CH709015A2 - Colour filter for digital sensor and method for adjusting a photo-finish camera with such a filter. - Google Patents

Abstract

The subject of the invention is an adjustment method for a photo-finish camera (3) equipped with a matrix color filter comprising the following steps: a first step (E1) for choosing a basic pattern having optical properties; of sensitivity (S), color quality (Q) and resolution (N) predefined according to stroke parameters; A second software selection step (E2) of a set of adjacent columns whose number corresponds to the width of said chosen base pattern; A third step (E3) of centering said set of adjacent columns on the finish line (2). The invention also relates to a color filter for a digital sensor and a corresponding photo-finish camera.

Description

Technical area

The present invention relates to an adaptive color filter comprising color pixel columns and white pixel columns for matrix digital sensor, as well as a parameterization method for such a filter in the context of use by a camera. photo finish.

State of the art

[0002] In the context of digital photography, photosensitive sensors formed of a mosaic of pixels have long been known; the sensors most used for digital cameras use, for example, CCD technology (charge-coupled device) or CMOS (complementary metal oxide semiconductor). In order to produce color images, color sensors are applied to these digital sensors, which are also in the form of a matrix of different colors, each of the colors being dedicated to a pixel of the sensor to which it is superimposed. The color that must be associated with each pixel on the resulting image is then obtained through image processing algorithms.

[0003] One of the best known and most used color filters is the Bayer filter, which is the most classic form of RGB filter (R for red, ie red, G for green, it is that is, green, and finally B for blue, that is, blue, these three colors being the basic colors allowing to obtain any other by combination of additive synthesis). Such a filter uses a basic pattern covering 4 pixels, which is repeated over the entire surface of the sensor, with two green pixels in opposite corners complemented by a blue pixel and a red pixel.

A disadvantage of this type of color filter is the considerable reduction in the sensitivity of the associated digital sensor due to significant attenuation of the amount of light that reaches him. Indeed, a good part of the wavelengths being absorbed by the different filters. For these reasons, different variants of the Bayer filter have since been proposed, in particular by replacing one of the two green pixels with a white pixel in order to improve the sensitivity of the sensor. In this case, the modified basic pattern is also repeated on the entire filter.

In the field of sports competitions, there are also known auxiliary timing devices based on image recognition, better known as photo-finish. Such systems allow an operator to decide between competitors crossing a finish line by viewing images taken successively by a high definition camera centered very precisely at the finish line. The analysis of the sequence of images, taken at successive given instants and consequently corresponding to different timed times, makes it possible to determine delayed, but nevertheless very quickly after the arrival, more precise times of the crossing of the line of each competitor, for example by the thousandth of a second and classify competitors reliably.

Cameras used by photo-finish type devices often include Linescan type CCD sensors (better known by the acronym LS-CCD) whose image capture rate is even higher than that of CCD cameras. customary, allowing a temporal resolution up to ten thousandths of a second. The first sensors used for such cameras have presented a very particular bar-shaped matrix structure, that is to say a single column of pixels aligned very precisely on the finish line. Nowadays we tend to use two-dimensional standard matrix sensors, so the selection of the column aligned with the finish line can be done by software.

For such photo-finish cameras equipped with LS-CCD sensors, filters of the aforementioned RGB type can of course be used to obtain color images. Nevertheless, these filters are not suitable to provide adequate quality in any type of race conditions. Indeed, depending on for example the time of the course of the competition or weather conditions, the ambient brightness can vary greatly, and thus greatly alter the quality of the images obtained. The same goes for race type considerations, which condition the speed of competitors according to the events and consequently the shooting parameters, including the exposure time.

[0008] There is therefore a need for photo-finish cameras provided with color filters for digital sensors free of these known limitations.

Brief summary of the invention.

An object of the present invention is to provide a new type of color filter that has adaptive optical properties, as well as a new method of adjustment particularly practical of a photo-finish camera using such a filter.

