DE102017220243B3 - METHOD AND DEVICE FOR DETECTING A BLOCKING OF A RADAR DEVICE, DRIVER ASSISTANCE SYSTEM AND VEHICLE - Google Patents

Abstract

The invention relates to a method for detecting a blockage of a radar device (5) based on a radar image (10a), which is obtained based on radar data acquired by the radar device (5), wherein the radar image (10a) has a plurality of pixels, wherein the Pixels are assigned at least one attribute and wherein at each cycle the following steps are performed: predicting (S1) a prediction radar image (10b) using the radar image (10a) of the previous cycle; Detecting (S2) new radar data by the radar device (5); Assigning (S3) measurement data clusters to the corresponding pixels of the radar image (10a) based on the acquired new radar data; Generating (S4) an updated radar image (10b) by updating the attributes of the pixels of the radar image (10a) based on the measurement data cluster associated with the respective pixel; Comparing (S5) the prediction radar image with the updated radar image (10b); and detecting (S6) whether a blockage of the radar device (5) has occurred based on the comparison of the prediction radar image and the updated radar image (10b).

Description

The invention relates to a method and a device for detecting a blockage of a radar device, a driver assistance system for a vehicle and a vehicle.

Modern vehicles typically include multiple radar devices for obtaining information about the environment of the vehicle. The radar devices are located close to the ground around the vehicle and are exposed to whirled up dust, mud, fog, ice and snow. Accordingly, a blockage of the radar device poses a serious problem because it results in a reduced quality of the radar data acquired by the radar device. The blocking can be done quickly, for example due to slush. However, slow changes are possible as well, for example due to a slowly freezing water film on the radar device.

Accordingly, it is very important to detect changes and anomalies as early as possible in order to avoid false predictions based on erroneous radar data.

Typically, anomalies are detected using three criteria: range, timeout, and object loss.

One method of detecting a blockage is from the document US 20090243912 A1 known. Within a virtual detection zone, a target object is detected, target-related information is collected, and based on the collected information, it is determined whether a "missed" alert signal should be generated.

The document WO 2004006119 A2 relates to an anomaly detection system for detecting anomalies in a medium having known or predictable statistical characteristics.

Statistics-based detection of anomalies is generally difficult. For example, the radar image may not always be displayed in a Gaussian distribution. Accordingly, anomaly detection based on means or variances generally only works for low-dimensional data streams. There may be a lack of clear boundaries between abnormal data and radar data of sufficient quality that affect exceptional environments. For example, an abnormality detection based on a time-out is difficult when the vehicle is running on a road with no traffic and no edge trim. If there is a layer of water or snow on the road, there are almost no reflections and the system may erroneously report a stall.

In addition, different attributes of a radar image may generally correlate with each other, which is separately lost by calculating averages for each attribute.

Determining anomalies based on histogram averages and low pass filters, as is commonly done for range criteria, can inadvertently eliminate crucial information. Tracking objects can keep objects alive for several cycles, even if the radar has lost the object. In addition, a response to a sudden change may be too slow and limited. For safety reasons, however, a sudden blockage, for example due to mud from the roadside, should be detected immediately. Slow reaction times can lead to misrecognition and wrong decisions of a driver assistance system.

In view of the foregoing, it is therefore an object of the present invention to provide an approach for detecting radar blocking as early as possible.

According to the invention, a method is provided for detecting a blockage of a radar device based on a radar image, as specified in claim 1. The invention further provides an apparatus for detecting a blockage of a radar device based on a radar image as recited in claim 13. Furthermore, the invention relates to a driver assistance system for a vehicle as specified in claim 14, and a vehicle as specified in claim 15.

According to a first aspect, therefore, the invention provides a method for detecting a blockage of the radar device based on a radar image. The radar image comprises a plurality of pixels, wherein the pixels are assigned at least one attribute. At each cycle, a predictive radar image is predicted using the radar image from the previous cycle. Furthermore, the radar device detects new radar data. Based on the acquired new radar data, measurement data clusters are assigned to the corresponding pixels of the radar image. An updated radar image is generated by updating the attributes of the pixels of the radar image based on the measurement data cluster associated with the respective pixel. The prediction radar image is compared with the updated radar image, and based on the comparison of the prediction radar image and the updated radar image, it is detected whether a jam of the radar device has occurred.

