FR2729484A1 - Reading of identification marks from liquefied gas bottles for domestic use - Google Patents

Abstract

The reading of identification marks commences with an image of the marks, applies digital processing to the image, then applies the results to a neural net for final identification. Three chromatic views are taken of the identification marks, and these are digitised. The chromatic view with the best contrast is selected. The selected image is processed to black and white, the boundaries of the characters identified, and the characters digitally re-oriented so they are upright and sequential. The centre of gravity of each character is determined and character data relative to the centre of gravity delivered to a neural net, which identifies the character by comparison with a library of characters.

Description

METHOD AND DEVICE FOR READING ELEMENTS OF IFETIFICATIOI CARRIED BY GAS BOTTLES
LIQUEFIED FUEL FOR DOMESTIC USE, AND THEIR USE.

 The present invention relates to a method and a device for reading identification elements carried by cylinders of liquefied combustible gas for household use (butane, propane, technical mixtures or liquefied petroleum gas, which will be designated below by the generic name GPL). It also relates to their use in a chain for conveying and filling gas cylinders.

 We know that the G.P.L. is stored in gas cylinders for domestic use at room temperature, in the liquid phase, under the pressure of a gaseous sky. Indeed, at atmospheric pressure it would vaporize instantly, while, under pressure it remains liquid, which, at identical volume, can store about 250 times more energy. Burners are generally designed to use gas and it is imperative, because of this energy ratio, that they do not receive G.P.L. liquid.

 The G.P.L. in the liquid phase is characterized by a relatively high coefficient of volume expansion, it is therefore imperative, to avoid a risk of rupture in hydraulic thrust of the liquid phase, to maintain above the latter a gaseous sky capable of absorbing the dilations liable to occur due to temperature variations.

LPG storage tanks, including cylinders, are classified in pressure vessels and are therefore subject to various constraints, in particular a periodic hydraulic test (5 or 15 years) and an individual identification specifying
- their full weight, says Tare,
- the year of expiry of the test,
- the year of manufacture,
- the name of the manufacturer,
- the serial number,
- the capacity, the nature of the liquefied gas and the pressure.

 The information relating to the test due date and the tare is carried by a tablet set on the valve wheel and name Tare tablet. The weight of the full bottle is entered in kilogrania, with mention of the hectograms; the test due year is identified by the combination of a number and the color of the patch.

 These two pieces of information are used during the filling operations, because, on the filling lines, on the one hand, it is necessary to carry out a preliminary sorting to orient towards the test the bottles which have reached the test deadline and , on the other hand, the tare weight is essential for correct filling and weighing control. These two pieces of information are updated with each test, by changing the tablet, however, the other information is permanent and is written in the metal of the bottle.

 It is the charge of gas that is sold to the customer, the cylinder being made available to him for a deposit.

 The gas charge must be in accordance with the quantity to which the transaction relates (the most common bottles in France are intended for 13 kg of gas). It happens that bottles return to the filling incompletely emptied and, for certain uses, a systematic periodic exchange is carried out, whatever the use made of the bottles and therefore the quantity of gas which they are likely to still contain; to respect the gaseous sky, they must naturally leave with the correct weight of G.P.L. and especially not overfilled.

 The French fleet of bottles by G.P.L. represents approximately 50,000,000 units. This park has a relatively long lifespan: there are many circulating bottles produced during the 1950-1960 decade and renewal takes place very slowly.

 This fleet of bottles is also characterized by an important diversity of characteristics (number of weld beads, sheet thickness, height of the bottle, diameter of the flange ...), which is due as much to historical causes as to technical causes (intended use for butane, propane, fuel mixture, for example). I1 results, in particular, significant deviations in tare if we consider the case of the most common bottles, provided for a gas load of 13kg, we find some with "full" weights distributed between 21 and more 29 kg. There are also significant differences in height, up to 10 cm, in particular the height of the tap head, which carries the tare tablet, and therefore of its position relative to the conveyor during filling.

