; and, selecting the point threshold in response to the determined statistical property of the gray-scale value.

5 34. The system of Claim 33 further comprising: a group of point threshold values from which to select the point threshold, the group of point threshold values being generated in response to calibration of the system.

35. The system of Claim 25 further comprising: 0 a second laser being capable of transmitting a second signal through the measurement space, the second laser having a coarse transmission mode and a fine transmission mode; a second camera being capable of receiving the second signal; a second set of point cloud data being acquired by utilizing the second 5 laser to transmit the second signal over the item and utilizing the second camera to receive the reflections of the second signal wherein the array generator utilizes the first and second sets of acquired point cloud data to construct the array.

36. The system of Claim 35 further comprising: 5 a third set of point cloud data being acquired by utilizing the first laser to transmit the first signal over the item and utilizing the second camera to receive the reflections of the first signal wherein the array generator utilizes the first, second, and third sets of acquired point cloud data to construct the array.

o 37. The system of Claim 36 further comprising : a fourth set of point cloud data being acquired by utilizing the second laser to transmit the second signal over the item and utilizing the first camera to receive the reflections of the second signal wherein the array generator utilizes the first, second, third, and fourth sets of acquired point cloud data to construct the array. 5

38. The system of Claim 37 further comprising : a third laser being capable of transmitting a third signal through the measurement space, the third laser having a coarse transmission mode and a fine transmission mode; 0 a third camera being capable of receiving the third signal; a fifth set of point cloud data being acquired by utilizing the third laser to transmit the third signal over the item and utilizing the third camera to receive the reflections of the third signal wherein the array generator utilizes the first, second, third, fourth, and fifth sets of acquired point cloud data to construct the array. 5

39. The system of Claim 38 further comprising : a sixth set of point cloud data being acquired by utilizing the first laser to transmit the first signal over the item and utilizing the third camera to receive the reflections of the first signal wherein the array generator utilizes the first, second, third, fourth, fifth, and sixth o sets of acquired point cloud data to construct the array.

40. The system of Claim 39 further comprising: a seventh set of point cloud data being acquired by utilizing the second laser to transmit the second signal over the item and utilizing the third camera to receive the reflections of the second signal wherein the array generator utilizes the first, second, third, fourth, fifth, sixth, and seventh sets of acquired point cloud data to construct the array.

41. The system of Claim 40 further comprising: 5 an eighth set of point cloud data being acquired by utilizing the third laser to transmit the third signal over the item and utilizing the first camera to receive the reflections of the third signal wherein the array generator utilizes the first, second, third, fourth, fifth, sixth, seventh, and eighth sets of acquired point cloud data to construct the array.

o 42. The system of Claim 41 further comprising: a ninth set of point cloud data being acquired by utilizing the third laser to transmit the third signal over the item and utilizing the second camera to receive the reflections of the third signal wherein the array generator utilizes the first, second, third, fourth, fifth, sixth, seventh, eighth, and ninth sets of acquired point cloud data to construct the array. 5

43. The system of Claim 25 further comprising: a first axis, the first camera and the first laser lying on the first axis; and, a second axis, the second camera and the second laser lying on the second axis. 0

44. The system of Claim 43 wherein the first and second axes are parallel.

45. The system of Claim 44 wherein both the first camera and the second camera are located between the first laser and the second laser. 5

46. The system of Claim 38 further comprising: a first perimeter, the first, second, and third cameras lying on the first perimeter; and, a second perimeter, the first, second, and third lasers lying on the second o perimeter.

47. The system of Claim 45 wherein the first, second, and third cameras are spaced 120° about the center of the first perimeter, and the first, second, and third lasers are spaced 120° about the center of the second perimeter. 5

48. The system of Claim 47 wherein the first and second perimeters are concentric circles, respectively, the first circle being contained within the second circle.

49. A computer readable medium having an application therein to facilitate 5 dimensioning of an item, or group of items, located within a measurement space, the medium comprising: a first segment for determining the approximate location and extent of an item; a second segment for acquiring a first set of point cloud data by utilizing a 0 first laser to transmit a first signal over the item and utilizing a first camera to receive the reflection of the first signal; a third segment for constructing an array from the first set of acquired point cloud data; and, a fourth segment for determining a rectangular prism to contain the s constructed image, the rectangular image having a height, length, and breadth.

50. The medium of Claim 49 further comprising : a fifth segment for utilizing a second laser and a second camera to determine an approximate location and dimension of the item; 0 a sixth segment for acquiring a second set of point cloud data by utilizing the second laser to transmit a second signal over the item and utilizing the second camera to receive the reflection of the second signal; and, an seventh segment for constructing the array by merging the first and second sets of acquired cloud data. 5

51. The medium of Claim 50 further comprising: an eighth segment for acquiring a third set of point cloud data by utilizing the second camera to receive the reflection of the first signal; and, a ninth segment for constructing the array by merging the first, second, o and third sets of acquired cloud data.