According to a second aspect, the invention provides a device for detecting a blockage of a radar device based on the radar image. The radar image has a plurality of pixels, wherein the pixels are assigned at least one attribute. The device includes a computing unit that is configured to perform multiple steps each cycle. Specifically, using the radar image of the previous cycle, the arithmetic unit predicts a prediction radar image, receives new radar data detected by the radar device, and assigns measurement data clusters to the corresponding pixels of the radar image based on the acquired new radar data. The computing unit also generates an updated radar image by updating the attributes of the pixels of the radar image based on the measurement data cluster associated with the respective pixel. The arithmetic unit compares the prediction radar image with the updated radar image. The arithmetic unit detects whether a blockage of the radar device has occurred based on the comparison of the prediction radar image and the updated radar image.

According to a third aspect, the invention provides a driver assistance system for a vehicle comprising a radar device adapted to generate radar data and a device for detecting a blockage of the radar device based on the radar data.

According to a fourth aspect, the invention provides a vehicle comprising a driver assistance system.

Various preferred features of the invention are set forth in the dependent claims.

Measured radar data is continuously combined to create a kind of "film" of the environment of the vehicle. The environment of the vehicle must change according to the laws of physics, i. H. new objects can emerge at the borders, and slow changes can occur around the vehicle. If an anomaly is detected in the movie, the change does not match the expected laws and blocking is detected. To define what it means that the continuous data flow or data stream meets the expected laws, i. H. changes naturally, a comparison object is required. This comparison object is provided by providing a prediction radar image based on the history of the measured radar data. The difference between the prediction radar image and the updated radar image should be small, i. H. only correspond to reasonable deviations.

Larger, unnatural deviations can be detected by statistical methods, in particular martingale algorithms and interchangeability tests. Anomaly detection is generally applicable to any radar image and does not require a particular statistical distribution of the attributes, such as a Gaussian distribution.

In addition, anomalies are detected immediately and possible false predictions based on incorrect radar data can be avoided. The quality of radar-based predictions is improved. Improved quality goes hand in hand with better radar-based decision-making and vehicle safety can be increased.

In accordance with another embodiment of the method, the predictive radar image is predicted based on a Kalman filter method. Kalman filtering takes into account a series of prior radar images and uses Bayesian inference to predict a common probability distribution across the variables for future time points.

According to a preferred embodiment of the method, measurement data clusters are assigned to the corresponding pixels of the radar image based on a spatial position associated with the measurement and a spatial region corresponding to the pixel. The measurements may include information such as position (s), velocity, radar cross section (RCS), and microdoppler effect information concerning objects or reflection points at the spatial position.

According to a further embodiment of the method, it is determined whether a difference between at least one of the attributes, which is determined based on the measurement, differs by more than a predetermined threshold from the corresponding attribute of the pixel having a corresponding spatial area containing the includes the spatial position associated with the measurement. If so, the measurement is associated with a pixel having a corresponding spatial area adjacent to the spatial location associated with the measurement. The measurements can thus be assigned to neighboring pixels in order to provide a radar image with better contrast. For example, if both moving and stationary objects are detected in the same pixel, the system searches for an adjacent pixel and assigns one of the two types of clusters to the adjacent lower-probability pixel.

According to another embodiment of the method, updating the attributes of a pixel of the radar image comprises extracting new attribute values for each measurement associated with the pixel, calculating weighted averages for each attribute based on the extracted new attribute values, and updating the attributes of the pixel using the calculated one weighted average of each attribute.

According to another embodiment of the method, calculating the weighted averages for each attribute comprises weighting the extracted new attribute values based on a Mahalanobis distance between a spatial position associated with the corresponding measurement and a midpoint of a spatial region corresponding to the pixel.

