DE102019219144A1 - Method and system for image evaluation for object recognition in the vicinity of vehicles - Google Patents

Abstract

The invention relates to a method for evaluating an image (34) of an environment (12) of a vehicle (10) using an object recognition algorithm (32) based on predetermined parameters (PAR) in order to identify objects (20) using the image (34) the surroundings (12) of the vehicle (10) to be recognized. In order to improve the reliability and quality of the detection of objects (20), the invention provides that the parameters (PAR) are specified as a function of a geographical position (POS) of the vehicle (10). The invention also proposes a system (100) for evaluating respective images (34) of respective surroundings (12) of a multiplicity of vehicles (10-1, 10-2, ...) in order to identify respective objects (20) in the respective The surroundings (12) of the vehicles (10-1, 10-2, ...) can be recognized.

Description

The present invention relates to the field of object recognition in the vicinity of vehicles, and in particular to a method for evaluating an image of an environment of a vehicle, an object recognition algorithm operating on the basis of predetermined parameters being used to identify objects in the vicinity of the vehicle on the basis of the image . The invention also relates to a system for evaluating respective images of respective surroundings of a multiplicity of vehicles in order to recognize respective objects in the respective surroundings of the vehicles on the basis of the respective images.

Processes for image evaluation carried out on board vehicles are known in various designs from the prior art and are becoming increasingly important, especially in the field of automotive technology for road vehicles, for example in the implementation and improvement of comfort functions (e.g. parking aid, cruise control, lane departure warning, automatic traffic sign recognition, etc.) or safety functions (e.g. obstacle recognition, etc.).

Particularly high demands on the reliability and quality of an object recognition accomplished with such an image evaluation method result, for. B. for autonomous vehicles in which control processes such as steering, accelerating and decelerating are partially or fully automated.

The image of the surroundings of the vehicle can, for. B. be recorded by means of at least one video camera arranged on the vehicle, the z. B. is sensitive in the visible spectral range and provides digital representations (image data) to a digital electronic evaluation device on board the vehicle, in which the object recognition algorithm runs in order to recognize objects (e.g. lane markings, traffic signs, other vehicles, pedestrians, etc.) .

From the prior art, there are known object recognition algorithms in various configurations that are suitable for this purpose and that work on the basis of predetermined parameters (algorithm parameters).

In particular, against the background of the requirement in many applications for image evaluation to be carried out very quickly (in "real time") with limited computing capacity in an evaluation device, it is fundamentally a major challenge to achieve object recognition with high reliability and quality, i. H. For example, reliably recognizing actually existing objects of a certain type (e.g. vehicles driving ahead) as such and at the same time avoiding incorrect recognition of certain objects that actually do not exist at all (or actually represent objects of a different type).

A certain improvement in this regard has already been achieved through the use of so-called artificial intelligence (AI), i. H. the use of self-learning and / or dedicated trained object recognition algorithms that were trained with the aid of reference objects (preferably of a large number) with the aim of achieving the highest possible rate of correct object recognition.

Corresponding training data can, for. B. determined by test drives with a vehicle and subsequently made available to the object recognition algorithm as training data.

For the practical implementation of AI, e.g. For example, an artificial neural network with a large number of neurons and connections between these neurons can be used, in which case the parameters of the object recognition algorithm are given by characteristics (e.g. weightings) of the latter connections between neurons.

Nevertheless, there is still room for improvement in terms of the reliability and quality of object recognition when using AI.

It is an object of the present invention to show a novel way with which the reliability and quality of a detection of objects in the vicinity of vehicles can be further improved by means of an image evaluation of the type mentioned at the beginning.

According to a first aspect of the invention, this object is achieved in a method of the type mentioned at the outset in that the parameters of the object recognition algorithm are specified as a function of a geographical position of the vehicle.

In this way, when evaluating the image, information that is relevant to object recognition but cannot be derived from the image as such can be used in order to significantly improve object recognition in practice. The invention is based on the knowledge that in particular z. B. Occurrence probabilities of objects in practice are location-dependent and thus the information about the geographical position of the vehicle can be used to improve object recognition.