 It should also be mentioned that, for reasons of economic optimization between the logistics and transport costs, and those of multiplication of filling establishments and also the regulatory and safety constraints which limit the possibilities of establishment, it is necessary to for operators to use the exchange of filling services or subcontracting, so that competing establishments are therefore called upon to fill the bottles of their colleagues. It may be added that the bottles are not assigned to a filling establishment and that they occur successively relatively randomly at filling, both geographically and temporally: a bottle from a high-consumption customer can very quickly return to filling and, at the same time following, having been given to a customer having a backup or troubleshooting use, stay very long before returning, possibly an establishment different.

 The tare pellets are made of anodized aluminum, according to an interprofessional specification. The "tare" is entered in the form of a three-digit number: that of tens of kilograms, of size less than the following two, that of kilograms, followed by a comma, and that of hectograms. The test due date is indicated according to an interprofessional convention by the combination of the color of the patch and a number positioned across, below the "tare". The specification clearly defines the characters to be used: font, size, weight, orientation of the number relative to the test due date etc., which should be strictly identical on all the pellets. In fact, this is not the case. and one can note a considerable disparity between pellets.

In addition, the tare pellets undergo alterations due
- with the effect of time and the use of the bottles, since, the bottles being normally tested every fifteen years, the markings of the pellets must be able to be used during all this period
- to the stains present on the pellets, which can hardly be cleaned at economically acceptable conditions in industrial operation;
- crimping of the pellets, which amputates frequ4mmpnt part of the marking.

 The filling of the bottles includes a succession of industrial operations, which take place in a highly automated, very specific establishment, taking into account the characteristics of the G.P.L.

 The materials and equipment must be usable in an explosive atmosphere zone.

The nominal rate of filling equipment of the
Applicant is, for example, 1,200 bottles / hour.

On a filling line, given that it is not possible to easily follow the individual path of the bottles (the cycle time is not strictly constant at each station and each bottle, and some bottles are ejected when we detect disorders, such as absences or deterioration of certain accessories, or failure to close or open the tap, etc.), the pastilles must be read three times
- at the sorting station, to evacuate the bottles at the end of the test,
- at the filling station, to adjust the filling scale,
- at the weighing control station.

 The very high reliability required from reading the tare tablet, under these conditions of cadence, makes this work very painful: at the filling station, this operation is normally carried out by a two-hour shift between the different operators present.

 To overcome this drawback, one could naturally envisage carrying out a new marking of the tare pellets or other identification elements, for example using a bar code, suitable for reading by using computer and video equipment. Such an operation would however prove to be extremely costly, given the size of the existing fleet of gas cylinders. In addition, a very long transitional period, in which the new marking system and the existing system would be used jointly, would be necessary. Furthermore, if the new marking system were to be affixed to the pellets comprising the usual marking system, the available area would be extremely limited.

Finally, such recognition systems would themselves be sensitive to soiling, alterations and degradations and to the problems of orientation and positioning of the pellets already encountered with the usual systems.

 The present invention aims to avoid these problems, by proposing a method and a device for reading the tare pellets of G.P.L. bottles. of the existing fleet, which can be immediately used with the marking system already present on these tablets.

 The invention also aims to provide a method and a device for reading the tare pellets of G.P.L. bottles, which does not require human intervention.

 Finally, the invention aims to propose a method and a device for reading the tare pellets of LPG bottles, which is compatible with the very short delays, generally less than three seconds, which are available for this purpose on bottling lines. between successive bottles.

To this end, the invention firstly relates to a method for reading an identification element carried by a bottle of liquefied combustible gas, in particular a tare tablet, this method being characterized in that:
(1) three chromatic shots of the identification element are carried out almost simultaneously, leading to three chromatic images, the first red, the second green and the third blue, of this element
(2) we digitize each of these images
(3) the chromatic image having the best contrast is selected
(4) on the selected image, the actual position of the identification element is sought, from its dimensions, and the fraction of this image corresponding exactly to the image of this element is determined
(5) a binarization of the image of the identification element is carried out, so as to make the characters it contains appear in black on a white background or vice versa and to eliminate the nuances
(6) the image thus binarized is converted into a file of all the pixels of this image or of a polygon circumscribed to this image
(7) the background color of the image is determined using the pixels of the three images, respectively red, green and blue
(8) the binarized image is straightened to bring it to a predetermined position
(9) in this position, we locate and isolate the characters appearing on this image
(10) we search for the center of gravity of each of the characters thus identified and we process them by a filter around this center of gravity
(11) the characters thus treated are transmitted to a neural network, which identifies them by comparison with characters contained in a library.