According to a further embodiment of the method, the attributes comprise at least one of a radar cross section, a relative speed and a probability of error detection.

According to another embodiment of the method, comparing the prediction radar image with the updated radar image comprises at least one of an interchangeability test and a martingale difference test.

According to one embodiment, comparing the predicted radar image with the updated radar image comprises determining at least one respective zone for each of the prediction radar image and the updated radar image, the zone having a plurality of pixels. At least one of an interchangeability test and a martingale difference test is applied to the respective zones. The zones may be determined based on differences between the corresponding pixels of the radar image and the prediction radar image. If such differences exceed a certain threshold, the pixel is assigned to each zone.

In accordance with another embodiment of the method, comparing the prediction radar image with the updated radar image includes determining an extraneous measure indicative of a difference between the prediction radar image and the updated radar image using a classifier. Based on the foreignness measure, the interchangeability is checked. Based on the interchangeability test, a martingale is calculated.

According to another embodiment of the method, the classifier is determined based on a support vector machine that is generated based on learning data. The classifier determines distances between the updated radar image and / or the prediction radar image from a hyperplane generated by the Support Vector Machine. The Support Vector Machine is thus generated based on offline training data.

In accordance with another embodiment of the method, if a stall has been detected, a stall area is identified based on the comparison of the prediction radar image with the updated radar image.

In another embodiment, a derivative of the foreign function may be analyzed to determine anomalies.

According to an embodiment of the method, the martingale value is evaluated using a test including the Doob's maximum inequality test and / or a test involving the Hoeffding-Azuma inequality to identify a concept change and blocking of the radar device is detected when a concept change is identified ,

• 1 ein schematisches Blockdiagramm einer Vorrichtung zum Erkennen einer Blockierung einer Radarvorrichtung gemäß einer Ausführungsform der Erfindung zeigt;
• 2 ein beispielhaftes Radarbild zeigt;
• 3 ein beispielhaftes neues Radarbild zeigt;
• 4 ein schematisches Blockdiagramm eines Fahrerassistenzsystems gemäß einer Ausführungsform der Erfindung zeigt;
• 5 ein schematisches Blockdiagramm eines Fahrzeugs gemäß einer Ausführungsform der Erfindung zeigt; und
• 6 ein schematisches Ablaufdiagramm eines Verfahrens zum Erkennen einer Blockierung einer Radarvorrichtung gemäß einer Ausführungsform der Erfindung zeigt.

For a more complete understanding of the invention and the advantages thereof, in the following description, embodiments of the invention will be explained in more detail, with reference to the accompanying drawing figures in which like reference numerals designate like parts and in which:

• 1 shows a schematic block diagram of a device for detecting a blockage of a radar device according to an embodiment of the invention;
• 2 shows an exemplary radar image;
• 3 shows an exemplary new radar image;
• 4 shows a schematic block diagram of a driver assistance system according to an embodiment of the invention;
• 5 shows a schematic block diagram of a vehicle according to an embodiment of the invention; and
• 6 a schematic flow diagram of a method for detecting a blockage of a radar device according to an embodiment of the invention shows.

The accompanying drawings are included for a deeper understanding of the present invention. The elements of the drawings are not necessarily drawn to scale with respect to each other. It is further understood that certain process steps in a particular Order can be described, but can be performed simultaneously or in a different order.

1 shows a schematic block diagram of a device 1 for detecting a blockage of a radar device.

The device 1 includes an interface 2 which is connected to the radar device and adapted to receive radar data from the radar device.

The device 1 further comprises a computing unit 3 that with the interface 2 connected and designed by the interface 2 to analyze the received radar data.

The arithmetic unit 3 is configured to generate a radar image and to update the radar image based on the radar data. The radar image includes a plurality of pixels. Each pixel represents a specific spatial area with a given dimension in an environment of a vehicle. The arithmetic unit 3 assigns attributes to each pixel. The attributes relate to objects detected by the radar device within the area associated with the pixel. The attributes may include a radar cross-section, a relative velocity, and a probability of detecting a possible object within the spatial region corresponding to the pixel. The attributes may be averaged over clusters associated with the pixel, with each element of the cluster representing individual measurements of the attributes based on the radar data.