For example, the probability of a tractor appearing in a rural area is greater than in the center of a large city, the probability of small people (children) appearing in the vicinity of a playground is greater than it is further away, etc. Through a specific specification of the algorithm parameters in Depending on the geographical position, such correlations can be used advantageously in the context of object recognition in order to improve the reliability and quality of the object recognition.

The image of the surroundings of the vehicle can, for. B. be recorded by means of an image recording device provided in the vehicle, the image recording device for this purpose, for. B. may have one or more video cameras arranged on the vehicle. Each such video camera can e.g. B. be sensitive in the visible spectral range. Alternatively or additionally, z. B. at least one IR-sensitive camera can be provided. The image recording device can take the image (or video image sequences) z. B. as digital representations (image data) to a digital electronic evaluation device in which the object recognition algorithm runs. As an alternative or in addition to a video camera, the image recording device can also use other sensors such as, in particular, e.g. B. have a lidar system and / or a radar system, which can also be arranged on board the vehicle. Within the scope of the invention it can also be provided that to obtain image data representing the image, sensors arranged or located outside of the vehicle are also used (e.g. sensor devices arranged in other vehicles) whose data are e.g. B. via telecommunication (z. B. "V2x") to the evaluation device.

In one embodiment, the object recognition algorithm is designed to assign respective probabilities of occurrence to predetermined types of objects and to take them into account for the recognition of objects, these probabilities of occurrence being determined by the specified parameters (i.e. at least partly determined or influenced by them). If z. For example, when a “normal” set of certain parameters is specified, the occurrence of an object type “ball (e.g. football)” and an object type “plastic bag” are assigned the same probability of occurrence and the algorithm then searches for “objective”, i. H. only makes a decision based on criteria based on the image information of the image (e.g. in favor of "ball"), then by specifying a different set of parameters based on the changed associated probabilities of occurrence, the decision in the otherwise same situation could also be in favor of the other object type ( Plastic bag) or z. B. in favor of a completely different property type (e.g. "animal").

The predetermined types of objects can e.g. B. be defined according to a classification of basically possible objects in the vicinity of a vehicle.

The predetermined types of objects can e.g. B. comprise at least one type of object selected from the group consisting of lane delimitation, lane marking, traffic sign, traffic light system, vehicle, person.

The types of objects can be defined adapted to the respective specific application, and for example, as an alternative or in addition to a more roughly defined object type such as. B. "Vehicle" also has a finer classification z. B. based on the type of vehicle, resulting, for example, in more meaningful object types such as cars, trucks, motorcycles, scooters, bicycles, bicycles with trailers, scooters, etc. . Each of the above-mentioned object types) each have to be classified more finely. In the case of the property type "Person", z. B. between "large person" (adult) and "small person" (child) and with the object type "traffic sign" could z. B. distinguish between traffic signs relating to stationary traffic and traffic signs relating to moving traffic, etc.

Advantageously, with the invention of the geographical position-dependent occurrence probabilities in the object recognition from the object recognition algorithm, in particular z. B. be taken into account in the decision-making process in cases of doubt.

For example, could thus z. B. for a geographical position in a very rural area, a probability of occurrence for agricultural vehicles (z. B. tractors etc.) is greater and z. B. a probability of occurrence for scooters can be assumed to be smaller than the respective probability of occurrence at geographical positions that are within a city.

The parameters to be specified for the object recognition algorithm can advantageously influence the decision-making process of the object recognition algorithm in the direction of more reliable and more precise object recognition.

• - Übermitteln von Anpassungsdaten von einem Server an eine den Objekterkennungsalgorithmus durchführende Auswerteeinrichtung, wobei die Anpassungsdaten von dem Server basierend auf der geografischen Position des Fahrzeuges und basierend auf in dem Server für ein bestimmtes geografisches Gebiet gespeicherten positionsabhängigen Umgebungsdaten ermittelt wurden, und
• - Anpassen der Parameter für den Objekterkennungsalgorithmus basierend auf den von dem Server an die Auswerteeinrichtung übermittelten Anpassungsdaten.

