Camera measurement system

[0002] The Applicant claims the benefit of his Provisional Application, Ser. No. 60/200,062, filed Apr. 25, 2000.

[0003] The present invention relates to the field of tailoring and provides a measurement system, wherein a user in a remote location, or in a store, can quickly be measured for any number of garments using a camera.

[0004] Prior art tailor measuring systems require that the subject wear a special tight-fitting garment with target points on the garment, and full body scans or multiple cameras. Additionally, in the prior art, body features, such as the neck, had to be detected by hand or with specially marked points.

[0005] Development of a camera based measurement system which can automatically determine tailor body measurements with one camera, few pictures and without requiring the customer to wear special garments represents a great improvement in the field of tailoring and satisfies a long felt need of the public.

[0006] The present invention is a camera based measurement system which can automatically determine tailor body measurements with one camera, few pictures and without requiring the customer to wear special garments. The present invention will automatically detect the neck and other body features and can perform the function with a single camera, and a relatively small number of views, typically four.

[0007] This invention captures the silhouette of a user not his or her image. In a preferred embodiment, customers face the camera with their arms held away from their body at about 45 degrees. Preferably, customers should wear a swimming suit or any tight fitting garment for the silhouette making process. Although a camera takes a picture it is immediately converted to a silhouette to perform the calculations. The silhouette is then compressed into a very small file that contains only the customer's 9 body measurements. No one ever looks at or saves any actual photographs. The image that is captured is transient and is lost as the image is processed into a silhouette. This technology enables derivation of the nine traditional tailor measurements—height, neck, chest, waist, hip, seat, sleeve length, sleeve inseam and pant inseam—necessary to fit a garment to the human body. Collecting the data at a retailer who has installed the technology takes approximately two minutes.

[0008] Two silhouettes are needed: front and side. This technology measures customers faster, more consistently, more accurately and less intrusively than traditional hand measuring. The input and transmission of measurement data is standardized and automated, eliminating errors resulting from recording measurement data by hand.

[0009] The digital silhouettes become a cyber mannequin. This technology computes both two- and three-dimensional measurements from the data captured by a digital camera and represented by a silhouette. This data is electronically processed to match online retailer's garment specifications to facilitate the perfect fit.

[0010] When shoppers arrive at participating online apparel stores, armed with their nine body measurements—height, neck, chest, waist, hip, seat, sleeve length, sleeve inseam and pants inseam—their data is matched to the retailer's existing computerized pattern storage system. Instead of guessing which size to buy, shoppers will know that the garment they receive has been electronically selected to best match their body.

[0011] While it is possible for shoppers to download the software and instructions at home and create their own body measurement data file, participating retailers offer the measurement technology of this invention in the privacy of their dressing rooms.

[0012] An appreciation of the other aims and objectives of the present invention and an understanding of it may be achieved by referring to the accompanying drawings and description of a preferred embodiment.

[0013] FIG. 1 is a side elevational representation of a booth that can be used to practice this invention.

[0014] FIG. 2 is an overall flowchart of this invention

[0015] FIG. 3 is a chart providing more detail regarding the calibration step.

[0016] FIG. 4 is a chart providing more detail regarding the image collection step.

[0017] FIG. 5 is a chart providing more detail regarding the feature detection step.

[0018] FIG. 6 is a chart providing more detail regarding the tailor measurement step.

[0019] FIG. 7 is a flowchart of the feature detection step.

[0020] FIG. 8 is a flowchart of the calibration step.

[0021] FIG. 9 is a flowchart of data flow in this invention.

[0022] The present invention provides a camera measurement system, wherein a user in a remote location, or in a store, can quickly be measured for any number of garments using a camera. The process begins with the camera focusing on a plain background. Blue has been found to work well. Two images are recorded on film, digitally or by some other means. If the images are initially recorded optically on film, any one of a number of commercially available software packages can convert them into digital images (see FIG. 9). These first two views are taken without any subject or customer (person to be measured), one of the background, and one of a calibration target. Any number of objects can be utilized for a calibration target, however, it has been found that a T shaped target works particularly well. The target must have markings in units of measurement such as inches and fractions or centimeters and millimeters. In an alternative embodiment, the calibration and background images may be combined.