The arithmetic unit 3 predicts a prediction radar image using the current radar image. The prediction radar image may be obtained using a Kalman filter from the current radar image, preferably also taking into account the history of the radar image. The prediction radar image is stored in a memory unit of the device 1 saved.

Now the radar device detects new radar data passing through the interface 2 received and to the arithmetic unit 3 be sent. The arithmetic unit 3 updates the radar image based on the acquired new radar data. The radar data corresponds to clusters or sets of measurements that can be assigned to a respective spatial area. Each cluster is thus assigned to the pixel that corresponds to the respective spatial area. The cluster can also be assigned to neighboring pixels, if in this way sharper separated values of the attributes can be obtained. In addition, impossible or improbable clusters can be filtered out.

Each cluster has several attributes that correspond to the attributes of the pixel. Values of the attributes are used to adjust or update the current values of the attributes. Thus, a new radar image is generated. The attributes of the pixels of the new radar image may be obtained using weighted averages of the corresponding value of the attribute of each element of the cluster associated with the pixel, along with the previous value of the attribute. A respective weight may be calculated using a Mahanalobis distance to the value of the pixel, i. H. a center of a pixel-corresponding spatial area.

As an example, in 2 a radar picture 10a with pixels P11 shown until Pnm. The size of the radar image is determined by the spatial range that can be detected by the radar device. For each pixel, respective measurements of one of the attributes are shown, the three different values A1 . A2 . A3 of the attribute. The values can be, for example, moving objects A1 , unknown objects A2 and static objects A3 correspond. The pixel in turn becomes one of the values based on an averaging process over the values of the respective measurements A1 . A2 . A3 assigned. In addition, if no or insufficient measurement (s) are obtained for a pixel, the attribute of the pixel is assigned the value "no object". This in 2 The scenario depicted corresponds to a vehicle that is at a spatial position that corresponds to the middle lower edge of the radar image and that travels along a highway with obstacles along the left and right lanes.

Each attribute or combination of attributes can be assigned a specific color value, depending on the value of the attribute or the combination of attributes. Thus, the arithmetic unit generates 3 a "Radar movie" that shows the temporal changes of the attributes of the radar image.

Now, the arithmetic unit compares the prediction radar image with the updated radar image. First, the arithmetic unit 3 identify certain zones within the radar image, each zone corresponding to a plurality of connected pixels. The zones may be determined based on a difference between the attributes of the new radar image and the prediction radar image. If the difference for one or, alternatively, all of the attributes exceeds respective thresholds, the pixel is assigned to a zone.

According to a first test, the interchangeability of the prediction radar image and the new radar image can be calculated. The interchangeability can also be determined for each zone and / or for the entire radar image. Interchangeability can be tested using a Martingale value.

von Zufallsvariablen ist eine finite Folge von Zufallsvariablen Z1, ..., Zn austauschbar, wenn die gemeinsame Verteilung p(Zl...Zn) im Rahmen einer beliebigen Permutation der Indices der Zufallsvariablen invariant ist.Interchangeability is defined as follows. For a consequence $$\left\{Z_i:1\le i<\infty \right\}$$

of random variables is a finite sequence of random variables Z1 , ..., Zn interchangeable, if the common distribution p (Zl ... Zn) is invariant within any permutation of the indices of the random variables.

derart, dass jedes Mn eine messbare Funktion von Zl...Zn für alle n=0, 1, ... ist, d. h. insbesondere: M0 ist ein konstanter Wert, und der bedingte Erwartungswert von Mn+1, ausgehend von M0, ..., Mn, ist gleich Mn, E(Mn+1|M1...Mn)=Mn.A martingale is a sequence of random variables $$\left\{M_i:1\le i<\infty \right\}$$

such that each Mn is a measurable function of Zl ... Zn for all n = 0, 1, ..., ie in particular: M0 is a constant value, and the conditional expectation of Mn + 1, starting from M0,. .., Mn, is equal to Mn, E (Mn + 1 | M1 ... Mn) = Mn.