In one embodiment, the method also has the following steps:

• - Transmission of adjustment data from a server to an evaluation device performing the object recognition algorithm, the adjustment data being determined by the server based on the geographical position of the vehicle and based on position-dependent environmental data stored in the server for a specific geographical area, and
• Adaptation of the parameters for the object recognition algorithm based on the adaptation data transmitted from the server to the evaluation device.

The term “server” is to be understood broadly in the context of the invention as a device in which position-dependent environmental data are stored and which can determine corresponding adaptation data for predetermined geographical positions (of the vehicle). The server can, in particular, be a central server (e.g. stationary data center, including cloud-based servers) which provides the aforementioned adaptation data for a large number of vehicles. Alternatively, however, the server can also be arranged entirely or partially in the vehicle, for example implemented as a functional component of an evaluation device arranged in the vehicle. In particular for applications of the invention with a limited (smaller) geographical extent, the environmental data can also be kept on board the vehicle and z. B. be updated from time to time via a telecommunications connection (cellular network, mobile Internet connection, etc.).

The geographical area for which the position-dependent environmental data is stored in the server preferably has a size of at least 100 km ² , more preferably at least 10,000 km ² , or even more, and can e.g. B. the territory of a certain state (Germany, USA, China etc.) or the territory of an association of states (e.g. European Union etc.) or be limited by such.

In one embodiment variant, the environmental data as such already represent the adjustment data or contain the adjustment data, so that the determination of the adjustment data to be transmitted to the evaluation device then z. B. by selecting a special subset of the environment data, namely corresponding to the geographical position of the vehicle in question, can be realized from the environment data stored in total in the server.

Alternatively or additionally, however, z. For example, a determination algorithm can be used by means of which, based on the geographical position of the vehicle, environmental data relevant to the vehicle is selected from the entirety of the position-dependent environmental data and further processed to form the adaptation data relevant to the vehicle. The server can here, for. B. can be implemented as a completely external device (outside the vehicle), or completely or at least partially by a device arranged in the vehicle with appropriate storage and computing capacity (e.g. as a functional component of an evaluation device on board the vehicle) .

For example, the environmental data could here, for. B. specify more general characteristics of a relevant environment (e.g. "rural area", "urban area", "inner city area", "playground area" etc.), which are then used by the determination algorithm to generate the relevant specific adjustment data. The adjustment data can then e.g. B. contain specific specifications of the parameter specification for the object recognition algorithm and / or in particular z. B. contain these parameters as such (or at least a part of the parameters to be adjusted).

• - Erfassen der geografischen Position des Fahrzeuges, und
• - Übermitteln der erfassten geografischen Position des Fahrzeuges an den Server.

In a further development, the method also has the following steps:

• - Detecting the geographical position of the vehicle, and
• - Transmission of the recorded geographic position of the vehicle to the server.

The detection of the geographical position of the vehicle can in particular, for. B. by means of a location device arranged in the vehicle. In particular, satellite-supported localization of the vehicle can be provided for detecting the geographical position of the vehicle, for example using a GPS receiving device arranged in the vehicle.

The transmission of the detected geographical position of the vehicle can, for. B. using a cellular network and z. B. using an (often already existing) Internet connection of the vehicle.

In one embodiment, at least one component of the evaluation device, or z. B. the entire evaluation device, arranged in the vehicle, in which case the transmission of the adaptation data from the server to the vehicle or to this component of the evaluation device can be implemented by a so-called geocasting of the adaptation data to vehicles located within the geographical area.

When geocasting certain data, here z. B. the adjustment data provided for a specific vehicle, transmitted to recipients within a spatially delimited geographical area.