[0023] After the subject-less images are taken, a number of views of the subject are taken. Four has been found to work well. Those views are back, front, left side and right side. As described in more detail below, the digital image will be converted into a silhouette, and the silhouette used to detect certain body points (such as armpits and crotch, shoulders, hands, waist, neck, etc.). Once those body points are detected, the system compares distances on the image of the subject to the known distances in the calibration target and then calculates any number of measurements of the body.

[0024] The present invention can be practiced using a home digital camera or a home film camera 14. It has been found that using a 5′×6′×7′ booth with a 12′ camera-to-subject length yields excellent results. Mirrors 18, 22 may reduce the size of the booth considerably. See FIG. 1. The entire image capturing process presently takes about one minute to complete. There are four different images that need to be captured for each person 10.

[0025] Front Image:

[0026] The individual 10 stands looking straight ahead. Feet are approximately 12 inches apart. The arms should hang downward and slightly extended from the body. There should be about 8 inches between the side of the body and the arm itself. The hand should be clasped into a fist, with the thumb wrapped over the pointer finger.

[0027] Right Image:

[0028] The individual 10 should stand perfectly straight with their feet close together. The arms should hang directly to the side of the body. Posture is important here. The hands remained clasped into a fist, with the thumb wrapped over the pointer finger.

[0029] Rear Image:

[0030] The rear image mirrors the front view. The individual 10 stands with arms hanging downward, extending from the side of the body by about 8 inches. Hands are clasped into a fist, with the thumb wrapped over the pointer finger.

[0031] Left Image:

[0032] Follow the same procedure that was used for the right image. The individual 10 must stand with feet together, arms directly at each side. Hands are clasped with the thumb wrapped over the pointer finger.

[0033] The best description of this invention may be derived from the Figures. FIG. 2 provides an overall flowchart of this invention. FIG. 3 provides more detail regarding the calibration step, while FIG. 8 is a flowchart of this step. The five cross points of the calibration target are the four end points and the center. FIG. 4 provides more detail regarding the image collection step. FIG. 5 provides more detail regarding the feature detection step, while FIG. 7 is a flowchart of this step. The shoulder points are the sides of the neck, as seen from the front view, and the narrowest tilted portion of the neck as seen from the side view. The shoulder point is the point at which the change in slope of the shoulder is greatest. All measurements are off the centerline. Measurements, for example the waist measurement, maybe obtained by creating ellipses that pass through the waist detection points of each of the four views. The formula to determine these measurements from the multiple ellipses can be found in commonly available mathematical texts.

[0034] FIG. 9 is a flowchart of data flow in this invention. The preferred frame digitizer is Computer Eyes I RT Monochrome, Gray-scale Video Frame grabber—1992, available from Digital Vision, Inc., 270 Bridge Street, Dedham, Mass. 02026, (617) 329-5400. However, other frame digitizers are available and could be used. Lview Pro is the preferred are for converting JPG image files to BMP (bitmap). The BMP to CE format conversion is done by translating image pixel data for a line of 640 bytes with 8 bit monochrome intensity to a bit plane order from most significant to least significant as 8 groups of 80 bytes with 8 bits of plane data. Numerous algorithms are possible for this translation. RLE is a run length encoded compression of a silhouette or binary image on a line by line basis using the start and end of the image line segments. The inverse process is used to reconstruct a BMP format which is converted to a CE format. One good reference for mathematical functions used in the measurement process is “Numerical Recipes In C The Art of Scientific Computing”—Second Edition Cambridge University Press, 1992. Images of photos from digital scanners are in a line by line format. Each line is converted to 8 bit monochrome format of length 640 pixels. Line formats are converted to CE format as described above.