Interchangeability is tested using the martingale by observing a new data point. A learning algorithm outputs a positive martingale value that reflects the level of evidence that violates the null hypothesis of interchangeability; H. no anomaly, was determined.

The arithmetic unit 3 can calculate the strangeness of a difference or change between the prediction radar image and the new radar image. The calculation of the foreignness is preferably for each by the arithmetic unit 3 certain zone performed. In addition, the strangeness for the entire radar image can be determined. The strangeness measure may be reduced using a classifier that may be implemented by a support vector machine. The Support Vector Machine can be generated based on offline learning data or observations. The strangeness depends on how much the new radar image differs from the prediction radar image. For each attribute of each pixel, a respective measure of foreignness can be obtained. From the extrinsic measures, p-values for the new radar image are extracted.

According to a statistical theorem, p-values pl, p2, ... output by a random p-value function V are evenly distributed in the interval [0,1], provided that input values z1, z2, .. I are generated by means of an exchangeable probability distribution in the input space. This property of output p-values is violated if the interchangeability condition is no longer met. Accordingly, by examining the distribution of the p-values, the interchangeability of the new radar image and the prediction radar image can be verified or falsified.

Now a random Supermartingal is constructed as follows: $$M_n^{\left(\epsilon \right)}={\displaystyle \underset{i=1}{\overset{n}{\Pi }}\left(\epsilon p_i^{\epsilon -1}\right)},$$

When θn = 1, the p-value function V is deterministic.

Now a concept change can be determined. A sequence of data points with a concept change can be understood as the concatenation of two data segments. By moving a data point Z from a sequence to a position in the next sequence, the data point will pick up. Accordingly, the interchangeability condition is violated. A conceptually stable data stream requires interchangeability. Therefore, the lack of interchangeability suggests a concept change.

When a concept change is detected, the p-values output from the random p-value function become skewed and the p-value distribution is no longer uniform. The Kolmogorov-Smirnov test states that the p-values are no longer uniformly distributed after a concept change. The null hypothesis "The output p-values are uniformly distributed" can be rejected at a significance level a = 0.05 and after generating a sufficient number of data points. The skewed p-value distribution can be used in a martingale test to determine an anomaly because small p-values inflate martingales.

According to a second test, the martingale difference for a change from the prediction radar image to the new radar image may be calculated. The calculation can be carried out for the zones and / or for the entire radar image. After a new radar image has been generated, a hypothesis test is made to determine if there is a concept change in the data stream. The decision may be based on whether the interchangeability condition is violated, which in turn is based on the martingale value.

The hypothesis tests may be based on the Doob's maximal inequality or on the Hoeffding-Azuma inequality. The null hypothesis H0 : "There was no concept change in the data stream" with the alternative hypothesis H1 : "A concept change occurs in the data stream". The test will continue as long as a first martingale test results: $$0<M_n^{\left(\epsilon \right)}<\lambda ,$$

Where A is a positive number. The hypothesis is rejected if $$M_n^{\left(\epsilon \right)}\ge \lambda ,$$

Alternatively, the test continues as long as a second martingale test results: $$0<\left|M_n^{\left(\epsilon \right)}-M_{n-1}^{\left(\epsilon \right)}\right|<t,$$

Where t is a positive number. The test will be rejected if: $$\left|M_n^{\left(\epsilon \right)}-M_{n-1}^{\left(\epsilon \right)}\right|\ge t,$$

The algorithm is preferably based on an offline learning process for generating a support vector machine. The Support Vector Machine is a set of parameters and thresholds that are hard-coded for the online algorithm. A currently identified tagged stream can be used to generate the Support Vector Machine as an offline classifier. The foreignness measure described above may consist of the distances starting from the hyperplane of the radar image.

