DE102020126930A1 - Method for determining the point in time for cleaning a cooking chamber of a cooking appliance - Google Patents

Abstract

The present invention relates to a method for determining the time for cleaning a cooking chamber (6) of a cooking appliance (30), in which camera images (32) of a camera (8) directed into the cooking chamber (6) are fed to evaluation electronics (40). which uses the camera images (32) to determine the degree of soiling of the cooking chamber (6) using software-supported image analysis. In order to determine the right time for cleaning the cooking chamber (6) with improved accuracy, it is proposed that several camera images (32 ) the brightness values (60) of several or all pixels of the camera images (32) via an averaging (70) using a metric and subsequent limit value formation (80) or via a limit value formation (80) and subsequent averaging (70) using a metric, it is decided whether cleaning is required.

Description

The present invention relates to a method for determining the time for cleaning a cooking chamber of a cooking appliance, in which camera images from a camera directed into the cooking chamber are fed to evaluation electronics, which uses the camera images to determine the degree of soiling of the cooking chamber using software-supported image analysis.

From Scripture DE 101 28 024 A1 it is known to determine the beginning and/or the duration of a cleaning cycle for thermal cleaning of a cooking chamber, for example by means of pyrolysis as an example of a cleaning cycle with a thermal cleaning process, based on the evaluation of the optical light emitted by a light source into the cooking chamber and detected by an optical sensor to control radiation. The optical sensor measures how much the light intensity in the cooking chamber has decreased compared to a reference value. The optical sensor is arranged in the cooking chamber wall opposite the light source. This method is disadvantageous because the light intensity detected by the optical sensor is only measured in the small area of the arrangement of the optical sensor and the measurement result is at least also falsified by extraneous light that enters the cooking chamber through the appliance door. Shading effects from trays or racks in the cooking space are not recognized and are not taken into account. The respective measurement is limited to the window size to the optical sensor and is inaccurate. It is not possible to discern the nature of the pollution. The reference value must also have been saved beforehand, although it remains unclear what should actually serve as a reference value.

From Scripture DE 10 2016 206 483 A1 it is known in a commercial cooking appliance to determine when cleaning is necessary by comparing a current image of a cooking surface recorded by a camera with a reference image of a dirty cooking surface or a cooking product that requires cleaning. The point of this procedure is to determine the right time to carry out the cleaning. The images captured by the camera are recorded using electronic image sensors such as CCD or CMOS sensors. The recorded images can be further processed digitally. It is necessary, however, to provide suitable reference images for the respective cooking appliance in question. Since the cooking chamber is often presented differently in a currently recorded camera image than in the reference image, depending on user behavior, especially in the case of cooking appliances for the consumer sector, for example due to grates, baking trays, baking paper, pots or leftovers in the cooking chamber, this method for measuring fails Determining the right time for a cleaning is very common. The specific procedure for image evaluation is not described. It leads to unreliable results.

In Scripture DE 10 2014 116 709 A1 describes a method for determining the degree of soiling of a cooking appliance, in which a sensor surface is illuminated at an angle by a light source and the light reflected by the sensor surface is received by a camera in order to compare the light emitted by the light source with that received by the camera close light to a degree of contamination. Since both the light source and the camera are outside the cooking chamber, but the dirt accumulates inside the cooking chamber, this method is also imprecise and unreliable.

From Scripture KR 2016 006 9344 A it is known to use a camera to take a photo of the interior of a cooking appliance in order to then use the photo to determine when the cooking chamber needs to be cleaned by evaluating the image. Here, too, there is a lack of a concrete description of how the image evaluation should take place.

The general problem is that a cooking appliance is cleaned too infrequently and cleaning is postponed even though it could help the function of the cooking appliance and the result of the cooking process. Fat spatters adhering to the interior surfaces of a cooking appliance make up a significant part of the soiling of a cooking appliance.

It is disadvantageous if a user of the cooking appliance could postpone the required cleaning for so long without specific information that soot particles originating from the dirt are deposited on the food to be cooked during a cooking process in an excessively dirty cooking appliance. A light source that is intended to illuminate the cooking chamber can be darkened by soiling, so that it no longer illuminates the cooking chamber sufficiently to be able to observe the food well during a cooking process. The image quality of camera images can deteriorate over the long term if they are not cleaned. This is particularly unfortunate when the camera images are projected onto a display of the cooking appliance or a smart device that is independent of the cooking appliance and the food to be cooked can no longer be seen optimally on the camera images. As a result, the cooking progress is recognized more poorly and may be incorrectly evaluated. The aesthetic pleasure of seeing the cooking progress can also be affected. Delayed cleaning of the cooking chamber can result in an increased workload, or pyrolytic cleaning can result in unpleasant odors as the dirt burns off. The reaction products from the combustion of pollution can also exceed permissible limit values for the room air without the user noticing.

Accordingly, it is the object of the present invention to create a method that enables improved accuracy when determining the right point in time for cleaning the cooking chamber of a cooking appliance.

The object is achieved for a generic method in that, from several camera images, the brightness values of several or all pixels of the camera images are decided whether cleaning is required by averaging using a metric and subsequent forming a limit value or using a limiting value and subsequent averaging using a metric.

It is generally difficult to distinguish whether a dark spot is caused by soiling or by the food being cooked. Stains are also harder to identify when the background, such as a baking sheet, is very dark. However, the food changes over several cooking processes. However, the soiling does not change its position, at most further stains appear. In order to determine whether a pixel shows a dirty spot, it can be checked, for example, whether the pixel falls below a certain brightness value. Recognition can be significantly improved by evaluating multiple images or an image combined from multiple images. Experience has shown that soiled regions tend to be darker and less variable over multiple cooking cycles than unpolluted regions. For a single image, it is not known whether a dark spot is caused by pollution or by a local feature of the scene. However, when viewing images from multiple events, the dirt is always in the same place while the scene changes. The dirty parts of the image change less than the immediate surroundings in a sequence of several camera images. The use of multiple images allows dirt to be detected without the user having to take any specific action.

This applies even if camera images are used for averaging and limit value formation, on which no food placed in the cooking chamber can be seen. Different lighting conditions in the individual camera images, which originate from light sources that can be located inside and outside the cooking chamber, different insertion positions of baking trays and racks and other objects placed in the cooking chamber, such as baking tins, can already result in a sufficient variance between the camera images for evaluation purposes result. Occasional images of an empty cooking chamber are particularly interesting because they show completely different brightness values and structures than camera images with food to be cooked in the cooking chamber. Particular differences can arise here with regard to the image values of the individual pixels of such camera images.