In this embodiment of the invention, the spatially delimited area z. B. a size of preferably at least 0.01 km ² , in particular at least 0.05 km ² and / or a maximum of z. B. 5 km ² , in particular a maximum of 1 km ² .

It should be borne in mind (e.g. when using geocasting) that the adaptation data can be created and transmitted to vehicles all the more relevant and useful, the smaller spatially delimited areas are selected (e.g. in the context of geocasting) certain (identical) adjustment dates are assigned. This results in z. B. an expedient maximum extent of these individual areas in order to ensure a high benefit of the adjustment data. On the other hand, however, a certain minimum extent of these areas can also be appropriate, e.g. B. on the one hand because the probabilities of occurrence of objects on very short distances (e.g. over a distance of about 100 m) change rather imperceptibly in practice, and on the other hand because the effort for communicating the adaptation data with decreasing size of the area to the z. B. arranged in the vehicle evaluation device increases accordingly, especially when the vehicle is moving relatively quickly.

In one embodiment, the position-dependent environmental data stored in the server represent features in an area around the geographic position in question and / or probabilities of occurrence that can be assigned to the predetermined types of objects in this area.

As already explained above by way of example, features represented by the environmental data can e.g. Specify whether it is a rural area or an urban area, e.g. B. such a classification of an environment containing the geographical position can also be provided in a more refined manner. In the rural area, for. For example, a distinction can be made between “motorway or expressway area in rural areas” and “rural areas characterized by rural roads and / or smaller villages”. In particular, the position-dependent environmental data z. B. contain quantitative information on the respective extent of the presence of predetermined features.

In one embodiment, the features include the presence of predetermined types of structures, in particular buildings, in the area around the geographic position in question. In particular, these types can, for. B. each be defined by a specific use of the structure or building.

Structures, in particular buildings, are always erected for a specific purpose, and the presence of certain types of structures in the area around the relevant geographical position can have a major influence on the probability of occurrence of predetermined types of objects, in particular depending on the purpose of use. So is z. B. in the vicinity of playgrounds, kindergartens and elementary schools, the probability of occurrence of children increases accordingly. Similarly, z. B. in the vicinity of larger shopping centers, the probability of occurrence of people with (large) shopping bags, etc. increases.

The above examples also illustrate a dependency of the probabilities of occurrence of objects, which is often given in practice, as a function of time (e.g. time of year, day of the week and / or time of day). For kindergartens and elementary schools etc. there are e.g. B. seasonal fluctuations (because of vacation times) and weekday-dependent fluctuations (because of weekends and public holidays) and time-of-day-dependent fluctuations (because of typical start and end times).

Such dependencies can also advantageously be taken into account within the scope of the invention, for example by specifying the parameters to be specified for the object recognition algorithm as a function of time (in addition to the dependence on the geographical position of the vehicle). A “time component” of the surroundings of the vehicle can also advantageously be taken into account in this way. The parameters can be specified for the object recognition algorithm as a function of the time at which the relevant image was recorded (and at which the image evaluation for object recognition is accordingly preferably carried out (“in real time”)).

In the embodiment already explained, with the transmission of adaptation data from a server to an evaluation device performing the object recognition algorithm, this can advantageously, for. B. can be implemented in the context of the determination of the adaptation data carried out by the server based on the stored environmental data.

In one embodiment variant of this, the stored environmental data already contain at least one such time dependency of a feature and / or a probability of occurrence, see above that this time dependency is also reflected in the adjustment data and thus in the parameters to be specified. For this purpose, z. For example, it can be provided that the environment data stored in the server are continuously updated (so that the adjustment data are time-dependent).

In one embodiment, the features include the presence of predetermined types of temporary events, in particular events, in the area around the geographic position in question, in particular these types each being defined by a specific purpose of the event.

In this way, additional, in particular one-off and / or relatively short-term influences on the probabilities of occurrence of objects can be taken into account during object detection (e.g. temporary construction measures), particularly regardless of the presence of predetermined types of structures or buildings in the area around the relevant geographical position , sporting, cultural or political events, traffic accidents, etc.).