These goals are achieved by a digital sensor color filter formed of a two-dimensional array of pixels, each pixel corresponding to a given color, the pixel array being formed of an alternation of first columns of color pixels and second columns of white pixels, characterized in that each first column of color pixels comprises a basic sequence of pixels repeated over the entire first column of color pixels, the basic sequence of pixels comprising at least three pixels of three different colors.

These goals are also achieved by a setting method for a photofinish camera comprising such a color filter, characterized in that it comprises the following steps:

[0012] Particular embodiments of the invention are defined in the dependent claims.

An advantage of the present invention is to continuously optimize the optical properties of color photos obtained through the proposed filter, regardless of the conditions of use.

Another advantage of the proposed solution is to allow a quick and efficient implementation of the adjustment of optical parameters chosen for shooting, and in particular, according to a preferred embodiment in which all adjustments are made. by software adjustment, without requiring any movement of the photofinish camera itself according to any of its degrees of freedom. Thus the fine adjustments of the alignment of the photofinish camera with respect to the finish line can be made once and for all, the filter settings being obtained only by software processing. The definitive locking of all degrees of freedom and the absence of physical manipulation of the photofinish camera thus greatly simplifies the adjustment operations while allowing to adapt to a maximum of types of possible race conditions.

Brief description of the figures

Examples of implementation of the advantageous invention are indicated in the description and illustrated by the appended figures in which: <tb> fig. 1 <SEP> respectively illustrates a top view of a photofinish camera used in the context of the invention, aligned with a finish line; <tb> fig. 2 <SEP> illustrates a scheme of a filter according to a preferred embodiment of the invention, and different basic patterns that can be retained; <tb> fig. FIG. 3 illustrates a diagram of the various steps of the method of setting a photofinish camera according to a preferred embodiment of the present invention.

Example (s) of Embodiments of the Invention

FIG. 1 shows a block diagram of a photo-finish camera used in the context of the present invention. The photo-finish camera 3, preferably LS-CDD, is aligned on a finish line 2 of a track 1 preferably comprising several lanes 10 in which the various competitors operate. Note that the example of an athletics track 1 is totally indicative and not limiting, photo-finish cameras 3 can also be used in cycling races, track racing, or for horse racing.

The photo-finish camera 3 preferably comprises 4 degrees of freedom, including three in rotation and one in translation. In fig. 1, only the degree of freedom in translation T and a first degree of freedom in rotation R1 about a vertical axis are illustrated; nevertheless the camera is preferably rotatably mounted around a ball 8 which simultaneously has second and third degrees of freedom in rotation R2, R3 additional to manage the alignment of the optical axis 4 of the camera 3 on the line of 2. The photo-finish camera 3 comprises a digital sensor 6 formed of a matrix of pixels, as well as an image processing device 7 supplying a digital signal 9 at the output corresponding to the image detected by the sensor. . In order to form color images, the photo-finish camera 3 furthermore comprises a particular color filter 60 whose optical properties can be adjusted according to various predefined parameters, discussed below. According to a preferred embodiment, this color filter 60 is affixed directly to the pixels of the digital sensor 6, so that by misnomer we speak of pixels also for the filter in order to denominate the different colors covering a pixel of the digital sensor 6.

Such a color filter 60 whose pixel patterns correspond to a preferred embodiment of the invention is illustrated in FIG. 2. The width 61 and the total height 62 of the filter are counted in rows and columns of pixels, which are generally between 1024 and 2048 pixels. It can be seen that the filter 60 is formed of an alternation of first columns 611 of color pixels and second columns 612 of white pixels W, each first column 611 being juxtaposed with a second column 612. Such a configuration of a color filter 60 per column is particularly suitable for a photo-finish camera that uses at most a few columns of pixels aligned on the finish line to allow the most precise time stamping possible by associating these columns at a given time, while maximizing the size of the image, whose height is precisely determined by the number of pixels in the column.

Each of the first pixel columns 611 comprises a base sequence A comprising at least 3 different colors, in order to guarantee a good color quality of the image obtained by additive synthesis based solely on it and not requiring it. no other pixel in another column. According to the preferred embodiment illustrated, each first column 611 of color pixels thus comprises at least one blue pixel B, one green pixel G and one red pixel R. Nevertheless, as an alternative, the colors used could also consist of cyan, magenta and yellow.