The use of one, several or only camera images showing the food to be cooked for the averaging and limit value formation can nevertheless be technically sensible because the food to be cooked in the cooking chamber during a particular cooking process statistically does not have exactly the same texture and color as in another cooking process. In particular, local gray value or color fluctuations, such as those caused by light dough placed in the cooking compartment, red tomato sauce and salami on a pizza, brown pastries, cakes and roasts, yellow cheese and green vegetables and herbs in casseroles and the like during the respective baking processes. balance each other out by averaging and adopt a gray or color value averaged from the individual camera images. Each camera image of the cooking space produced during a cooking process, regardless of whether it is a color or black-and-white image, thus differs more clearly from other camera images of the cooking space. This applies even if a user only ever cooks pizza in the cooking device, for example, because even then the topping and the position of the pizza in the cooking chamber differ from previous cooking processes. The items to be cooked shown on camera images result in a greater variance in the image values of the individual pixels or pixel fields from which the camera images are composed as digital data.

A variance in the image values can arise in particular if the food to be cooked is moved in the cooking chamber. Camera images can also be used that were taken from various successive cooking processes with different items to be cooked. Finally, different Hel Values of the individual pixels can also be obtained from a comparison of camera images from a single cooking process if the food to be cooked changes sufficiently during the cooking process, for example rises and/or browns.

The more camera images with different representations of the cooking chamber and/or the food being cooked are averaged, the better the evaluation result tends to be. However, even fewer camera images can suffice for an assessment if they offer sufficient variance in the brightness values of the individual pixels for assessment purposes. In order to get a qualitatively sufficiently reliable assessment, the cooking appliance must have been used, which requires a certain amount of time. When looking at the cooking chamber over a longer period of use, further dirt spots may appear, which then also do not change in terms of location and hardly change in terms of their shape and structure until the next cleaning is carried out.

In any case, the brightness values of pixels change when averaging to form an average image, regardless of whether the camera images used for the averaging show the food to be cooked or not, while dirt spots are at the same place and, with regard to their shape and brightness, are at least approximately the same during further cooking processes stay the same. This is also independent of the use of a color or a monochromatic camera.

Averaging to generate an average image by means of a metric is a mathematical process used to calculate image values of pixels with one another in order to obtain information about regions that do not change or no longer change and which can therefore be regarded as dirty regions . An average image can be formed from the multiple camera images by calculating, for example, an arithmetic mean value or a median value as a metric from the gray or color values for a pixel point or a pixel field that corresponds to one another in the camera images. The advantage of a median value can be seen in the fact that individual outlier values have less of an impact on the mean value. For example, if there are fat deposits on wall surfaces in the cooking chamber, these can be clearly seen on the average image as dark spots that contrast in their area and at the edges with the brightness or color values of the surrounding image components of the average image.

When this description refers to the "average image" on the basis of which evaluations are made, this does not always mean the exact average image in the narrow sense that results from the averaging of the gray or color values of the individual pixels from the camera images , but in a broader sense also those images and the gray or color values contained therein and other information that result from further processing of the average image in the narrower sense, such as a binarized image in which the pixels are only digitally labeled as "dirty ' or 'not soiled' are shown. If derivations formed from the average image are also meant, this is indicated in each case.

However, other metrics can also be used to form an average value. For example, the dirty pixels are not only generally darker than non-dirty pixels, but also have a lower standard deviation or variance. Therefore, instead of the arithmetic mean, the standard deviation or the lower variance can also be determined. Other methods can also be used as a metric for forming an average for combining multiple images, such as forming a geometric or harmonic mean, forming averages in different color spaces, determining weighted averages, determining spatial or temporal gradients of brightness and/or color, the use of high-pass, low-pass or other filters, without this list of examples being limited to the methods mentioned.

Depending on how high the resolution of the average image is, the evaluation electronics can calculate image values for each individual pixel or also for pixel fields in which the image values of several pixels are viewed together. By considering pixel fields, the complexity of the calculations can be reduced without a loss of quality in the assessment of the degree of soiling necessarily having to occur.

Information about the degree of soiling is extracted from the average image or a derivation formed from it by subjecting the average image to a software-supported image evaluation. The image evaluation can use the known statistical methods and/or suitable filters techniques using suitable filters for individual image values such as the brightness or color values of individual pixels and pixel fields, area distributions of certain brightness or color values, and the like. The image evaluation is no longer dependent on information that is outside the camera images or the average image or a derivation formed therefrom, such as reference images or other reference values.

Experiments have shown that it is possible to detect the degree of soiling of a cooking chamber regardless of whether an averaging is carried out first using a metric and only then is the limit value determined, or whether the two steps are carried out in reverse order. In both cases, the results differ gradually, but in both cases the degree of soiling of the cooking chamber is clearly identified. It is also possible to determine the degree of contamination by performing both calculation methods and then combining the results of both calculation methods to form an assessment of the degree of contamination.

Once the degree of contamination of the cooking chamber has been determined using the information extracted from the average image or a derivation formed therefrom, the evaluation electronics can easily determine a point in time for cleaning the cooking chamber that is appropriate for the degree of contamination. In particular, the evaluation electronics can output a signal that a cleaning process is recommended when the degree of soiling determined using the average image or a derivation formed therefrom has reached a threshold value for soiling that is no longer acceptable. The threshold value can be defined, for example, via a percentage of the area of the areas identified as dirty in the total area of the average image or a derivation formed therefrom and/or the area percentage of at least one single coherent area identified as dirty in the total area of the average image or a derivation formed therefrom. Larger pixel fields recognized as dirty can also be taken into account with a disproportionate weighting in the evaluation, because larger areas of dirt make cleaning the cooking chamber seem advisable.

In order to decide whether cleaning is necessary, it should be noted that it is usually a matter of giving the user a message that cleaning should be carried out immediately when the threshold value for carrying out cleaning has been reached. However, it is of course also possible to report to the user a prognosis value that, based on the soiling determined, cleaning should be carried out at the latest after ten additional cooking cycles, for example, or that cleaning should be carried out within the next five cooking cycles, for example. The problem with forecasts is that they cannot be sufficiently accurate because it is not known to what extent the cooking chamber could become soiled as a result of the cooking cycles that are still to come. When baking rolls, the risk of contamination is much lower than when roasting a duck. It therefore makes sense to promptly request cleaning when a degree of soiling is detected that makes a cleaning cycle necessary, for example by issuing a message "Cleaning necessary".