In one embodiment, the probabilities of occurrence to be assigned to the predetermined types of objects in the area around the geographic position in question were determined empirically.

This empirical determination can in particular by z. B. systematically planned and performed test drives take place with a vehicle, it being possible to determine respective position-dependent and / or time-dependent occurrence probabilities for a plurality of predetermined types of objects.

Coming back again to the possibility already explained of storing position-dependent environmental data in a server. B. be provided, the relevant (expediently relatively large) geographical area by means of an (imaginary) grid, in particular z. B. regular grid, e.g. B. a square grid or a hexagonal grid, to be divided into a plurality of individual grid areas (z. B. squares or hexagons) and to provide constant environmental data within each grid area. The determination of the adjustment data for a specific vehicle can then be done, for. B. be done in such a way that those environmental data are used that are stored for the grid area within which the geographical position of the vehicle is located. As an alternative to an exclusive use of this environmental data, however, z. B. be provided to perform an interpolation of the environment data to be used and / or the adaptation data to be transmitted when determining the adaptation data, for which the environment data z. B. of several grid areas (including the grid area in which the vehicle is located) can be used. In the case of such an interpolation (e.g. to implement a weighted interpolation), in particular e.g. For example, the distances between the geographical position of the vehicle and the neighboring grid areas in the relevant grid are taken into account (the smaller this distance, the greater the “weight” of the data stored or determined for the relevant grid area).

• - einen ersten Algorithmusteil, der dazu ausgebildet ist, anhand des Bildes eine vorläufige Erkennung der Objekte in der Umgebung des Fahrzeuges durchzuführen,
• - einen zweiten Algorithmusteil, der dazu ausgebildet ist, basierend auf den vorgegebenen Parametern das Ergebnis der vorläufigen Erkennung der Objekte in der Umgebung des Fahrzeuges zu korrigieren.

In one embodiment, the object recognition algorithm has:

• - A first algorithm part which is designed to carry out a preliminary detection of the objects in the vicinity of the vehicle on the basis of the image,
• a second algorithm part which is designed to correct the result of the preliminary detection of the objects in the vicinity of the vehicle based on the specified parameters.

The preliminary detection of the objects carried out by the first part of the algorithm can e.g. B. specify each “provisionally recognized” object by a quantitative “object probability”, which indicates a degree of certainty (correctness) of the recognition of the object in question. B. for each provisionally recognized object this quantitative information is corrected based on the probability of occurrence assigned to the relevant type of object (e.g. by multiplication with a corresponding correction factor). Then z. B. when a predetermined threshold value is exceeded by the corrected quantitative extent, the respective preliminary object recognition is confirmed (or revoked if this threshold value is undershot).

In a further development of this embodiment, the first algorithm part has an KI module that works on the basis of KI parameters determined by training.

The AI module can in particular, for. B. an artificial neural network with a plurality of neurons and connections between the neurons, the latter connections are each specified by weighting factors (or more general transfer functions), which in turn are specified by the KI parameters.

In this embodiment, for. B. be provided that the first part of the algorithm, in particular if this has an KI module or is formed from such an KI module, works independently of the specified parameters and thus only the second part of the algorithm works based on the specified parameters. In this case, AI is used advantageously, but disadvantageously cannot use the information inherent in the parameters provided within the scope of the invention. This use is only made by the second part of the algorithm by way of a correction of the preliminary object recognition.

In the following, an embodiment is described which overcomes this potential disadvantage and uses AI more extensively, so to speak.

In such an embodiment, the object recognition algorithm has an KI module that works based on KI parameters determined by training, these KI parameters being specified as a component of the parameters to be specified for the object recognition algorithm as a function of the geographical position of the vehicle by being specified as a function of the geographical position of the vehicle, these KI parameters are selected as an KI parameter set assigned to the geographical position of the vehicle from a large number of KI parameter sets which have been empirically determined in advance and stored as position-dependent KI parameter sets for a specific geographical area.