[0035] FIG. 6 provides more detail regarding the tailor measurement step. Once the measurements are obtained, they can be used in any number of ways. In one embodiment the present invention electronically supplies that information to one of any number commercially available automatic fabric cutting devices, providing the user with a made to measure garment. In another embodiment, the measurements are provided directly to the user or the user's clothier to properly select commercially available garments. In a third embodiment, the measurements are compared to the standardized patterns of one or more manufacturers, to determine which size garment from that particular manufacturer is appropriate to the individual user. In another embodiment, the user's measurements are available through an Internet site, from which the user can then go to Internet sites of various clothiers. In such an embodiment, when the user selects a garment, the present invention will automatically compare the user's measurements to the patterns used by that clothier and either inform the user of the correct size garment to order, or automatically insert the correct size in the appropriate place in the on-line order form.

[0036] In another embodiment of the present invention, the user may simply take his or her own photograph with a known reference target in the picture or take his or her picture and a separate photograph of a known reference target. These photograph(s) are then submitted to a central processing center, electronically digitized or in analog format, or by any other means including mail. The central processing center then scans the photograph(s), if necessary to convert into digital form, and then proceeds to process them as described above.

[0037] It should be clear that any number of images and views may be taken, and any number of measurements may be generated, from a single view. For example one can measure only the user's waist or make a large number of body measurements, in order to create a suit that exactly fits the user's body.

[0038] Feature detection and measurement computations are key components that are used to derive the many intricate body landmark points required for creating individual body measurements. The feature location and body measurement computation technology consists of a digital camera, a desktop PC and proprietary software. This proprietary software captures and registers multiple body images in three dimensions, locates feature points in multiple images, and computes body measurements using both linear and circular models. Reference is made to a visible body structure or landmark; an example would be the outer corner of the eye. Often axis intersect points are used as a beginning of a measurement. From these landmark body structures, geometric relationships are formed with other body landmarks, which define the required measurement. The garment size data is compiled from a number of these body measurements. This body measurement system reverses this process. The starting point is a surface represented by the points captured by a camera. The data from multiple two-dimensional digital images is combined to form an input stream to compute measurements. The final data is an input stream of three-dimensional numerical values representing the feature surface of the subject.

[0039] Camera image processing removes background surface data and transforms the image into planes of information from which silhouettes, or image edges, are derived. These silhouettes are catalogued by orientation to the axis of a body so that a hierarchy of feature detection algorithms can be applied to identify numerical points of data. This data represents the same landmarks described in the dictionary of measurements. Thus the detection of body features, which a tailor would look for or feel on a subject, are now found by an analysis of their data values. The body feature points are therefore the highest resolution unit of measurement.

[0040] Having established a vocabulary of landmark skeletal and musculature features in a digital language, the technology is capable of extracting the desired body measurements. This process applies the same geometrical relationships as described in the definitions of the measurement. From a given landmark, the measurement is taken in the prescribed geometric direction until the next digital landmark is contacted.

[0041] Thus image processing transforms the digital camera image of discrete points into a series of two-dimensional silhouettes, using planes to intersect the output. Fuzzy logic locates the appropriate body features in the silhouette planes, and a rule-based process is used to select feature points associated with the required measurements. To compute linear and circular tailor measurements from the landmark features, a unique combination of analytic geometry equations is applied.

[0042] The algorithms used in this invention are divided into two groups. First, the determination of the location of body feature points on binary silhouette images of the subject. Second the computation of tailor measurements using the feature point data and mathematical models of tailor measures.

[0043] Feature Point Location Algorithms

[0044] Five feature point silhouette images are formed by subtracting the background image from the subject image. The five subject images are as follows:

[0045] 1 Front view of subject

[0046] 2 Left side view of subject

[0047] 3 Rear view of subject

[0048] 4 Right side view of subject

[0049] 5 Front view of calibration fixture

[0050] The fifth image is used to determine the number of pixels per inch in the horizontal and vertical directions for calibration of the tailor measurements. Five points are located on the cross shaped calibration fixture. The points are the top, left, right, and bottom point defined by a lower tab on the fixture. The following example file is produced by this process that consists of four records with three fields per record. The first field is a designator where 9xxx represents calibration and 001 to 004 specifies the four points for extremes of the four cross points in image pixels where (0,0) is in the top left corner:

[0051] 9001 230 207

[0052] 9002 230 618

[0053] 9003 27 412

[0054] 9004 434 412

[0055] The process for locating these features is to search in a single direction until a large (as defined by fuzzy logic) black line of pixels is detected. The four search directions are top down, left to right, right to left, and bottom up.