The evaluation electronics can evaluate the need for cleaning after each cooking process, but timing is also possible, for example after every third or fifth cooking process, or the evaluation is based on operating parameters such as the preselected temperature, a grill that is switched on, the duration of a cooking process or a combination of these options.

The software-supported image processing can be used to classify pixels and/or pixel fields from the camera images or the average image or a derivation formed therefrom as dirty or not dirty, and the evaluation electronics determine the degree of soiling from the proportions of the pixels and/or pixel fields classified as dirty and not dirty of the cooking chamber. For example, the evaluation electronics can assess the cooking chamber as in need of cleaning if the proportion of pixels and pixel fields classified as dirty in the total area of the average image or a derivation formed therefrom reaches or exceeds 5%. Of course, the proportion can also be higher or lower depending on the programming of the evaluation electronics. The image evaluation of the camera images or the average image or a derivation formed therefrom can take place independently of how users of the cooking appliance individually equip the cooking chamber, for example with grates, baking trays, baking paper or other vessels. The image evaluation of the average image or a derivation formed therefrom is therefore much more accurate in detecting the soiling, and the point in time at which a cleaning process is to be carried out can be determined much more precisely. That The cooking appliance can also remind the user that the cooking chamber needs to be cleaned at a point in time before the contamination has a negative effect on the food to be cooked, the camera image of the cooking chamber no longer shows the cooking food in a satisfactory quality, or unpleasant odors in the case of pyrolytic cleaning occur that are no longer classified as tolerable by the user or are no longer tolerable due to the pollution in the room air.

Depending on the technical equipment of the cooking appliance, cleaning can then be carried out manually by the user or thermally using a pyrolysis program.

The invention can be implemented easily and inexpensively in cooking appliances, since a suitable camera is already installed as standard in some cooking appliances by professional users, but also in those in the consumer sector. The manufacturing costs incurred when implementing the invention in a cooking appliance are then very low. The cameras are often camera systems that have software-supported image analysis. The image evaluation makes it possible to see whether food has been placed in the cooking compartment. Camera images can then be produced which show the cooking chamber with or without the food to be cooked placed in it. However, other sensor technologies are also possible, such as ultrasound, odor sensors or other technologies, or combinations of such technologies, which can be used to determine whether camera images showing an item to be cooked are being taken with the camera. The sensor signal output by the existing cooking product detection sensor system then only has to be evaluated by suitable software, which forms part of an evaluation electronics system. This requires little programming effort. A physical computing capacity on which the software can run is also often already provided in the operating electronics in a cooking appliance, so that existing equipment technology can also be used with regard to the computing capacity to implement the invention and no additional costs are incurred here either. The evaluation electronics can then consist solely of a software package that is loaded onto the control electronics of the cooking appliance.

In cooking appliances in which there is no camera, this can of course also be built into the cooking appliance only for the inventive application. It is possible both with already existing cooking product detection sensors and with devices to be installed separately in a cooking appliance, to design the evaluation electronics as a separate computing chip with software programmed on it, which is used to implement the invention in the operating electronics of a cooking appliance. In the case of operating electronics that have remote control functions, the software package can of course also be integrated into the remote control function, such as an app on a smartphone or a tablet computer.

According to one embodiment of the invention, an average image is first calculated during averaging using a metric and subsequent limit value formation, then an average brightness value is calculated for a majority or all pixels of the average image, then the average brightness values of the pixels are compared with a limit value that contains a binarizing evaluation threshold for the differentiation of the average brightness values between a pixel rated as dirty or not dirty, and then based on the number of pixels rated as dirty after the limit value comparison, a decision is made as to whether cleaning is required.

This method of calculation results in an average image in which variably bright pixels in the respective camera images are shown weakened in their gray tone value by averaging, while the pixels in which darker gray tone values are specified without a greater variance in all camera images contain the gray tone values for these Pixels remain in the dark area. When considering limit values, the average image can be simply binarized by showing, for example, all pixels with a gray tone value <20 as dirty and all pixels ≥20 as not dirty in a derivation of the average image. The binarized derivation can now be evaluated by determining how many pixels are considered dirty and whether this number exceeds a threshold value above which cleaning is considered necessary.

If a gray tone value is mentioned here, this information initially refers to gray tone images that have been recorded by a monochromatic camera. In the case of a gray tone image, each pixel of an image is assigned a value that represents its brightness or light intensity on a scale from 0 to 255, for example. 0 stands for a black pixel and 255 for a white pixel. If a color camera is used, a gray tone image can first be generated from the color image. Commercially available color cameras generate color information by placing a wavelength-selective sensor in front of the sensor chip translucent lattice, a so-called Bayer filter, which, depending on its position in the lattice, allows red, green or blue light to pass through. A so-called RGB image can then be reconstructed from the red, green and blue color components. Each pixel has three values, these represent the light intensity in the red, green and blue areas of the optical spectrum. A gray image can now be generated from the RGB image by reducing the color information to pure brightness information. In the simplest case, the gray value is determined for each pixel as the average of the red, green and blue values. It is thus possible to form gray tone values even when using color cameras. It is crucial for the further procedure that an image is available in which the light intensity and thus the brightness are represented by the values of the individual pixels.

According to one embodiment of the invention, during the limit value formation and subsequent averaging, an individual brightness value is first calculated using a metric for a majority or all of the pixels of each camera image, then the individual brightness values of the pixels are compared with a limit value, which is a binarizing evaluation threshold for distinguishing between the individual brightness values between a pixel of the camera image assessed as dirty or not dirty, a binary contamination image is created for each camera picture from the pixels assessed as dirty or not dirty, and then an average image is calculated from a majority or all pixels of the binary dirt images, in which the gray level value of each pixel corresponds to a contamination probability, and the gray level values of the pixels are calculated to form an average contamination probability value of the average image et, which is compared to a limit value and if the limit value is exceeded, cleaning is assessed as necessary.

When binary contamination images are created beforehand and an average image is subsequently formed, there is no binary contamination image, but rather a contrasting image in which contamination probabilities are represented via gray scale values of the respective pixels. Depending on the conditions of use, this can result in better assessment results than when using an exchanged sequence of averaging and limit value formation.

According to one embodiment of the invention, some or all of the several camera images are colored images in which the color channels of the individual pixels are evaluated differently or further calculated without the color values.