Based on the result of at least one of the first and second tests, the arithmetic unit decides 3 whether an anomaly is detected in the data stream. Thus, based on the above criteria, a partial or complete blockage of the radar device can be calculated.

As an example, in 3 a new radar picture 10b shown taking into account new radar data. New moving objects and unknown objects were identified in a central area near the vehicle, positioned in a spatial area corresponding to the center of the lower edge of the radar image. However, the prediction radar image is much more similar to the one in 2 shown original radar image. In particular, the central areas of the prediction image and the new image differ significantly from each other. Accordingly, the first and second tests described above determine the deviations between the prediction radar image and the new radar image 10b do not correspond to natural fluctuations. In other words, the new radar image 10b and the predictive radar image can not be linked together in a statistically natural way. In the central area, a blockage is detected, for example due to an insect on the radar sensor.

If a blockage has been detected, the arithmetic unit tries 3 to restore the radar image based on the previous radar image and the prediction radar image. Otherwise, if no blockage is detected, the radar device makes new measurements and the blockage detection is repeated.

In order to also detect slow partial blocking, a relative image may be added, with each pixel of the relative image including thresholds for the attributes that should not be exceeded. The relative image corresponds to the maximum degradation of each pixel in the structure. The arithmetic unit 3 compares the new radar image with the relative image as well. If at least one of the attributes of at least one of the pixels exceeds the threshold value of the corresponding pixel of the relative image, blocking is detected.

In general, the algorithm can make abnormal changes in just a few radar cycles, i. H. preferably at most 20 radar cycles.

4 provides a schematic block diagram of a driver assistance system 4 dar. The driver assistance system 4 includes a radar device 5 , which is adapted to detect radar data and the radar data to an interface of a device 1 for detecting a blockage of the radar device 5 transferred to. The device 1 may be configured according to any of the above outlined embodiments.

The driver assistance system 4 may be further configured to control driving functions of a vehicle based on the result of the stall detection. For example, the radar data may be used to analyze the environment of the vehicle and the vehicle may be steered, accelerated or decelerated based on the result of the radar data. After a stall has been detected, the radar data can either no longer be analyzed or only partially analyzed to control the driving functions.

In addition, the driver assistance system may be configured to output a warning signal to a driver of the vehicle when a blockage has been detected. The radar device may also be disabled after detection of a stall.

5 shows a schematic block diagram of a driver's toy 6 , which is a driver assistance system 4 as described above.

6 FIG. 10 is a flowchart illustrating a method of detecting a blockage of a radar apparatus. FIG 5 represents.

In the first process step S1 A radar image is predicted using the radar image or multiple radar images of the previous cycle or cycles. The radar image has a plurality of pixels, with attributes associated with the pixels, as described above. The prediction can be based on a Kalman filter.

In a further process step S2 the radar device detects new radar data.

Based on the acquired new radar data, measurement data clusters are assigned to the corresponding pixels of the radar image, S3 ,

The radar data is analyzed S4 to generate an updated radar image. Based on the radar device, measurement data clusters are generated and assigned to respective pixels. The attributes of the pixels of the radar image are updated based on the measurement clusters.

In a further step S5 the prediction radar image is compared with the updated radar image. The comparison may include the interchangeability conditions and martingale algorithms outlined above.

Based on the comparison of the prediction radar image and the updated radar image, the radar device becomes locked 5 recognized S6 , The detection can be any of the above for the device 1 include described steps.