The invention also relates to the use of such a reading process in a continuous conveying and filling chain of liquefied gas, characterized in that the position of this chain to which each bottle is transferred is determined according to of the color and characters identified by the reading process defined above. More precisely, it will thus be possible to process each bottle, depending on the results of reading the identification element or the tare tablet with which it is equipped.
- at a sorting station, which disposes of the bottles at the end of the test
- at a filling station, where the filling scale can be adjusted to the desired weight
- at a weighing control station, to compare the target weight with the measured weight and draw the necessary conclusions.

 Phase (1) of the process can be implemented by filming in sequence the three chromatic planes, red, green and blue, which constitute the image of the taring disc considered and this image is then transmitted (phase 2) to the video card from a computer equipped with appropriate software, which digitizes these plans.

 In practice, each of the bottles which pass on the conveyor of the filling line is stopped and gripped by a mechanical or pneumatic device, which lifts it up to bring it into contact with a centering stop of the tap head, under the shooting system, then immobilizes it there and triggers the shots, after which the bottle is lowered onto the chain and evacuated.

 The correct positioning of the tap head can be detected by a light beam optical system, which makes it possible to start shooting with an ultra-short and repetitive response time.

 Step (1) is very important, because the better the quality of the images on which the computer software will work, the better the performance.

The lighting (spectral composition, vertical orientation, inclined, grazing, concentrated or diffuse, or combination of different lighting ...) plays a key role in obtaining "good" images, that is to say images presenting a maximum contrast between the characters and the background, while being free of reflections and minimizing the contrast between the patch and the crimping imprints which appear hollow on its periphery. It is also necessary, of course, that the focusing of the objective is correctly carried out on the patch and that the latter does not move during the capture of the image.

 Alternatively, instead of immobilizing each bottle under the shooting system, it is also possible to film the image of the pastille to be read at once. This leads to working with a camera with three sensors and with strobe lighting. In fact, in 100 micro-seconds (maximum duration of a flash), a pellet of 12 millimeters in diameter moves 0.03 millimeter on a conveyor driven at 18 meters / minute, that is to say from 0 , 25% of its diameter, which does not produce any significant blurring of the image (even with very good definition, there would be less than 300 pixels in the image for the 12 millimeters of a dot).

 Step (3) is important. In fact, by choosing the monochrome plane which presents the best contrast for character recognition, one compensates for the effect of the color and the state of the pastilles on their readability.

 We will then operate on the luminance of this plane, which we can assimilate to more or less dark variants of gray.

 The plan selected in step (3) includes the image of the patch, but this image is not rigorously centered, because, during data entry, there is no adequate benchmark.

 In step (4), the software will search in this plane for the actual position of the patch, from its theoretical dimensions, taking into account the focal length of the objective used and the shooting distance. This will allow treatment to continue on the fraction of the plane corresponding exactly to the image of the patch being identified (or in practice to a square circumscribed to this theoretical circle), with correct centering of the patch.

 An original search method has been developed by the Applicant: it consists of passing over a plurality of virtual circles over the entire surface of the image, with diameters slightly greater than that of the pellet and retaining only that of which the standard deviation of the luminances of the peripheral pixels is the lowest.

 In step (5), a binarization threshold is sought and the image of the patch is binarized, to erase the shades of gray and form an image comprising only "whites" and blacks, this which "normalizes" the marking, the characters appearing in black on a white background, while, on the pastilles, depending on the color of the background, the characters are in white or in black.

This "normalized" image, which can only be used for further processing, is reduced to a file of all the pixels of the square circumscribed by the patch found in the previous step, each assigned a "0" or a "1 "(step 6).

 According to an advantageous embodiment of step (5), to take into account the extremely variable renderings of the used pellets and of the different colors on the images obtained, instead of working with the same thresholds for all the images, one realizes a histogram of the densities of "gray" of the plane considered and the analysis of this histogram makes it possible to determine the optimal threshold for the binarization of this image as well as the color of the characters (black, or white). Binarization is then carried out using these elements (threshold and color of the characters).