According to one embodiment of the invention, the camera images are produced by the camera through a transparent layer that separates the camera from the cooking chamber, with the transparent layer forming part of the inner wall of the cooking chamber. The transparent layer can consist of a pane of glass, for example. Soiling such as fat splashes that occur in the cooking chamber during the cooking process are deposited on the interior surfaces of the cooking chamber and thus also on the transparent layer. The soiling of the transparent layer is thus similar to the soiling of the inner surfaces in the rest of the cooking space and is therefore representative of the degree of soiling of the cooking space overall. Since the transparent layer is located between the cooking chamber and the camera, the dirt that collects on the transparent layer covers the camera's view of the food to be cooked that is behind the dirt in the cooking chamber from the camera's point of view. This means that the image values for soiling on the transparent layer in the area of the relevant pixels or pixel fields do not change or at least hardly change, even in an average image or a derivation formed from it, because the camera images no longer contain different types of food, whose image values could be averaged, but only the view of the pollution. The image values of the contamination on the transparent layer contrast well in the average image or a derivation formed therefrom with the image areas surrounding the contamination, which show averaged image values from the image values of the individual images of the item to be cooked, which differ from one another, through the items to be cooked recorded during different cooking processes.

According to one embodiment of the invention, the multiple camera images show multiple different cooking processes and/or loading situations. In order to be able to detect contamination on an average image or a derivation formed therefrom with the necessary accuracy, a certain number of camera images is required, via which a statistical smoothing of the color values with characteristically distributed image values for individual pixels or pixel fields is established. When several camera images are averaged, brightness and/or color values are obtained for the individual pixels or pixel fields from which soiled areas can be identified. The more camera images from different cooking processes are included in the averaging, the better the recognition. The changes that can be detected between different camera images of a single cooking process are often too small to be reliable draw conclusions about the degree of contamination of the cooking chamber. Even when considering fewer than five camera images, the averaging effects are often still too blurred. This only applies if the food in the cooking chamber changes sufficiently during the cooking process, for example by rising or browning, or is moved on a turntable in a microwave, for example. In such cases, just a few camera images, possibly even from just one cooking process, can be sufficient to determine whether the cooking chamber needs to be cleaned.

According to one embodiment of the invention, the evaluation electronics evaluate the quality of the average image or a derivation formed from it using an evaluation algorithm, and a point in time for cleaning the cooking chamber is only determined if the quality of the average image or a derivation formed therefrom has been assessed as sufficient. If the averaged image is averaged from camera images that have too little variance with regard to the image values for the individual pixels or pixel fields, soiling in the cooking chamber cannot be identified with sufficient certainty and clearly. A small variance can easily be detected by an evaluation algorithm by comparing individual pixel values from different images. There are various stochastic methods for this. The camera images can, for example, vary slightly if the same homogeneous-looking food is always being cooked and/or only a single small food is being cooked in the same place in the cooking chamber, the food is not being moved in the cooking chamber, or the camera images all show the same empty cooking chamber. In such a case, further camera images of the cooking chamber are required in order to obtain a variance in the image values for individual pixels or pixel fields that is necessary for detecting soiling. In such a case, the evaluation electronics should not give any recommendation as to when cleaning is required. The electronic evaluation system only makes a recommendation for the time of cleaning when the evaluation algorithm considers the quality of the average image or a derivation formed from it to be sufficient to serve as the basis for a recommendation.

According to one embodiment of the invention, the degree of contamination of pixels or pixel fields of the average image or a derivation formed therefrom is weighted differently in different sectors of the average image or a derivation formed therefrom. This measure is advantageous for compensating for imbalances in the assessment of the degree of soiling of the cooking chamber, which result, for example, from a focusing of the camera image and/or the perspective with which the camera is directed into the cooking chamber. For example, if the camera is aimed at an angle from above into the cooking chamber, without a correction, soiling that is found in the average image or a derivation formed from it in the front area from the camera view would be weighted more heavily than soiling that is in the front of the camera due to the perspective distortion located in a more distant area. This perspective effect can be corrected by weighting the detected contamination differently, depending on the sector, when determining the degree of contamination. When recording and averaging camera images through a transparent layer, however, such perspective distortions would not play a role if the transparent layer is arranged flat in front of the camera lens. Here, however, different weightings of pixels and pixel fields recognized as dirty can result due to optical distortions in the average image or a derivation from the lens optics formed therefrom, which represents different sectors of the camera image in different sizes. A sector-dependent correction of the weighting of the detected contamination is also possible here.

Another option is to weight the outer areas of a camera image, in which the side walls can be seen, for example, less heavily than the center of the cooking space, where the food is usually located. This measure can improve the averaging of the camera images. If camera images are transmitted and displayed, or if additional evaluation functions are carried out based on camera images, these are significantly less disturbed by dirt in the outer areas than by dirt in the center of the picture. In such cases, the visual assessment of the degree of soiling by a user looking into the cooking chamber from the outside can deviate from the assessment of the degree of soiling by the method according to the invention. For the display of the camera image or the additional evaluation functions, however, the evaluation using the method according to the invention is advantageous because the evaluation of the degree of soiling is based on the camera images that are displayed and/or additionally evaluated.

According to one embodiment of the invention, the classification of the pixels and/or pixel fields of the average image or a derivation formed therefrom takes place taking into account the probabilities of the respective classification. For example, a first pixel or a first pixel area can be dirty as "very likely", "probable" or "unlikely" and/or a second pixel or a second pixel area can not be dirty as "very likely", "likely" or "unlikely". be rated. More or fewer gradations in the classification of probability are possible. The differentiated classification of the individual pixels or pixel areas results in a more precise situational image of the degree of soiling of the cooking space in the cooking appliance. Such a soft classification with, for example, a total of six different classification levels could be superior to a digital yes/no classification because the certainty with which a pixel or a pixel field was assigned a classification is included in the assessment of the degree of contamination.

According to one embodiment of the invention, the average image is converted into a gray image. An average image can show color fluctuations that are not due to the photographed food or soiling, but to other effects, such as uneven lighting of the cooking chamber. The color fluctuations can be removed from the average image by converting the image values of the pixels or pixel fields into a gray image.

According to one embodiment of the invention, the gray image or the average image is converted into a differential image by filtering. In the gray image or averaged image, dirty spots in the image values of the pixels and pixel arrays may have a higher gray value than other spots on the averaged image, or the dirty spots may have a particular color value that other spots in the averaged image do not have. By using filters, the points on the gray image or the average image in a difference image can be made particularly recognizable, for example by using thresholding to hide other image values.

According to one embodiment of the invention, structures with a small size are removed from the average image, the gray image and/or the difference image in a correction run. Due to the random nature of the original camera images, small local variations that do not represent contamination cannot be avoided. However, structures with a small size can be ignored or removed via morphological operations. This takes place in a correction run to which the average image, the gray image and/or the difference image are subjected. In the correction run, the structures recognized by the image processing software are smoothed over a large area, particularly in the area of possible contamination. The image remaining after the correction run shows the spots where dirt is located in the average image with a high level of hit accuracy.