[0056] The first image is the front view of the subject and can be searched for the following thirty six feature points as shown in the example file. Only eighteen or the thirty six point definitions are currently used. The unused spare points have negative values in fields two and three. Each point is defined by the first field which has 1xxx to indicate the front views and 001 to 036 to indicate the feature point. These are table driven and can be redefined in the software. For example, the first point is the top of head located at the highest large black line of pixels along the body center line.

[0057] 1001 195 146

[0058] 1002 181 239

[0059] 1003 231 239

[0060] 1004 145 356

[0061] 1005 270 353

[0062] 1006 151 416

[0063] 1007 266 416

[0064] 1008 140 473

[0065] 1009 280 473

[0066] 1010 135 515

[0067] 1011 282 515

[0068] 1012 206 256

[0069] 1013 284 279

[0070] 1014-32-32

[0071] 1015 209 557

[0072] 1016-32-32

[0073] 1017 208 416

[0074] 1018-32-32

[0075] 1019-32-32

[0076] 1020-32-32

[0077] 1021-32-32

[0078] 1022-32-32

[0079] 1023-32-32

[0080] 1024-32-32

[0081] 1025 235 256

[0082] 1026 178 256

[0083] 1027-32-32

[0084] 1028-32-32

[0085] 1029-32-32

[0086] 1030-32-32

[0087] 1031-32-32

[0088] 1032-32-32

[0089] 1033-32-32

[0090] 1034-32-32

[0091] 1035-32-32

[0092] 1036 337 526

[0093] The second image is the left side view of the subject and can be searched for the following twelve feature points as shown in the example file. Each point is defined by the first field which has 4xxx to indicate the left side views and 001 to 012 to indicate the feature point. These are table driven and can be redefined in the software. For example the first point is the top of head located at the highest large black line of pixels along the body center line.

[0094] 4001 193 152

[0095] 4002 179 271

[0096] 4003 228 255

[0097] 4004 151 326

[0098] 4005 258 326

[0099] 4006 143 416

[0100] 4007 234 416

[0101] 4008 141 458

[0102] 4009 254 458

[0103] 4010 147 505

[0104] 4011 266 505

[0105] 4012 203 263

[0106] The third image is the rear view of the subject and can be searched for the following twelve feature points as shown in the example file. Each point is defined by the first field which has 3xxx to indicate the rear view and 001 to 012 to indicate the feature point. These are table driven and can be redefined in the software. For example the first point is the top of head located at the highest large black line of pixels along the body center line.

[0107] 3001 197 145

[0108] 3002 174 237

[0109] 3003 225 237

[0110] 3004 133 352

[0111] 3005 268 349

[0112] 3006 143 415

[0113] 3007 258 415

[0114] 3008 132 466

[0115] 3009 270 466

[0116] 3010 128 512

[0117] 3011 277 512

[0118] 3012 199 237

[0119] The fourth image is the right side view of the subject and can be searched for the following twelve feature points as shown in the example file. Each point is defined by the first field which has 2xxx to indicate the right side views and 001 to 012 to indicate the feature point. These are table driven and can be redefined in the software. For example the first point is the top of head located at the highest large black line of pixels along the body center line.

[0120] 2001 175 144

[0121] 2002 163 246

[0122] 2003 212 262

[0123] 2004 137 322

[0124] 2005 244 322

[0125] 2006 161 416

[0126] 2007 254 416

[0127] 2008 142 458

[0128] 2009 256 458

[0129] 2010 135 505

[0130] 2011 252 505

[0131] 2012 187 254

[0132] All feature points in all images are located by the same general algorithm with only controls for direction and stopping rules. For example, the top of head is located by searching down from the top of the image until a large black pixel line is detected to determine the vertical location. The horizontal location is the middle point of the large line. Similarly, the crotch in the front and rear images is located by searching up from the bottom between the spread legs (two black lines separated by a long white line of pixels ) until a large black line is detected. These binary image searching methods are well known and described in numerous image processing texts and papers. The process is continued until all points are located in all images. The resulting points in pixel coordinates are written to the feature files as shown in the five examples above.