In a further development of this embodiment, the KI parameters are additionally specified as a function of the time (when the image was taken) by using the KI parameters as a function of the geographical position of the vehicle and the time as a function of the geographical position of the vehicle for this time assigned KI parameter set is selected from the KI parameter sets which have been empirically determined in advance and saved as position and time-dependent KI parameter sets.

In this embodiment, it cannot be ruled out that an object recognition implemented with the AI module, as in the embodiment described above, is subsequently corrected based on the specified parameters. However, this is not mandatory, since the KI parameters specified for the KI module as part of the parameters to be specified are already dependent on the geographical position of the vehicle.

The required position-dependent KI parameter sets can be used empirically, in particular z. B. be determined by test drives with a vehicle. The decisive factor here is that the training of the AI module does not use training data that was obtained from different or different geographical positions of the vehicle (test vehicle), but rather for each geographical position (or e.g. for a grid area of the above-mentioned type) a special / separate training is carried out and thus training data obtained for a specific geographical environment (e.g. grid area) are used to determine corresponding AI parameters (assigned to the geographical environment).

If this embodiment is used in combination with the already described use of a server in which position-dependent environment data are stored for a specific geographical area, it is expedient for these environment data to contain the corresponding KI parameters as such. The determination of the adaptation data to be transmitted to a specific vehicle consists in this case in selecting the corresponding AI parameters based on the geographical position of this vehicle and transmitting them as adaptation data to the evaluation device (e.g. located in the vehicle).

• - eine Vielzahl von Fahrzeugen, die jeweils aufweisen:
  • - eine Bildaufnahmeeinrichtung (z. B. wie oben bereits beschrieben) zum Aufnehmen eines Bildes einer Umgebung des Fahrzeuges,
  • - eine Bildauswerteeinrichtung, die dazu ausgebildet ist, das Bild unter Verwendung eines basierend auf vorgegebenen Parametern arbeitenden Objekterkennungsalgorithmus auszuwerten, um Objekte in der Umgebung des Fahrzeuges zu erkennen, und ferner dazu ausgebildet ist, die Parameter für den Objekterkennungsalgorithmus basierend auf Anpassungsdaten anzupassen,
  • - einen Server (z. B. wie oben bereits beschrieben), der dazu ausgebildet ist, für ein bestimmtes geografisches Gebiet positionsabhängige Umgebungsdaten zu speichern und für geografische Positionen innerhalb des geografischen Gebiets jeweilige Anpassungsdaten zu ermitteln,
  • - optional (im Falle eines außerhalb der Fahrzeug befindlichen Servers) ein Kommunikationssystem, das dazu ausgebildet ist, die von dem Server für die jeweiligen geografischen Positionen der Fahrzeuge jeweils ermittelten Anpassungsdaten von dem Server an die Bildauswerteeinrichtungen der jeweiligen Fahrzeuge zu übermitteln. Denkbar ist jedoch auch, dass jedes der Fahrzeuge selbst mit einem derartigen Server ausgestattet ist (z. B. wie oben bereits beschrieben, z. B. beinhaltend einen lokalen Speicher für Umgebungsinformationen zwecks Bereitstellung der Anpassungsdaten), so dass allenfalls eine lediglich lokal (d.h. an Bord der einzelnen Fahrzeuge bzw. innerhalb einer dort jeweils vorgesehenen Rechnereinrichtung bzw. Auswerteeinrichtung) stattfindende Kommunikation der Anpassungsdaten erforderlich ist.