According to one embodiment of the invention, specific areas of the average image, the gray image or the differential image are ignored by the evaluation electronics. The areas of such false positives can be image areas in which contamination is indicated although there is actually no contamination there. Shadows from the lighting can be found in such areas, for example, or the areas are covered by components of the cooking appliance. If such image areas were included in the assessment of the degree of soiling, the result would be falsified. This is avoided by ignoring these areas. Depending on the device, these areas can be located at different points in the images. Accordingly, the areas of the false positives can be programmed device-specifically into the operating software, or the operating software has a self-learning program with which areas of the false positives are recognized in a self-learning manner and excluded from the evaluation.

According to one embodiment of the invention, after the cooking chamber has been cleaned, the evaluation electronics delete previously recorded camera images. Of course, after the cleaning has been carried out, pictures taken before the cleaning may no longer be used, since they show dirt that is no longer present. Therefore, the memory must be reset as soon as a cleaning has been performed. Before a new assessment can be made, a sufficient number of camera images must also be created in order to be able to carry out a new assessment. If a pyrolysis is carried out, the evaluation electronics know this and can automatically empty its memory. If the customer cleans manually, a memory for the current degree of soiling can be provided for this case, in which the last evaluated degree of soiling has been established. The status of the memory can be queried with each new evaluation. If this changes towards clean, previously recorded camera images must also be deleted because the current status apparently no longer corresponds to the previously determined degree of soiling.

• 1: Gargerät mit einer Kamera,
• 2: Verfahrensablauf mit Mittelwertbildung und anschließender Grenzwertbildung, und
• 3: Verfahrensablauf aus 2 mit getauschter Abfolge.

The invention will be explained in more detail below using exemplary embodiments. Show it:

• 1 : Cooking device with a camera,
• 2 : procedure with averaging and subsequent limit value calculation, and
• 3 : Procedure off 2 with reversed sequence.

An embodiment of the invention is in 1 shown purely schematically and will be described in more detail below. 1 shows a cooking appliance according to the invention with a camera.

In the 1 a cooking appliance 30 according to the invention is shown as an example. The cooking appliance 30 is embodied here as a baking oven and has a housing 2 in which a cooking chamber 6 delimited by a housing wall 4 is arranged. The baking oven also has a 2D camera 8 designed as a fixed-focus camera as an optical sensor, with an opening 4.1 for the camera 8 being formed in the housing wall 4 and covered by a lens 10 designed as a converging lens made of borosilicate glass.

In the cooking chamber 6, a cooking chamber lighting 11 designed as an oven lamp is attached to a cooking chamber ceiling 4.2 of the housing 4. The baking oven light 11 is designed here at the same time as a camera light 11 . Accordingly, the fixed focus camera 8 in the present embodiment can be constructed more simply. Alternatively, the cooking chamber lighting 11 can also be arranged on a side wall or consist of a combination of ceiling and side lighting.

The camera 8 is designed to take pictures of the interior of the cooking chamber 6 and is arranged accordingly on the housing 2 . Since the cooking chamber 6 is relatively wide and flat, a lens 8.1 of the camera 8 has a correspondingly large opening angle on the object side. The camera can also be designed as a plenoptic camera, TOF camera or stereo camera, ie using 3D technology. Cameras designed as thermal imaging cameras are also conceivable. In addition to single images, the recording of image sequences and moving images is also possible.

The opening 4.1 is closed essentially gas-tight by the converging lens 10 and a seal 12 arranged between the converging lens 10 and the cooking chamber ceiling 4.2 of the housing wall 4 and designed as a glass fiber seal. Alternatively, a metal gasket with or without a graphite coating can be used. Correspondingly, the vapors produced during a cooking process taking place in the cooking chamber 6 of a cooking item 16 placed on a cooking item support 14 cannot escape through the opening 4.1 in an undesired manner. The camera 8 can also be arranged in a lateral position or in any other position relative to the cooking chamber 6, so that it can also take useful photos of the cooking chamber 6 and the dirt present therein from these positions for evaluation purposes.

In addition, the converging lens 10 made of borosilicate glass and the glass fiber seal 12 protect the camera 8 from substances contained in the vapor and from the high temperatures arising during the cooking process. This is particularly important since the cooking appliance 30 is equipped as a self-cleaning cooking appliance 30, ie with a pyrolysis function. Far higher temperatures result in pyrolysis operation.

The cooking chamber 6 is thermally insulated from the environment in a manner known to those skilled in the art by a substantially circumferential insulation layer 18 . The insulating layer 18 is only partially shown in the figure.

The opening 4.1 in the cooking chamber wall 4 is arranged here in the center of the cooking chamber ceiling 4.2 of the housing wall 4. The opening 4.1 is kept as small as possible in order to have to break through the insulation layer 18 as little as possible at this point. The opening 4.1 is shown significantly enlarged in the figure for the sake of clarity.

The camera 8 is arranged further away from the cooking chamber ceiling 4.2 and thus from the cooking chamber 6 in a cooler area of the housing 2 due to the insulating layer 18, but also for the purpose of reducing the temperature load on the camera 8 by the thermal radiation.

The opening 4.1 can be made so small because a converging lens 10 is used here. The converging lens 10 widens the beam path of the light, symbolized by dashed lines 19 in the figure, through the opening 4.1. Despite the small opening 4.1, this enables the cooking chamber 6 to be shown completely in the essential areas on the images recorded with the camera 8, despite a reduced opening angle of the lens 8.1 of the camera 8 on the object side. In this way, the angle of view in the plane of the page shown in the figure can be increased from approximately 50° to approximately 110°. The aforementioned angle information cannot be inferred from the schematic representation. On camera images 32 marked with the im or camera 8 arranged adjacent to the cooking chamber 6, even minor soiling on the side walls of the cooking chamber can be clearly seen.

Since the opening 4.1 is relatively small, the opening 4.1 can be arranged in the region of a top-heating element 20 in the cooking chamber ceiling 4.2, without the opening 4.1 and the top-heating element 20 having a negative effect on one another. For example, the top heater 20 does not appear on the images captured by the camera 8. The same applies to the cooking chamber lighting 11; the cooking chamber lighting 11 is also arranged on the cooking chamber ceiling 4.2 of the cooking chamber wall 4 in such a way that there is no undesired interaction between the cooking chamber lighting 11 and the upper heating element 20.