In FIG. 2, the base sequence A is identical for each first column 611, namely the succession, starting from the top to the bottom, of exactly three pixels: a first red R, a second green G, and a third blue B. This red-green-blue sequence is thus repeated on the entirety of each of the first color columns 611. The fact of using only three pixels per sequence makes it possible to improve the color quality Q, which is modeled in the context of this the invention is inversely proportional to the number of pixel lines necessary to obtain the color image, as explained further in detail in relation to FIG. 3. Alternatively, one could use a sequence corresponding to a modified Bayer type pattern where the base sequence repeated on each first column 611 would be the succession of four pixels instead of three, namely: a green G, a blue B , then another green G and finally a red R. This base sequence variant would be advantageous in terms of sensitivity, because the green color is not as absorbent as blue and red. Nevertheless, such an improvement would be to the detriment of the color quality, since it would be necessary to use 4 lines of pixels instead of three. Moreover, having identical sequences for each base sequence A of each column makes it possible to limit the number of columns necessary to form basic patterns, and thus maximize the resolution of the image.

As can be seen in FIG. 2, the arrangement of the basic sequences A is not exactly identical for each first column 611 of color pixels. Indeed we can notice that each line of pixels has the same sequence of a sequence of pixels formed of a red R, a green G, then a blue, as for each first column 611 of color pixels, with simply white pixels W interposed. Such symmetry properties are obtained by shifting each base sequence A by one pixel from a first column 61 of color pixels to the next. In other words, if for a given pixel line, the pixel of a first column 611 of color pixels is of a certain color, for example red, the pixel of the same color will end up for example a higher line or more bottom, on the next first column 611 of pixels and a line lower - or respectively higher - on the first column 611 of pixels of previous color. According to the preferred embodiment illustrated - see on the right of FIG. 2 - The base sequences are shifted one pixel up by scanning the color pixel columns one by one from left to right. Therefore only the first columns 611 of "modulo 3" color pixels are exactly the same, taking an arithmetic formulation, that is to say that only the first, fourth, seventh, tenth and so on. first color pixel 611 columns are identical, as are the second, fifth, eighth, and so on. and the third, sixth, ninth, and so on. This is the reason why dotted lines have been drawn to materialize a fictitious demarcation line of exact replication of the 6-column pattern which is effectively repeated on the whole of the filter 60. The advantageous optical properties of such a shift of the sequences base A have been verified empirically with respect to filters for which all lines are of identical color. Indeed, in addition to the obvious symmetry considerations of the rows relative to the columns on the entire color filter 60, such an arrangement facilitates the color determination for each pixel by algorithmic calculation.

With the aid of the color filter 60 illustrated in FIG. 2, it is possible to take pictures in black and white, when for example the brightness is so limited that it is impossible to consider taking a picture in color, using only a second column 612 of white pixels W centered on the finish line. However, the filter 60 makes it possible to define different patterns making it possible to adapt gradually to different frame rates, conditioning the frame rate and consequently the exposure time, and more exactly four basic patterns determining different sensitivities, extending over a number of columns between 1 and 3 and comprising respectively 0.1 or 2 second columns of white pixels W.

The four basic patterns M1, M2, M3 and M4 here contain a number of pixels which is a multiple of three, corresponding to the three color pixels respectively red R, blue B and green G and which are systematically used. Thus each of these patterns extends over a height L of three pixels 3, while the width, equal to the number of pixels, is variable; the setting of the width of the pattern influences the resolution and in parallel the sensitivity of the sensor. The detail of each of the four preferred motifs illustrated is given below:

The optical properties obtained using each basic pattern are conditioned, inter alia, by the following 3 parameters:

To evaluate overall the optical performance of each pattern, it will be possible to multiply the scores obtained for each of the sensitivity parameters S, Q of color quality, and N of resolution above; however, the choice of the pattern can be made taking into account a possible weighting, in particular on the sensitivity parameter S. The value of the parameter Q is preferably fixed, so we will preferably seek the best compromise between the sensitivity parameters S and N of resolution.