LIST OF REFERENCE NUMBERS

1

contraption

2

interface

3

computer unit

4

Driver assistance system

5

radar device

6

vehicle

10a

radar image

10b

new radar picture

Pij

pixel

Claims (15)

A method of detecting a blockage of a radar device (5) based on a radar image (10a) obtained based on radar data acquired by the radar device (5), the radar image (10a) comprising a plurality of pixels, the pixels having at least one attribute and with each cycle performing the following steps: Predicting (S1) a prediction radar image using the radar image (10a) of the previous cycle; Detecting (S2) new radar data by the radar device (5); Assigning (S3) measurement data clusters to the corresponding pixels of the radar image (10a) based on the acquired new radar data; Generating (S4) an updated radar image (10b) by updating the attributes of the pixels of the radar image (10a) based on the measurement data cluster associated with the respective pixel; Comparing (S5) the prediction radar image with the updated radar image (10b); and Detecting (S6) whether a blockage of the radar device (5) has occurred based on the comparison of the prediction radar image and the updated radar image (10b). Method according to Claim 1 wherein the prediction radar image is predicted based on a Kalman filter method. Method according to Claim 1 or 2 wherein measurement data clusters are assigned to the corresponding pixels of the radar image (10a) based on a spatial position associated with the measurement and a spatial region corresponding to the pixel. Method according to Claim 3 wherein a measurement is associated with a pixel having a corresponding spatial area adjacent to the spatial location associated with the measurement when a difference between at least one of the attributes determined based on the measurement is greater than a predetermined threshold of deviates from the corresponding attribute of the pixel having a corresponding spatial area that includes the spatial position associated with the measurement. The method of any one of the preceding claims, wherein updating the attributes of a pixel of the radar image (10a) comprises: Extracting new attribute values for each measurement associated with the pixel; Calculating weighted averages for each attribute based on the extracted new attribute values; and updating the attributes of the pixel using the calculated weighted average of the respective attribute. The method of claim 1, wherein calculating the weighted averages for each attribute comprises weighting the extracted new attribute values based on a Mahalanobis distance between a spatial position associated with the corresponding measurement and a midpoint of a spatial region corresponding to the pixel. The method of any one of the preceding claims, wherein the attributes include at least one of a radar cross-section, a relative velocity, and a probability of detecting a misrecognition. The method of any one of the preceding claims, wherein comparing the prediction radar image with the updated radar image (10b) comprises at least one of an interchangeability test and a martingale difference test. Method according to Claim 8 wherein comparing the prediction radar image with the updated radar image (10b) comprises determining at least one respective zone having a plurality of pixels for each of the prediction radar image and the updated radar image (10b) and applying at least one of an interchangeability test and a Martingaldifferenztest on the corresponding zones. Method according to one of Claims 8 or 9 wherein comparing the prediction radar image with the updated radar image (10b) comprises the steps of: determining an extraneous measure indicative of a difference between the prediction radar image and the updated radar image (10b) using a classifier; Checking for interchangeability based on the degree of foreignness; and calculating a martingale value based on the interchangeability test. Method according to Claim 10 wherein the classifier is determined based on a support vector machine generated based on learning data, wherein the classifier determines distances between the updated radar image (10b) and / or the prediction radar image from a hyperplane generated by the support vector machine. A method according to any one of the preceding claims, wherein, when a blockage has been detected, a blocking area is identified based on the comparison of the prediction radar image with the updated radar image (10b). A device (1) for detecting a blockage of a radar device (5) based on a radar image (10a) obtained based on radar data acquired by the radar device (5), the radar image (10a) comprising a plurality of pixels, the pixels at least one attribute is assigned, the device (1) comprising a computing unit (3) which is designed to carry out the following steps during each cycle: Predicting a prediction radar image using the radar image (10a) of the previous cycle; Receiving new radar data detected by the radar device (5); Associating measurement data clusters with the corresponding pixels of the radar image (10a) based on the acquired new radar data; Generating an updated radar image (10b) by updating the attributes of the pixels of the radar image (10a) based on the measurement data cluster associated with the respective pixel; Comparing the prediction radar image with the updated radar image (10b); and Detecting whether a blockage of the radar device (1) has occurred based on the comparison of the prediction radar image and the updated radar image (10b). A driver assistance system (4) for a vehicle (6), comprising: a radar device (5) adapted to generate radar data; and a device (1) according to Claim 13 , Vehicle (6), the a driver assistance system (4) after Claim 14 includes.

Images