[0133] Tailor Measurement Algorithms

[0134] The tailor measurement process reads all five of the feature files and produces a measurement file as shown in the following example. The nine tailor measurement entry file consists of records that have an identifier field and a value field. For example, the first record is the subject height labeled 1001 with a value of 69.18 inches. The fields are labeled for clarity by “—name of tailor measurement” that does not appear in the file.

[0135] 1001 69.18—height

[0136] 3001 15.39—neck

[0137] 3002 38.13—chest

[0138] 4001 31.61—waist

[0139] 4002 35.32—jacket

[0140] 4003 37.67—seat

[0141] 7001 32.96—sleeve

[0142] 5001 30.67—inseam pants

[0143] 7002 16.59—inseam sleeve

[0144] The following is a listing of the computer program to compute all nine tailor measurements with each computational algorithm identified. For example, the height measurement 1001 is computed first by summing the four image features labeled n001 where n goes from 1 to 4 into variable sum and adding the offset of the calibration from the ground of thirty six inches. Each of the measurements is listed in order and the computation is easily understood by anyone skilled in the art of computer programming using visual basic. This is a test program developed to verify and validate the correct operation of the C/C++ program implemented in VMS for reasons of efficiency and speed. The computations agree identically to six decimal places.

[0145] Public Sub measurements( )

[0146] ′compute VMS measurements from features

[0147] Dim i, j, k, l, n As Integer

[0148] Dim sum, sumd, a, a1, b, c, arc, d, e, f As Double

[0149] Dim bb0, bb1, bb2, bb3, bb4, bb5 As Double

[0150] ′Measurement 1001—Height

[0151] n=1

[0152] mnum(n)=1001

[0153] mnam(n)=“Height”

[0154] sum=0#

[0155] 1=0

[0156] For i=1 To 4

[0157] k=i*1000+1

[0158] j=Feature(k)

[0159] Ifj=0 Then

[0160] Debug.Print “Missing Top of Head in View”; k

[0161] Else

[0162] sum=sum+yy(j)

[0163] l=l+1

[0164] End If

[0165] Next i

[0166] a=sum/1

[0167] j=Feature(1015)

[0168] sum=Sqr(yy(j)*yy(j)+36#)

[0169] sum=sum−yy(j)

[0170] a=a+sum

[0171] mm(n)=a

[0172] Debug.Print mnam(n); mnum(n); mm(n)

[0173] ′Measurement 3001—Neck

[0174] n=2

[0175] mnum(n)=3001

[0176] mnam(n)=“Neck”

[0177] bb0=distance(1002, 1003)

[0178] bb1=distance(2002, 2003)

[0179] bb2=distance(3002, 3003)

[0180] bb3=distance(4002, 4003)

[0181] a=pi*(bb0+bb1+bb2+bb3)/4#

[0182] mm(n)=a

[0183] Debug.Print mnam(n); mnum(n); mm(n)

[0184] ′Measurement 3002—Chest

[0185] n=3

[0186] mnum(n)=3002

[0187] mnam(n)=“Chest”

[0188] bb0=distance(1004, 1005)

[0189] bb1=distance(2004, 2005)

[0190] bb2=distance(3004, 3005)

[0191] bb3=distance(4004, 4005)

[0192] a=pi*(bb0+bb1+bb2+bb3)/4#

[0193] mm(n)=a

[0194] Debug.Print mnam(n); mnum(n); mm(n)

[0195] ′Measurement 4001—Waist

[0196] n=4

[0197] mnum(n)=4001

[0198] mnam(n)=“Waist”

[0199] bb0=distance(1006, 1007)

[0200] bb1=distance(2006, 2007)

[0201] bb2=distance(3006, 3007)

[0202] bb3=distance(4006, 4007)

[0203] a=pi*(bb0+bb1+bb2+bb3)/4#

[0204] mm(n)=a

[0205] Debug.Print mnam(n); mnum(n); mm(n)

[0206] ′Measurement 4002—Jacket

[0207] n=5

[0208] mnum(n)=4002

[0209] mnam(n)=“Jacket”

[0210] bb0=distance(1008, 1009)