 In step (7), the color of the tablet is determined, taking into account its state of soiling and the chromatic nuances for the same original color or due to aging (navy blue to light blue, for example) .

 The search for color is carried out on the pixels determined as belonging to the background of the patch (those found "white" by binarization): for each of these pixels, we go from red, green, blue to HLS (Hue Luminance and Saturation ) and, for each of its components, a histogram is produced. Analysis of each of these three histograms (one per HLS component) makes it possible to determine the color of the patch. This information will be used, in combination with the reading of the number written across under the indication of Mtareff, to determine the date of expiry of the test.

 As the orientation of the tare tablet under the video camera is random, the image is straightened.

 For this purpose, we first seek to register in a virtual block each elementary drawing formed by groups of black pixels in the binarized plane. The software then determines for each of these blocks its main axis of inertia: if there are three axes parallel to each other and a fourth perpendicular, a first angle is determined. The software also searches for lines passing through points of the base of the blocks, which can confirm this angle or propose others: all the angles found are classified in order of preference according to criteria of likelihood and the software then performs successively all the corresponding rotations in the order of this classification, to select the one that seems to be the right one and rotate the image by this angle.

 To locate and isolate the characters carried by the tare label, the software then searches and locates on the rectified plan each of the expected characters (those which constitute the indication of tare and that necessary for determining the due date for test). For this, it performs the search for the location of the white spaces which separate the characters by testing possible separation lines and retaining those which cut the least number of "black" pixels (step 9).

 The software then prepares the characters it has located in the rectified plane to send them to the neural network (step 10). It searches for the center of gravity of each of them and passes a filter around this center of gravity, in particular to limit the effect of the variants of line thickness.

 The neural network functions by relying on a wise apprentice carried out on characters whose value it has been given. This training determines coefficients called weights. During recognition, we apply coefficients to the character to be recognized.

 The neural network has ten possible outputs corresponding respectively to the ten digits of O to 9 that a character can represent: for each character which is proposed to it, it will restore on each of these outputs a note between -1 and +1, which 'it will establish by comparison with the library constituted by learning (step 11). The analysis of the set of ten notes obtained on these ten exits for the character examined will make it possible to determine to which figure corresponds this character (the one which corresponds to the gratified exit of the best note) and also if this set of note converges enough on this conclusion for the result to be considered almost certain (if the convergence ratio of the notes on these ten outputs is insufficient, the software considers that it could not recognize the character considered).

 It is one of the advantages of the neural method chosen and an originality of this development to forge this judgment criterion of the result of the recognition of characters, which is essential in this application, since we require a quasi-certainty of accuracy. reading.

 The whole of this process is therefore simple and quick to implement, meeting the requirements of gas cylinder filling lines.

 The apparatus used for the implementation of this process constitutes another object of the present invention.

The invention therefore also relates to a device for reading an identification element carried by a bottle of liquefied combustible gas, in particular a so-called tare pellet, characterized in that it comprises
a camera capable of taking almost three chromatic images of the identification element simultaneously;
means for selecting the chromatic image presenting the best contrast
means for determining the fraction of this chromatic image which corresponds to the image of the identification element;
- Means for making appear in black on a white background, or vice versa, the characters that comprise the identification element; ;
- means for determining the background color of the identification element
- Means for bringing the image of the identification element to a predetermined position;
- means to locate and isolate the characters appearing on this image
- means to identify the characters carried by this image.

 The means for selecting the chromatic image of the identification element having the best contrast and for subsequently processing this image will simply comprise a computer with a video acquisition card and appropriate software.

 The camera will advantageously be equipped with a lighting system capable of providing good image quality and maximum contrast. an appropriate lighting system will be described below.

 As indicated above, means may be provided to immobilize and center the gas cylinder under the camera.

The accompanying drawings illustrate certain features of the invention. On these drawings
Figure 1 shows various tare pellets illustrating the diversity of characters encountered
Figure 2 illustrates the acquisition of three chromatic images of a patch and the choice of the best plan;
Figure 3 illustrates the binarization of the pastille image
FIG. 4 illustrates the rectification of the image of the patch;
Figure 5 illustrates the isolation of the characters from this patch
FIG. 6 illustrates the preparation of the results sent to the neural network for a pellet;
Figure 7 illustrates the results returned by the neural network for this same pellet
Figures 8 and 9 are views, respectively in elevation and in perspective, of a set of shots and lighting usable in the context of the invention.