Since the cooking appliance 30 is a cooking appliance 30 with a pyrolysis function, the camera 8 and the converging lens 10 are also arranged in a cooling channel 2.1 of the housing 2. The camera 8 and the converging lens 10 are in heat transfer connection with the cooling air guided in the cooling channel 2.1. The cooling channel 2.1 is designed here as an air duct 2.1 of the cooking appliance.

The air duct 2.1 is connected to the housing wall 4 in a flow-conducting manner by means of a flange 22, which is also resistant to chemicals and temperature, and the air duct 2.1 is connected to the environment in a substantially gas-tight manner.

The air duct 2.1 of the cooking appliance 30 is used to suck the vapors produced during cooking processes in the cooking chamber 6 out of the cooking chamber 6 and to discharge them into the open air. The vapor-air mixture is guided in the air duct 2.1 formed in the housing 2. The cooking appliance is also cooled in this way.

Due to the arrangement of the camera 8 and the converging lens 10 in the air duct 2.1, an additional active cooling device can be dispensed with here. Correspondingly, the cooking device 30 according to the present embodiment can be implemented with fewer components and thus more cost-effectively.

The individual components of the cooking appliance 30 according to the invention are partially shown at a distance from one another in the figure; see, for example, the distance between the insulation layer 18 and the cooking chamber ceiling 4.2 of the housing wall 4. This is only for a better overview. The components of the cooking appliance 30 according to the invention are connected to one another in a manner known to those skilled in the art, unless explicitly stated otherwise using the exemplary embodiment. The camera 8 can also be designed as a plenoptic camera, TOF camera or stereo camera, ie using 3D technology. Cameras designed as thermal imaging cameras are also conceivable. In addition to single images, the recording of image sequences and moving images is also possible.

A camera image 32 recorded by the camera 8 is transmitted to the evaluation electronics 40 in each case. For this purpose, the cooking appliance 30 is connected to the evaluation electronics 40 to which the camera images 32 taken by the camera 8 are transmitted. The evaluation electronics 40 have image processing software 46 with which the software-supported image processing is carried out. The individual camera images 32 produced by the camera 8, which are used for the image evaluation, can be stored in the evaluation electronics 40, but it is also possible to store them in the operating electronics 34 of the cooking appliance 30 or if the cooking appliance 30 has an Internet connection and is appropriately equipped to store in the cloud. The image processing software 46 is programmed in such a way that it calculates an average image 42 from a plurality of available camera images 32 as an image evaluation.

The evaluation electronics 40 extracts information about the degree of soiling of the cooking chamber 6 from the average image 42 . The operating electronics 34 can then communicate this point in time to an operator via the available displays, be it an indicator light on the cooking appliance 30, an indication on a display on the cooking appliance 30 or a display of a smart device that communicates with the cooking appliance 30 via WLAN or the Internet.

In the exemplary embodiment, the lens 10 represents a transparent layer through which the camera 8 produces camera images 32 of the cooking chamber 6 . Fat spatters that splash up from the food 16 to be cooked during a cooking process also adhere to the lens 10 , which forms part of the inner wall of the cooking chamber 6 on the side of the lens 10 that faces the cooking chamber 6 . Cover this dirt adhering to the lens 10 from the point of view of the camera at the points where they are located on the lens, the view from the camera of the food 16 to be cooked. In camera images 32 produced subsequently, these points of the camera images 32 have none or at most even with changing food 16 to be cooked in the cooking chamber 6 only minor changes in the image values of the pixels or pixel arrays in question. At the points of the lens 10 where there is no soiling, the food to be cooked 16 lying in the cooking chamber 6 during the respective cooking process can be clearly seen in the camera image 32, and the different textures and colors of the food to be cooked 16 in the cooking chamber 6 result different values for the image values for the relevant pixels or pixel fields in the camera images 32, which are then calculated into an averaged value by averaging in the average image 42.

The double arrow 48 indicates that the evaluation electronics 40 and the operating electronics 34 can communicate with one another beyond the point in time 44 by providing the information. The communication can relate in particular to cooking item identification, which can be a component of the operating electronics 34 or of the camera 8 . The cooking product recognition can be used to verify for the evaluation electronics 40 that the available camera images 32 show the cooking chamber 6 with the food to be cooked 16 placed therein and that the camera images 32 were produced on the occasion of different cooking processes.

The evaluation electronics 40 can further process the calculated average image 42 into a gray image 50 that can offer better evaluation options. The gray image 50 alone or in combination with the average image 42 can be further processed to form a difference image 52, in which certain values can have been filtered out of the image values for individual pixels or pixel fields in order to obtain an image that can be evaluated even better. The average image 42, the gray image 50 and/or the differential image 52 can be used by the evaluation electronics 40 to determine the degree of soiling of the cooking chamber 6.

In 2 a process sequence with averaging 70 and subsequent limit value formation 80 is shown, in order to then decide at decision point 90 whether there is a need for cleaning. Starting from a brightness value 60 of a pixel, a mean value for the brightness value of this pixel is first formed for this pixel from a plurality of camera images 32 using a metric. If this new brightness value determined from a number of camera images 32 is available, this brightness value can be entered in the average image 42 for this pixel. The average image is composed of a large number of brightness values calculated in this way for the individual pixels. In the limit value formation 80, the average image 42 is binarized in that a decision is made for each pixel examined in the context of the limit value formation 80 as to whether its brightness value 60 is above or below an assumed limit value. If the brightness value is above the limit value, the corresponding pixel is considered not to be dirty; if the brightness value is below, this pixel is considered to be dirty. A decision can then be made on the basis of the contamination found as to whether a cleaning cycle is required.

In 3 the procedure is over 2 , but in reversed sequence with a first limit value formation 80 and the subsequent mean value formation 70.