52. The medium of Claim 41 further comprising : a tenth segment for acquiring a fourth set of point cloud data by utilizing the first camera to receive the reflection of the second signal; and, an eleventh segment for constructing the array by merging the first, second, third, and fourth sets of acquired cloud data.

53. The medium of Claim 49 further comprising: a twelfth segment for compensating for lens distortion of the constructed array.

54. The medium of Claim 53 wherein the segment for compensating for lens distortion comprises: a thirteenth segment for utilizing an image point correction factor in cooperation with the acquired first set of point cloud data to adjust the location of each image point affected by radial lens distortion, wherein the image point correction factor being determined during calibration of the dimensioning system.

55. The medium of Claim 54 further comprising: a fourteenth segment for storing the image point correction factor in a calibration look-up table, wherein the image point correction factor being associated with an image point location.

56. The medium of Claim 49 fiirther comprising: a fifteenth segment for reducing noise from the image.

57. The medium of Claim 56 wherein the segment for reducing noise comprises: a sixteenth segment for determining a median pixel value for a predetermined area surrounding a pixel; and, a seventeenth segment for setting each pixel to its respective median pixel value.

58. The medium of Claim 56 wherein the segment for reducing noise comprises: an eighteenth segment for computing a spatial histogram of the point cloud data in a vertical direction; a nineteenth segment for computing a spatial histogram of the point cloud data in a horizontal direction; a twentieth segment for grouping points having a spatially detached value; a twenty-first segment for comparing the amount of points in a grouping against a predetermined value; a twenty-second segment for identifying each grouping having a lesser amount of points than the predetermined value; and, a twenty-third segment for removing each identified group.

59. The medium of Claim 56 wherein the segment for reducing noise comprises: a twenty-fourth segment for identifying points in a point cloud, each point having a height; a twenty-fifth segment for grouping the points by the height of each point; a twenty-sixth segment for comparing the amount of points in each group against a predetermined value; a twenty-seventh segment for identifying each grouping having a lesser amount of points than the predetermined value; and, a twenty-eighth segment for removing each identified group.

60. The medium of Claim 56 wherein the segment for reducing noise comprises: a twenty-ninth segment for identifying a position of each disjoint point in a measurement array; a thirtieth segment for comparing a height value of each disjoint point against a height value of a surrounding signal; and, a thirty-first segment for removing each disjoint point not matching the height value of the surrounding signal.

61. The medium of Claim 49 further comprising: a thirty-second segment for utilizing a point threshold during construction of the array.

62. The medium of Claim 61 further comprising: a thirty-third segment for identifying a gray-scale value for each acquired point; a thirty-fourth segment for utilizing each identified point to determine a statistical property of the gray-scale value; and, a thirty-fifth segment for defining the point threshold in response to the determined statistical property of the gray-scale value.

5

63. The medium of Claim 61 further comprising: a thirty-sixth segment for providing a group of point threshold values from which to select the point threshold, the group of point threshold values being determined in cooperation with calibration of the dimensioning system. 0

64. The medium of Claim 49 further comprising: a thirty-seventh segment for transforming the constructed array to a global coordinate system.

s 65. The medium of Claim 49 further including: a thirty-eighth segment for determining the dimensions of the rectangular prism by rotating a co-ordinate frame about the centroid of the constructed array through a plurality of angular increments; and, a thirty-ninth segment for measuring a distance from the centroid to the o edge of the array for each angular increment.

66. The medium of Claim 65 further including: a fortieth segment for storing each measurement; a forty-first segment for identifying a length measurement and a breadth 5 measurement; and, a forty-second segment for selecting a single length measurement and a single breadth measurement, wherein the selected measurements, in combination with the determined height of the item, compose the dimensions of the rectangular prism having the smallest volume, but which would contain the item. 0

67. The medium of Claim 49 wherein acquiring a first set of point cloud data comprises: a forty-third segment for coarsely transmitting the first signal in a first direction at an off-center location within the measurement space; 5 a forty-fourth segment for identifying a first edge of the item; a forty-fifth segment for finely transmitting the first signal in a second direction over the first edge, the second direction being opposite the first direction; a forty-sixth segment for coarsely transmitting the first signal in the second direction at the off-center location within the measurement space; a forty-seventh segment for identifying a second edge of the item; and, a forty-eighth segment for finely transmitting the first signal in the first direction over the second edge.