 FIG. 1 is simply intended to illustrate the diversity of the characters appearing on the tare pellets and the variety of their relative positions. As indicated above, these pellets are soiled in use, worn over time and partially amputated by crimping. They are also presented in absolutely arbitrary orientations on the tap heads of the bottles, on the filling lines, which makes their reading all the more complicated.

 The present invention aims precisely to provide a simple and rapid method for reading such pellets.

 Firstly, we will proceed with the acquisition of three similar chromatic images of the pastille, simultaneously or almost simultaneously, using a camera. These images, referenced R for red, V for green, and B for blue, appear in FIG. 2. They make it possible to determine the color of the patch, from the weight of the three components red, green and blue.

 As seen in Figure 2, the three images of the patch are of different quality. For the rest of the operations, we will keep only one image, the one with the best contrast, here the green chromatic image. This image is intended to be processed in black and white.

 On the saved image, which is a square of 200 by 200 pixels, we will start by searching for the patch, by finding a circle with a diameter close to that theoretically fixed (156 pixels).

 Once we have found the circle corresponding to the dot, we will operate on a square circumscribed to this circle, with a dimension of 156 by 156 pixels and we will look for the binarization threshold corresponding to the best contrast of the characters compared to the background . For this purpose, the image is considered to have various gray levels, for example 256 in total (level 0 corresponding to black and level 255 to white) and the histogram of the different gray levels of the square is calculated. circumscribed to the circle of the patch.

 Once the binarization threshold has been determined, the image of the patch is binarized, the characters of which then appear in black on a white background, as shown in FIG. 3.

 It is then necessary to straighten this image, because the patch has appeared in an absolutely arbitrary position below the camera.

 For this purpose, we use the technique of enclosing Paves ", that is to say that we enclose the different characters in" blocks "(rectangles or squares with sides all parallel to the same directions), as shown on FIG. 4, and we are looking for different lines, namely the line serving as the baseline, which passes through the bottom of the three characters giving the tare of the gas cylinder and those corresponding to the moments of inertia of each of the blocks .

 We then consider the angles of the moments of inertia with respect to the baseline and we correlate the moments of inertia, among which three, corresponding to the tare figures, will be roughly parallel, while the fourth (figure vintage) will be perpendicular to the previous ones.

 We then straighten the patch, trying all the relationships according to the angles found.

 Once the image is straightened at a certain angle, we try to isolate three characters, by determining the white spaces separating the characters.

 For this purpose, horizontal and vertical lines are passed through the straightened image and those which meet the fewest black pixels are kept (see FIG. 5).

 After having succeeded in isolating at least two characters in this way, corresponding to the units of kilos and grams, it remains to prepare them before sending them to the neural network which will identify them. For this purpose, we seek the center of gravity of the character and we pass a filter around this center of gravity.

 We then send the characters to the neural network.

This network works by relying on learning carried out on characters whose value it has been given.

This training determines coefficients called weights.

During recognition, these coefficients are applied to the character to be recognized.

 The neural network has ten possible outputs corresponding respectively to the ten digits from 0 to 9 that a character can represent. For each character which is proposed to him, he will restore, on each of these outputs, a note between -1 and +1, which he will establish by comparison with the library formed by learning (see Figure 7). The overall analysis of the ten scores obtained on these ten outputs by the character examined will make it possible to determine which figure corresponds to this character and whether this set of notes converges sufficiently on this conclusion so that the result can be considered almost certain.

 When taking color images of the tare capsule, the lighting system used is of great importance. It is important to obtain an optimal contrast, not to suffer from any reflection, whatever the color of the pastille considered, and to minimize the effect of the crimp marks.

 To this end, the Applicant has developed a set of shots with incorporated lighting, which is shown in FIGS. 8 and 9.