• Vor Beginn der erfindungsgemäßen Verschmutzungserkennung müssen von der Kamera 8 Kamerabilder 32 des Inneren des Garraums 6 aufgenommen werden. Dafür kann die Kamera 8 in einer Ausführungsform eine monochromatische Kamera 8 sein, die ein Grautonbild aufnimmt. Bei einem Grautonbild wird jedem Pixel eines Bildes ein Wert, der seine Helligkeit/Lichtintensität auf einer Skala von z.B. 0 bis 255 repräsentiert, zugewiesen. Dabei steht 0 für ein schwarzes, 255 für ein weißes Pixel. Wird eine Farbkamera 8 verwendet, so kann aus dem Farbbild zunächst ein Grautonbild erzeugt werden. Handelsübliche Farbkameras 8 erzeugen Farbinformationen, indem sich vor dem Sensorchip ein wellenlängenselektiv lichtdurchlässiges Gitter, ein sogenannter Bayer-Filter, befindet, das je nach Position im Gitter rotes, grünes oder blaues Licht durchlässt. Aus den roten, grünen und blauen Farbanteilen lässt sich dann ein sogenanntes RGB-Bild rekonstruieren. Dabei besitzt jedes Pixel drei Werte, diese repräsentieren die Lichtintensität im roten, grünen bzw. blauen Bereich des optischen Spektrums. Aus dem RGB-Bild lässt sich nun ein Graubild erzeugen, indem die Farbinformation auf eine reine Helligkeitsinformation reduziert wird. Im einfachsten Fall wird für jeden Pixel dazu der Grauwert als Mittelwert von Rot-, Grün-und Blauwert bestimmt.

$$Grau=\frac{1}{3}Rot+\frac{1}{3}Gr\ddot{u}n+\frac{1}{3}Blau$$

More specifically, the procedure is as follows:

• Before the contamination detection according to the invention begins, camera images 32 of the interior of the cooking chamber 6 must be recorded by the camera 8 . For this, in one embodiment, the camera 8 can be a monochromatic camera 8 that records a gray tone image. In a grayscale image, each pixel of an image is assigned a value that represents its brightness/light intensity on a scale from 0 to 255, for example. 0 stands for a black pixel and 255 for a white pixel. If a color camera 8 is used, a gray tone image can first be generated from the color image. Commercially available color cameras 8 generate color information by having a wavelength-selectively transparent grating in front of the sensor chip, a so-called Bayer filter, which allows red, green or blue light to pass through, depending on its position in the grating. A so-called RGB image can then be reconstructed from the red, green and blue color components. Each pixel has three values, these represent the light intensity in the red, green and blue areas of the optical spectrum. A gray image can now be generated from the RGB image by reducing the color information to pure brightness information. In the simplest case, the gray value is determined for each pixel as the average of the red, green and blue values.

$$G\mathrm{right}a\mathrm{and}=\frac{1}{3}ROt+\frac{1}{3}G\mathrm{right}\ddot{\mathrm{and}}n+\frac{1}{3}Bla\mathrm{and}$$

In practice it is usual to weight the green value more heavily and the blue value less, so that the formula can look like this, for example. $$G\mathrm{right}a\mathrm{and}=0.3ROt+0.6G\mathrm{right}\ddot{\mathrm{and}}n+0.1Bla\mathrm{and}$$

What is decisive for the further procedure is that there is a camera image 32 in which the light intensity is represented by the values of the individual pixels.

In a camera image 32 from a cooking chamber onto a baking tray with French fries, all pixels with values <20 (on a light intensity scale from 0 (black) to 255 (white)) were marked as "soiled" for evaluation, all other pixels were marked as "not -dirty" is marked so that the following binary image is created from the camera image 32:

see figure 4

The value 20 corresponds to a very dark gray on the scale used. If a pixel is even darker, the light intensity here is extremely low, which can indicate that the incidence of light is being disturbed by dirt at this point. The dirty pixels are marked in white in the figure above, the non-dirty ones are black. It is clearly visible that recognition from a single camera image often does not work and the food to be cooked has a major impact on recognition. For example, the spots between the French fries are incorrectly identified as soiling.

$$\frac{A\left[i,j\right]+B\left[i,j\right]+C\left[i,j\right]}{3}=D\left[i,j\right]$$

$$\frac{20+12+112}{3}=48$$

$$\frac{24+18+21}{3}=21$$

$$\frac{1+10+7}{3}=6$$

$$\frac{23+32+41}{3}=31$$

The use of multiple camera images allows dirt to be detected without the user having to take any specific action. Instead of a single camera image, multiple camera images from consecutive cooking processes are used. An average image 42 is calculated from the multiple camera images, in which the average brightness is calculated for each pixel using a metric. The equation below shows this for 3 sample images, each 2 x 2 pixels. In order to calculate the average image 42 = D, the arithmetic mean of the corresponding pixels from all three images A, B and C is calculated for each pixel in this exemplary embodiment. $$A=\left[\begin{array}{cc}20& 24\\ 1& 23\end{array}\right],B=\left[\begin{array}{cc}12& 18\\ 10& 32\end{array}\right],C=\left[\begin{array}{cc}112& 21\\ 7& 41\end{array}\right],\Rightarrow D=\left[\begin{array}{cc}48& 21\\ 6& 32\end{array}\right]$$

$$\frac{A\left[i,j\right]+B\left[i,j\right]+C\left[i,j\right]}{3}=D\left[i,j\right]$$

$$\frac{20+12+112}{3}=48$$

$$\frac{24+18+21}{3}=21$$

$$\frac{1+10+7}{3}=6$$

$$\frac{23+32+41}{3}=31$$

For example, the average image displayed below can be calculated from several camera images. This average image was determined from only six camera images, so that the different loads in the cooking chamber 6 can still be clearly identified. However, the loads in the cooking chamber 6 are already weakened by the averaging. In contrast to this, however, the contamination is always the same and is not weakened by averaging.

see figure 5

In this image too, all pixels with values < 20 can now be marked as dirty for binarization. The figure shown below shows the contamination determined from the figure above.

see figure 6

This figure already shows a much better match with the actual contamination than the individual camera image shown above. The contamination is shown light after binarization. Based on the number of dirty pixels, a decision can now be made about the need for cleaning. For example, if the pixels are more than 10% dirty, the customer can be informed that the cooking chamber 6 is dirty.

If more images are used to determine the average image, contamination detection can be improved. In general, the more images that are used, the better the recognition works. The image displayed below shows an average image 42 determined from around 100 images from the same cooking chamber 6.

see figure 7

The image displayed below shows the contamination determined using the above image in a binarized image.

see figure 8

With the exception of the regions in the upper corners, which are very dark and are therefore incorrectly marked, the detection corresponds to a very good approximation of the dirt that is actually present.

There are two possible sequences when calculating the mean value and limit value. The way in which first a mean value and then a limit value are formed was described above. Instead of first calculating an average image 42 and then recognizing the contamination from this image using a binarized image, the contamination detection can also first be carried out for each individual camera image using binarization and then the binary contamination images can be averaged using a metric. This can lead to a different result. If a binarization is first carried out individually for each camera image and then the mean value is formed, the result shown below is obtained with the same original camera images:

see figure 9

Here it can be seen that the result in this case is no longer a binary image in which a distinction is made between two categories "dirty, not dirty". Instead, each pixel is given a contamination probability, which is represented by the pixel's gray level in the resulting image. Depending on the application, this approach can deliver better results.