It can be seen that, of the four illustrated patterns, at least three basic patterns comprise a different number of white pixels W, namely 0.3 or 6 (corresponding respectively to the first basic pattern M1, the second and third patterns basic M2-M3, and finally to the fourth basic pattern M4). This allows the sensitivity to be adjusted to at least three distinct levels (high-medium-low). Moreover, the fact that the ratio between the white pixels and the color pixels can be respectively 0% (for the first basic pattern M1), 33% (for the third basic pattern M3), 50% (for the second basic pattern M2), and 66% (for the fourth basic pattern M4) makes it possible to further refine the granularity and the panel of the possible adjustment choices for this sensitivity parameter S, thus making it possible to adapt to a many types of races and weather conditions, typically including clear weather, cloudy weather, cloudy weather and night conditions. Finally, the fact that each of the basic patterns M1-M4 extends over at most 3 columns makes it possible to maintain a correct resolution N for the image obtained.

At the bottom of FIG. 2 are referenced various sets of columns that can be chosen according to the basic pattern selected in relation to the race conditions, that is to say in particular the brightness and the type of race conditioning among other things the speed of scrolling competitors at the level from the finish line. We thus distinguish:

It can therefore be found on the one hand that the determination of the number of columns N1-N4 corresponding to each basic pattern M1-M4 does not automatically make it possible to automatically select the right set of columns C1-C4 which must be aligned. on the finish line 2 to provide the desired optical properties. For example, the third and fourth basic units M3 and M4 extend over an identical number of 3 columns (i.e., have N3 = N4 = 3) but have different optical properties. In this case, shifting software manipulation may be necessary, as explained below with the aid of FIG. 3 which illustrates a sequence of adjustment steps of a photofinish camera 3 according to a preferred embodiment of the invention, and more particularly focused on the basic patterns of the filter illustrated in FIG. 2.

In the diagram of FIG. 3, the first step E1 consists in choosing a basic pattern having optical properties of sensitivity S, color quality Q and resolution N predefined as a function of race parameters such as the weather conditions and the type of race (the speed of scrolling the number of lines of pixels per second that can vary from 1000 lines per second for athletics to more than 10000 lines per second for car races, the speed for cycling and horse races being of the order of 3000 lines per second). According to the preferred embodiment using the color filter of FIG. 2, the color quality parameter is set because all basic patterns have a height L of 3 pixels.

In the second step E2, a number of columns N corresponding to the selected pattern is determined, and software is selected correspondingly a number of adjacent columns on the matrix. For the first basic pattern M1, a first number N1 of a single column will therefore be selected, for the second basic pattern M2 a second number N2 of two adjacent columns will be selected, and for the third and fourth patterns M3 and M4 will be selected. 3 adjacent columns (the third and fourth numbers N3 and N4 being equal, each to 3).

Then, a third centering step E3 of the set of adjacent columns obtained on the finish line 2 is performed. According to a preferred embodiment, the only software manipulation will be preferred, so that no physical movement of the relatively fastidious photofinish camera 3 is ever required once it has been installed. Alternatively, the camera can be moved physically using inverse kinematics algorithms to optimize manipulation and minimize the range of motion at different degrees of freedom among the three axes of rotation R1.R2, R3 and degrees of freedom in translation T1. In order to facilitate this centering step, whether it is performed by software alone or not, it will be possible to use the aid of an auxiliary reticle, as in the solution proposed in patent EP0898249. When the number of columns is odd, the center column is aligned on the finish line 2; when the number of columns is even, the operation is slightly more delicate because it is in theory to place the reticle between two adjacent columns. When an auxiliary reticle is used, it can be considered in practice that the third centering step E3 is complete when the reticle is aligned with one of the columns of the selected set of columns.