 This assembly comprises a parallelepipedic housing forming a reflector, integral with a camera 2, external to the housing 1 and arranged coaxially with a light-shielding tube 3, which passes through the housing and ends at the base thereof by a pyramid 4 without bottom, in translucent plastic, open to the outside. Two explosion-proof luminaires 5, of the fluorescent tube type, are housed in parallel in the housing 1, on either side of the tube 3, above the pyramid 4. The pyramid made of translucent plastic material makes it possible to obtain lighting very diffuse and without shade. One could naturally replace this pyramid by a cone or similar.

 Of course, other lighting systems could be used without departing from the scope of the invention.

Claims (11)

Claims  1. Method for reading an identification element carried by a bottle of liquefied combustible gas, in particular a tare tablet, this method being characterized in that  (1) three chromatic shots of the identification element are carried out almost simultaneously, leading to three chromatic images, the first red, the second green and the third blue, of this element  (2) we digitize each of these images  (3) the chromatic image having the best contrast is selected  (4) on the selected image, the actual position of the identification element is searched for, from its dimensions, and the fraction of this image corresponding exactly to the image of this element is determined.  (5) the image of the identification element is binarized, so as to make the characters which it contains appear in black on a white background, or vice versa;  (6) the image thus binarized is converted into a file of all the pixels of this image or of a polygon circumscribed to this image  (7) the background color is determined from the pixels of the three images, respectively red, green and blue  (8) the binarized image is straightened to bring it to a predetermined position  (9) in this position, we locate and isolate the characters appearing on this image  (10) we search for the center of gravity of each of the characters thus identified and we process them by a filter around this center of gravity  (11) the characters thus treated are transmitted to a neural network, which identifies them by comparison with characters contained in a library.  2. Method according to claim 1, characterized in that the chromatic views are carried out after immobilization of the gas cylinder relative to the camera.  3. Method according to claim 1, characterized in that the chromatic views are carried out simultaneously, with stroboscopic lighting, while the gas bottle scrolls opposite the camera.  4. Method according to one of claims 1 to 3, characterized in that the step (4) of finding the actual position of the position of the identification element on the chromatic image comprises passing over the entire image surface of a plurality of virtual circles of diameter slightly greater than that of the patch, in order to select the one whose standard deviation of the luminances of the peripheral pixels is the smallest.  5. Method according to one of claims 1 to 4, characterized in that, in step (5) of binarization of the image of the identification element, a histogram of the densities of the different gray variants of the image considered, in order to determine the optimal contrast threshold for binarization of the image.  6. Method according to one of claims 1 to 5, characterized in that the step (8) of straightening the image of the identification element comprises a phase of writing in a rectangle or "block" of each of the characters of this image, the search for each of these blocks of its main axis of inertia, the search for the line or lines passing through the base of the "blocks", the search by software of the angles of these different lines and the pivoting of the image from these different angles.  7. Method according to one of claims 1 to 6, characterized in that the step (9) of locating the characters appearing on the image comprises a phase consisting in searching for the white spaces separating the characters by passing from straight lines to across the image and retaining those that meet the least number of black pixels.  8. Method according to one of claims 1 to 7, characterized in that the neural network comprises ten distinct outputs, corresponding respectively to the ten digits 0 to 9 that can represent a character, a note between -1 and +1 being restored on each of these outputs for each of the characters offered, after comparison with the library.  9. Device for reading an identification element carried by a bottle of liquefied combustible gas, in particular a tare tablet, characterized in that it comprises  a camera capable of taking almost three chromatic images of the identification element simultaneously;  means for selecting the chromatic image presenting the best contrast  means for determining the fraction of this chromatic image which corresponds to the image of the identification element;  - Means for making appear in black on a white background the characters which comprise the identification element;  - means for determining the background color of the identification element  - means for bringing the image of the identification element;  - means for locating and isolating the characters appearing on this image;  - means to identify the characters that bear this image.  10. Device according to claim 9, characterized in that it further comprises a lighting system cooperating with the camera, this lighting system comprising, inside a housing (1) with reflecting walls, at least one light source (5) disposed outside a bottomless pyramid or conical structure (4), made of a translucent material, open towards the outside of the housing (1).  11. Use of a method according to one of claims 1 to 8 or of a device according to one of claims 9 and 10 in a continuous chain for conveying and filling bottles of liquefied combustible gas, characterized in that the Bottles at the end of the test are ejected from the filling line and oriented towards a test line.