[0001] This invention relates to a method of measuring an article such as an O-ring, for example with a view to selecting and supplying a replacement article. Background to the Invention

[0002] Online systems are known for identifying a product from an image up loaded from a smart phone or the like to enable the purchase of that product, for example as described in US2017/132486A1. It has also been proposed that online systems can use an image to assess the size of the article, for example by comparison with a reference article photographed alongside the article, for example as described in US2015/248719A1. Clothing sizes may be assessed and selected in this way.

[0003] However, for certain articles, such as sealing O-rings, obtaining the exact replacement can be critical; an incorrect size could result in failure of a seal and thus failure of the machine of which the article is a part. There is therefore a need to be able to determine accurately the correct replacement ar ticle from a range of such articles, for example from an on-line shop.

[0004] Conventionally, O-ring sizes have been assessed using a conical gauge, but because O-rings are formed from an elastic material there is a risk that the rings could become stretched in placing them on the gauge, thus lead ing to an incorrect assessment of the replacement ring. In addition, the gauge needs to be made with some precision to achieve the necessary accuracy, which means that it can be too costly to supply every O-ring user with one. Ac cordingly, there is a need for a simpler and more reliable way of measuring arti- cles.

Summary of the Invention

[0005] Accordingly, the invention provides a method of measuring an article, as set out in the Claims. [0006] The method of the invention permits, for example, the user to deter mine and purchase online the correct replacement for an article, by capturing a digital image, for example using a smartphone camera, of the article to be re placed located on to a calibrated reference background previously supplied, and then uploading the captured image to the online shop website.

Brief Description of the Drawings

[0007] In the drawings, which illustrate a method according to one exemplary embodiment of the invention:

Figure 1 is an image of an O-ring on a grid background;

Figures 2 and 3 are flowcharts illustrating successive stages in the meth od of the invention;

Figure 4 is an enlarged image of a section of the grid background illus trating the location of candidate points in identifying the primary grid;

Figure 5 is a further enlarged image of a section of the grid background illustrating the location of candidate points on the secondary grid;

Figure 6 is an enlarged view of a section of the image of Figure 1 , illus trating the final pass step 34 in Figure 3; and

Figure 7 is a flowchart illustrating the steps in using the data yielded by the method in the selection and purchase of a replacement O-ring. Detailed Description of the Illustrated Embodiment

[0008] In a typical procedure for purchase of a replacement O-ring online, after loading an app on a mobile device such as a smartphone, or after logging in to the purchase website, the user will place the O-ring 10 to be replaced on a printed grid 1 previously supplied by the O-ring supplier (Figure 1 ). The back- ground 1 1 comprises a primary rectilinear grid 12 and a secondary pseudoran dom grid 13 of triangles located within the primary grid. The secondary grid is configured so as to provide a unique angle between any two lines at a junction and thus: • Enables the determination of the x,y orientation of the background during image analysis - a square rectilinear grid cannot provide this information;

• Facilitates shadow detection, because detection of the secondary grid in shadowed regions of the image confirms that it is back ground;

• Provides positional information even when the grid is mostly oc cluded. The pseudo-randomness prevents aliasing to give a ro bust position for each point in the outline of the object; and

• The denser grid provides more positional information in the image while still involving a low ink cost in printing the background. The grid can still be measured manually..

[0009] The printed background may be provided as such by the supplier so that the dimensions can be guaranteed. Alternatively, the background may be provided as a downloadable document to be printed by the user. In the latter case, calibration will be needed. This may be achieved by the use of calipers, or even an accurate ruler, to verify the primary grid spacings. Alternatively au to-calibration might be achievable by including in the image of the article against the background a reference article of known accurate dimension, for example a coin. The grid may include lines of different colour to assist in identifying the lo cation of the article against the background.

[0010] Using, for example, a smart phone camera, the user will photograph the O-ring and background, producing a view such as in Figure 1. It will be seen that typical conditions may not produce an ideal image, for example shad ows may be present as a result of the lighting conditions.

[0011] As illustrated at step 20 of Figure 2, the user will upload the digital image so produced to the website or process it using the app in the mobile de vice. The following processing steps will then be carried out by the website server or the app:

Using a ridge detection algorithm 21 , candidate points that lie on the linear grid lines are discovered. A pre-trained neural-network based classifier 22 is used to remove spurious points that do not lie on the primary (linear) grid lines. A random sample consensus algorithm is used to identify candidate grid lines on the basis that they are lines passing through the most candidate points. Figure 4 illustrates the candidate points on the grid (circled for clarity).

• A robust fitting technique 23 is used to fit a polynomial function to the identified points. This allows for curvature from, for example, lens distortion or the paper not being entirely planar. Any outliers (lines that are not approximately orthogonal or overly curved) are discarded. This yields a list of polynomials that closely approxi mate the actual major grid lines in the image.