However, it may be noted that when the number of columns of the chosen pattern is odd, the centering of a set of adjacent columns equal to the pixel width of the pattern does not necessarily guarantee that the right set of columns, here the thirds and fourth set of columns C3 and C4 for the patterns M3 and M4, respectively, have been chosen. Indeed, by choosing any column of pixels of the filter matrix, the probability of selecting a first column 611 of color pixels is identical to that of selecting a second column 612 of white pixels W, and likewise by choosing a set of three adjacent columns of pixels, it is equiprobable that the set contains a single column of white pixels W on the three columns or two columns of white pixels. In this case, a fourth step E4, subsidiary and used only for an odd number N of columns (that is to say equal to 2P + 1, P corresponding to a mathematical modeling of the divider of the even lower number), consists shifting the adjacent column set aligned to the finish line by a single pixel, i.e. the equivalent of a column, to the left or to the right. It is indeed easy to see that if a column of white pixels W (ie a second column 612) is aligned with the finish line, it will suffice to shift the reticle on the column immediately to its left or to its right to be centered on a first column 611 of color pixels where the first base pattern M1 is found in the third centering step E3 for a first set of columns C1 which comprises only one. Thus similarly, shifting a set of three adjacent columns by one pixel to the right or to the left switches alternately from the third set of columns C3, corresponding to the third pattern M3 with two columns of color pixels, to the fourth set of columns C4, corresponding to the fourth pattern M4 which contains only one for two columns of white pixels W. When the number N of columns is even by cons, that is to say equal to 2P, the fourth step is never necessary.

Therefore, using the adjustment method described in FIG. 3 for a color filter 60 presenting the arrangement of first columns 611 of color pixels, with basic patterns R, G, B, shifted, interlaced with second columns 612 of white pixels W, we see that the manipulations required for the adjustment optical properties of the filter depending on the race conditions are extremely simple. Once the camera has been installed and all its degrees of freedom have been permanently locked, one can choose a first pattern determining optical properties corresponding preferably to normal operating conditions, or statistically most likely. This pattern extending over 1 to 3 columns, it will then be sufficient to select one or two additional columns maximum, or respectively to remove or deselect one or two at most, and then shift, if necessary, all columns obtained from a single pixel in width if the number of columns is odd. To further facilitate the centering operations, when they are performed using an auxiliary reticle as in the solution proposed in the patent EP0898249 of the applicant, we can choose to systematically select, by default, the column of pixels on which is aligned the reticle and complement it if necessary by additional columns of pixels in the second step E2 preliminary software selection for a set of columns whose number is strictly greater than one. The remaining manipulations to carry out the centering are then extremely summary and thus allow to have a maximum reactivity, which is often essential for demonstrations taking place outside, thus very exposed to variations of meteorological conditions, and in real time like for example track and field events at World Championships or Olympic Games.

This convenience of use is also worth not only for the initial settings of the photo-finish camera, but also to dynamically adjust the optical properties of the color filter 60 by changing the selection of the basic pattern. The fact of having at most 2 columns to add - or respectively remove, depending on the conditions - and having to make an offset of at most one column of pixels in width is all the more appreciable when the duration of Events are spread over several periods of the day, for example track and field events for which the qualifiers are held during the day and the finals often late in the evening or during the night. For example, if, for example, the first basic pattern M1 is selected by default for day-to-day use conditions, it will suffice to add a second adjacent column 612 of white pixels W to obtain the second base pattern M2, of increased sensitivity, and still another second column 612 of white pixels W in the evening, so that the two second columns 612 of white pixels W are located respectively on either side of the first column 611 of color pixels aligned on the line of arrival 2, to obtain the fourth basic pattern M4, whose sensitivity is even better, when the night will have fallen. And to adjust the color filter to an intermediate sensitivity, if necessary, we can go from the first base pattern M1 to the third base pattern M3, by adding two adjacent columns, respectively a second column 612 of white pixels W and a first column 611 of color pixels on the same side, then shifting the entire pixel. All these simple software selection operations of adding and / or possibly deselecting columns of pixels can thus make it possible to follow a test throughout its course by adjusting as best as possible and as and when the optical properties of the optical filter installed on the photo-finish camera 3.