[0211] bb1=distance(2008, 2009)

[0212] bb2=distance(3008, 3009)

[0213] bb3=distance(4008, 4009)

[0214] a=pi*(bb0+bb1+bb2+bb3)/4#

[0215] mm(n)=a

[0216] Debug.Print mnam(n); mnum(n); mm(n)

[0217] ′Measurement 4003—Seat

[0218] n=6

[0219] mnum(n)=4003

[0220] mnam(n)=“Seat”

[0221] bb0=distance(1010, 1011)

[0222] bb1=distance(2010, 2011)

[0223] bb2=distance(3010, 3011)

[0224] bb3=distance(4010, 4011)

[0225] a=pi*(bb0+bb1+bb2+bb3)/4#

[0226] mm(n)=a

[0227] Debug.Print mnam(n); mnum(n); mm(n)

[0228] ′Measurement 7001—Sleeve Length

[0229] n=7

[0230] mnum(n)=7001

[0231] mnam(n)=“Sleeve Length”

[0232] bb0=distance(1012, 1013)

[0233] d=xx(Feature(2003))

[0234] e=distance(2004, 2005)/2#

[0235] f=xx(Feature(2004))

[0236] arc=d−e−f

[0237] bb0=Sqr(arc*arc+bb0* bb0)

[0238] bb1=distance(1013, 1036)

[0239] sum=bb0+bb1−2#

[0240] mm(n)=sum

[0241] Debug.Print mnam(n); mnum(n); mm(n)

[0242] ′Measurement 5001—Pant Inseam

[0243] n=8

[0244] mnum(n)=5001

[0245] mnam(n)=“Pant Inseam”

[0246] j=Feature(1015)

[0247] a=yy(j)*yy(j)

[0248] sum=Sqr(a+36#)+1#

[0249] mm(n)=sum

[0250] Debug.Print mnam(n); mnum(n); mm(n)

[0251] ′Measurement 7002—Sleeve Inseam

[0252] n=9

[0253] mnum(n)=7002

[0254] mnam(n)=“Sleeve Inseam”

[0255] bb0=distance(1005, 1036)

[0256] sum=bb0−2.5

[0257] mm(n)=sum

[0258] Debug.Print mnam(n); mnum(n); mm(n)

[0259] End Sub

[0260] Public Function distance(ByVal f1, f2 As Integer) As Double

[0261] ′returns distance between two features

[0262] Dim i, j, k As Integer

[0263] Dim a, b, c, x1, x2, y1, y2 As Double

[0264] i=Feature(f1)

[0265] x1=xx(i)

[0266] y1=yy(i)

[0267] i=Feature(f2)

[0268] x2=xx(i)

[0269] y2=yy(i)

[0270] a=x1−x2

[0271] b=y1−y2

[0272] c=a*a+b*b

[0273] distance=Sqr(c)

[0274] End Function

[0275] The data produced by the system is compatible with the National Institute of Standards and Technology (NIST) Standard for the Exchange of product Model Data (STEP) and the NIST body dimensions for apparel

[0276] This technology can be used when shopping online or in-store at any participating retailer. Shoppers can register to use this technology at a participating online merchant or physical store. Once their body measurements are on file, they can begin the shopping process. Retailers may offer to download the software and instructions for shoppers who wish to create their files at home.

[0277] On-line stored body measurement data is password protected. A customer may allow another person access to their file by providing the correct username and password. Individuals may want to use this feature when gift buying or to allow an affiliated merchant to create a custom fit, one-of-a-kind garment.

[0278] Keeping accurate measurements on file is important. Therefore, customers should update their files when they gain or lose weight. All major manufacturers have computerized pattern storage systems, which will interface with the 9 body measurements online shoppers bring to their site.

[0279] This technology is better than hand measuring because it measures customers more consistently, more accurately, less intrusively and faster than traditional hand measuring. Also, the input and transmission of measurement data is standardized and automated, thus eliminating errors resulting from recording measurement data by hand.

[0280] The camera body measurement system has been described with reference to a particular embodiment. Other modifications and enhancements can be made without departing from the spirit and scope of the claims that follow.