In the result images shown above it can be seen that the upper corners of the image are emphasized like dirt, since they are only poorly illuminated in the given cooking chamber 6 and therefore appear constantly dark like dirt. To prevent this, these regions can be ignored in the evaluation. Alternatively, these inequalities can be corrected prior to evaluation. A median filter, in which each pixel value is replaced by the median of the surrounding pixel values, can be used for this, for example. If a median filter with a size of 201x201 px is applied to the average image shown above, which is created from 100 camera images, the following image is obtained:

see figure 10

If the average image shown above, created from 100 camera images, and the image corrected with the median filter are subtracted from one another, the difference image shown below is obtained:

see figure 11

Dirty areas are again highlighted on the difference image displayed above, it can be seen that the dark upper corners are not incorrectly recognized as dirty. Binarization can be performed again for the difference image in order to separate dirty from non-dirty pixels. The result of this is shown in the image shown below:

see figure 12

Here it can be seen that the large lower soiling is not fully recognized. Depending on the application, the brightness correction can be useful.

The invention is not limited to the present embodiment. For example, the teaching according to the invention can also be advantageously used in cooking appliances other than ovens. The cameras or other optical sensors may also be positioned elsewhere than suggested in the foregoing physical description, inside or outside the cooking cavity, such as in the cooking cavity access door or in front of the cooking cavity access door, such as on the cooking cavity facing side of a handle, and multiple cameras and/or optical sensors may be used.

Reference List

2

Housing

2.1

Cooling channel of the housing 2, designed as an air duct

4

Housing wall designed as baking muffle wall

4.1

Opening in the housing wall 4, for the camera 8

4.2

Cooking chamber ceiling of the housing wall 4

6

Cooking space designed as a baking muffle

8th

Camera designed as a fixed focus camera

8.1

Camera lens 8

10

Lens designed as a converging lens

11

Cooking chamber lighting designed as camera lighting

12

poetry

14

food carrier

16

food

18

insulation layer

19

Beam path of the light

20

Top heat radiator

22

flange

30

cooking appliance

32

camera image

34

control electronics

40

evaluation electronics

42

average picture

44

time

46

image processing software

48

data communication

50

gray image

52

difference image

60

brightness value

70

Averaging calculation

80

limit formation

90

Determination of cleaning requirement

QUOTES INCLUDED IN DESCRIPTION

This list of documents cited by the applicant was generated automatically and is included solely for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Patent Literature Cited

• DE 10128024 A1 [0002]
• DE 102016206483 A1 [0003]
• DE 102014116709 A1 [0004]
• KR 20160069344 A [0005]

Claims (17)

Method for determining the point in time for cleaning a cooking chamber (6) of a cooking appliance (30), in which camera images (32) of a camera (8) directed into the cooking chamber (6) are fed to evaluation electronics (40) which, on the basis of the camera images ( 32) determines the degree of soiling of the cooking chamber (6) via a software-supported image evaluation, characterized in that the brightness values (60) of several or all pixels of the camera images (32) are determined from several camera images (32) via an averaging (70) using a metric and subsequent Limit value formation (80) or a limit value formation (80) and subsequent averaging (70) is decided by means of a metric whether cleaning is necessary. procedure after claim 1 , characterized in that during the averaging (70) by means of a metric and subsequent averaging (80), first an average image (42) is calculated, then an average brightness value (60) is calculated for a plurality or all pixels of the average image (42), then the average brightness values (60) of the pixels are compared with a limit value, which represents a binarizing evaluation threshold for distinguishing the average brightness values (60) between a pixel rated as dirty or not dirty, and then based on the number of pixels as dirty after the limit value comparison assessed pixel it is decided whether cleaning is necessary. procedure after claim 1 , characterized in that in the limit value formation (80) and subsequent averaging (70) by means of a metric, first an individual brightness value (60) is calculated for a majority or all pixels of each camera image (32), then the individual brightness values (60) of the pixels be compared with a limit value, which represents a binarizing evaluation threshold for the differentiation of the individual brightness values (60) between a pixel of the camera image (32) evaluated as dirty or not dirty, a binary dirt image for each camera image from the pixels evaluated as dirty or not dirty (32) is created, and then an average image (42) is calculated from a plurality or all pixels of the binary contamination images, in which the gray level value of each pixel corresponds to a contamination probability, and the gray level values of the pixels to an average contamination probability value d es average image (42) are calculated, which is compared with a limit value and the cleaning is assessed as necessary if the limit value is exceeded. Method according to one of the preceding claims, characterized in that some or all of the multiple camera images (32) are colored images in which the color channels of the individual pixels are evaluated differently or further calculated without the color values. Method according to one of the preceding claims, characterized in that the camera images (32) are made by the camera (8) through a transparent layer which separates the camera (8) from the cooking chamber (6), the transparent layer being part of the Forms inner wall of the cooking chamber (6). Method according to one of the preceding claims, characterized in that the degree of soiling of the transparent layer serves as an indicator for the degree of soiling of the cooking chamber (6). Method according to one of the preceding claims, characterized in that the several camera images (32) show several different cooking processes and/or loading situations. Method according to one of the preceding claims, characterized in that the evaluation electronics (40) evaluate the quality of the average image (42) or a derivation formed therefrom using an evaluation algorithm and a point in time for cleaning the cooking chamber (6) is only determined if the Quality of the average image (42) or a derivation formed from it has been assessed as sufficient. Method according to one of the preceding claims, characterized in that the degree of soiling of pixels or pixel fields of the average image (42) or a derivation formed therefrom is weighted differently in different sectors of the average image (42) or a derivation formed therefrom. Method according to one of the preceding claims, characterized in that the classification of the pixels and/or pixel fields takes place taking into account probabilities of the respective classification. Method according to one of the preceding claims, characterized in that structures with a small size are removed from the average image (42) or a derivation formed therefrom in a correction run. Method according to one of the preceding claims, characterized in that the average image (42) is converted into a gray image (50). procedure after claim 12 , characterized in that structures with a small size are removed from the gray image (50) in a correction run. Method according to one of the preceding claims, characterized in that the gray image (50) or the average image (42) is converted into a difference image (52) by filtering. procedure after Claim 14 , characterized in that structures with a small size are removed from the difference image (52) in a correction run. Method according to one of the preceding claims, characterized in that certain areas of the average image (42), the gray image (50) or the difference image (52) are ignored by the evaluation electronics (40). Method according to one of the preceding claims, characterized in that the evaluation electronics delete previously recorded camera images (32) after the cooking chamber (6) has been cleaned.

Images