However, it will be understood that the basic patterns M1-M4 and the basic sequence A have been given by way of example only to illustrate a particularly preferred embodiment. However, other basic patterns extending over a larger number of columns, or comprising a base sequence A extending over a larger number of pixels, or comprising other colors is also conceivable without departing from the scope. of the invention, insofar as the centering and shifting software manipulations remain relatively simple. For example, it is possible, for example, to envisage alternate basic sequences in R, G, B, and cyan, magenta, yellow on two successive columns.

Furthermore, without departing from the scope of the present invention, the method of setting a photo-finish camera described above may also choose to use only any subset of patterns among the all four patterns M1 to M4 above, depending on the conditions of use, which require for example to use only two or three of the four reasons.

Claims (13)

A color filter (60) for a digital sensor (6), said color filter (60) being formed of a two-dimensional array of pixels, each pixel corresponding to a given color, said pixel array being formed of an alternation of first columns (611) of color pixels and second columns (612) of white pixels (W), characterized in that each first column (611) of color pixels comprises a base sequence (A) of pixels repeated on the set of said first column (611) of color pixels, said base sequence (A) of pixels comprising at least three pixels of three different colors. 2. Color filter (60) according to claim 1, characterized in that said base sequence (A) is identical for each first column (611). 3. Color filter (60) according to claim 2, characterized in that said basic sequence (A) comprises at least one blue pixel (B), a green pixel (G) and a red pixel (R). 4. Color filter (60) according to one of the preceding claims, characterized in that said base sequence (A) comprises exactly 3 pixels. 5. Color filter (60) according to the preceding claim, characterized in that each said basic sequence (A) of a said first column (611) of color pixels is shifted in height by one pixel relative to each of said base sequences (A) of said first columns (611) of adjacent color pixels. 6. Photo-finish camera (3) comprising a color filter (60) according to one of the preceding claims and a digital sensor (6) matrix, said photo-finish camera (3) being rotatably mounted around a ball joint (8) having three degrees of freedom in rotation (R1, R2, R3), and also has a degree of freedom in translation (T) in the direction of the course of the race. 7. A method of adjustment for a photo-finish camera (3) according to claim 6, characterized in that it comprises the following steps: A first step (E1) of choice of a basic pattern having optical properties of sensitivity (S), color quality (Q) and resolution (N) predefined as a function of stroke parameters; A second software selection step (E2) of a set of adjacent columns whose number corresponds to the width of said chosen base pattern; A third step (E3) of centering said set of adjacent columns on the finish line (2). 8. Adjustment method according to claim 7 for a photo-finish camera (3), said first step (E1) for selecting at least two basic patterns of sensitivities (S) different. 9. A method of adjustment according to claim 7 or 8 for a photo-finish camera (3), said first step (E1) allowing to choose at least four basic patterns of different sensitivities (S), extending over a number of columns between 1 and 3. 10. A method of adjusting a photo-finish camera (3) according to one of claims 7 to 9, said third step (E3) of centering being followed by a fourth step (E4) secondary fine tuning consisting of a offset in width of said set of adjacent columns by an integer number of columns of pixels, said fourth step (E4) being used only when said number of columns corresponding to the width of said base pattern determined at said second step (E2) is odd and strictly greater than 1. 11. A method of setting a photo-finish camera (3) according to claim 10, said fourth step (E4) requiring the offset in width of said set of columns of pixels of only one column of pixels, in n ' any sense. 12. A method of setting a photo-finish camera (3) according to one of claims 7 to 11, said third step (E3) of centering, and, if appropriate, said fourth step (E4) offset, does not consisting only of software adjustment steps. 13. A method of setting a photo-finish camera (3) according to one of claims 8 to 12, the transition from a set of predefined columns, corresponding to a predefined basic pattern, to another set of predefined columns. requiring only the addition or deletion of up to two columns of pixels, and the width offset of only one column of pixels.