• A line detection algorithm (see Von Gioi, R.G., Jakuowicz, J., Mo rel, J.M. and Randall, G., 2010, LSD: A fast line segment detector with a false detection control. IEEE transactions on pattern analy sis and machine intelligence, 32(4), pp. 722-732) is used to identi fy the minor grid lines, as highlighted in Figure 5. The position & orientation of each minor grid line on the grid is known in advance. From the detected lines (in image-space), the intersection points & the angles defined by the lines at each intersection point are ex tracted. Using the a-priori information about the expected angles at each intersection point, a random sample consensus algorithm is used to assign each ideal (model) intersection point (in image- space) a position in grid space, according to the grid measure ments already defined by the user.

• A best fit world space transfer function is found (24). This produc es a mapping that relates the x, y, z position of a pixel in the im age to a real X, Y, Z position in millimetres across the plane of the grid paper. The field of view of the lens, necessary for this calcula tion, is obtained from the smart phone using a software API.

• A circular Plough transform is applied to the image at 30 (Figure 3) to identify the approximate location of the O-ring in the image. If no circle is detected, the image is rejected. • Starting from the estimated centre of the O-ring, projections are made radially. Along these radial“spokes” candidate points that are likely to correspond to the true edge projected by the O-ring over the grid are selected (31 ). For each of these points, a de scription vector (descriptor) is extracted consisting of values of neighbourhood pixels. A pre-trained machine learning model (32) is used to classify these descriptors as either true O-ring edge points or spurious points (e.g. falsely detected gridlines; edges from shadows). Spurious points are discarded. If too many points have been discarded the images rejected. This results in vectors of estimated radius and width at each polar angle.

• A Savitzky-Golay filter is applied to the radii and width vectors to smooth out any noise (35), resulting in cleaned vectors of estimat ed radius and width at each polar angle, i.e. cleaned of noise.

• If the computed centre with the filtered vectors is substantially dif ferent from the original basis (33), the centre position is updated and the process returns to step 31.

• Using the filtered vectors as“hints”, a final pass 34 is made using spokes analysis again, but using interpolation to provide sub-pixel accuracy at the transition of the O-ring. This gives high precision cleaned vectors of estimated radius and width at each polar angle. Figure 6 illustrates the plotting of the mean edge line by reference to the changes in pixel intensity

• A Savitzky-Golay filter 35 is applied to the radii and width vectors to smooth out any noise

• Actual diameter and width are calculated (37) using the mapping from the grid, correcting for the 3D projection of the O-ring onto the grid. A prior assumption that the O-ring has a circular cross section is necessary for this correction.

• Figure 7: the user is asked if the O-ring is known to be from a cer tain standard, for example ISO 3601 , (70) (this question may be asked at the start of the input procedure). If it is of a known stand- ard, data from a list of known standards 71 are used in the next step;

• the calculated O-ring diameter and width is then used to identify a list of the most likely O-rings by performing a t-test (Student’s t- test) for the hypothesised diameter/width against each of the cata logue O-rings (73) with reference to the database/sales catalogue of O-rings held at 72;

• the likely O-ring(s) from the catalogue is(are) then proposed for purchase to the user, the transaction being carried out in conven tional online shop manner (74).

[0012] While the invention has been particularly described with reference to identifying and selecting the correct size of O-ring, it will be appreciated that the method of the invention may be applied to other generally planar articles where accurate size selection is important, for example gaskets. Since the measure ment is carried out by photographing against a planar grid, it is preferably appli cable to articles having only a minor thickness, as it is the plan of the article that is measured. While some corrections may be applied to account for the per spective error caused in the projection of a 3D object on to the grid, these cor rections become less reliable as the thickness increases.

[0013] While the invention conveniently uses an ordinary camera such as is associated with a smart phone, it will be appreciated that other forms of camera may be advantageous, for example a thermal imaging camera, the article being simply warmed by the user’s hand and imaged against a cooler background, thereby eliminating the effects of shadowing.

[0014] Additionally, it may be possible for the user to indicate to the system whether wear is present in the sample O-ring, allowing the prediction of a suita ble replacement to be modified accordingly. This could suitably be incorporated into the process before the imaging step. The user would be asked whether there is any significant wear on the indie or outside surface of the O-ring (not the top or bottom - i.e. effectively the faces of the ring). If the user answers that there is, the system would warn the user that both the diameter and width measurements could be inaccurate. The user may be offered the opportunity to over-ride the width measurement manually, and will be advised to take care when multiple suggested O-rings are close in dimensions.

[0015] Further, while the described method uses a rectilinear grid as the primary background, other patterns may be usable, for example circular or ellip- tical. The secondary grid may be a pseudo-random line pattern as illustrated in the Figures, but other forms can be used, such as a plurality of circles or sinu ous or squiggly lines.