According to a further aspect of the invention, a system for evaluating respective images of respective surroundings of a large number of vehicles is proposed in order to use the respective images to identify respective objects in the respective surroundings of the vehicles, this system having:

• - a variety of vehicles, each featuring:
  • - an image recording device (e.g. as already described above) for recording an image of the surroundings of the vehicle,
  • an image evaluation device which is designed to evaluate the image using an object recognition algorithm that works on the basis of predetermined parameters in order to recognize objects in the vicinity of the vehicle, and is also designed to adapt the parameters for the object recognition algorithm based on adaptation data,
  • a server (e.g. as already described above) which is designed to store position-dependent environmental data for a specific geographical area and to determine respective adaptation data for geographical positions within the geographical area,
  • - optionally (in the case of a server located outside the vehicle) a communication system which is designed to receive the information determined by the server for the respective geographical positions of the vehicles To transmit adaptation data from the server to the image evaluation devices of the respective vehicles. However, it is also conceivable that each of the vehicles is equipped with such a server itself (e.g. as already described above, e.g. containing a local memory for environmental information for the purpose of providing the adaptation data), so that at most a local (ie on board the individual vehicles or within a computer device or evaluation device provided there in each case, communication of the adaptation data is required.

• 1 ein Fahrzeug und einige beispielhafte Objekte in dessen Umgebung,
• 2 ein Blockdiagramm zur Veranschaulichung eines Verfahrens zur Bildauswertung, um Objekte in der Umgebung zu erkennen,
• 3 beispielhafte zu erkennende Arten von Objekten,
• 4 ein Blockdiagramm eines Systems zur Durchführung des Verfahrens von 2 gemäß eines Ausführungsbeispiels,
• 5 ein Blockdiagramm eines Systems zur Durchführung eines Verfahrens zur Bildauswertung gemäß eines weiteren Ausführungsbeispiels,
• 6 ein Blockdiagramm eines Systems zur Durchführung eines Verfahrens zur Bildauswertung gemäß eines weiteren Ausführungsbeispiels,
• 7 ein Blockdiagramm eines für eine Bildauswertung verwendbaren Objekterkennungsalgorithmus gemäß eines Ausführungsbeispiels, und
• 8 ein Blockdiagramm eines Objekterkennungsalgorithmus gemäß eines weiteren Ausführungsbeispiels.

The invention is described further below on the basis of exemplary embodiments with reference to the accompanying drawings. They represent:

• 1 a vehicle and some exemplary objects in its vicinity,
• 2 a block diagram to illustrate a method for image evaluation in order to recognize objects in the environment,
• 3rd exemplary types of objects to be recognized,
• 4th FIG. 3 is a block diagram of a system for performing the method of FIG 2 according to an embodiment,
• 5 a block diagram of a system for performing a method for image evaluation according to a further embodiment,
• 6th a block diagram of a system for performing a method for image evaluation according to a further embodiment,
• 7th a block diagram of an object recognition algorithm that can be used for image evaluation according to an exemplary embodiment, and
• 8th a block diagram of an object recognition algorithm according to a further embodiment.

1 shows a vehicle 10 and an environment 12th of the vehicle 10 , here as an example one in front of a vehicle driving on a street 10 current environment 12th . The figure shows an example of a surrounding scenario with various objects 20th illustrated.

The reference numbers of components provided several times, but having an analogous effect, such as e.g. B. the objects 20th , are numbered consecutively (each supplemented by a hyphen and a consecutive number). In the following, reference is also made to individual such components or to the entirety of such components by the non-supplemented reference number.

In 1 can be recognized as objects 20th for example a building 20-1 , a road marking 20-2 , a traffic sign 20-3 , an oncoming car 20-4 , a first taller person (adult) 20-5 and a second smaller person (child) 20-6 .

Regardless of whether it is the vehicle 10 to a conventional, ie by a user (driver) controlled vehicle, or z. B. is an autonomous vehicle in which control processes are partially or fully automated, it is advantageous if an image of the environment 12th of the vehicle 10 recorded and evaluated using an object recognition algorithm to use the image to objects such as here z. B. the objects 20-1 to 20-6 in the neighborhood 12th to recognize.

In the example of 1 assume that the vehicle 10 is equipped with an image recording device (e.g. front camera) (not shown in the figure) in order to enable the vehicle to be operated 10 continuously such images of the environment 12th in front of the vehicle 10 to record.

It is also assumed that the vehicle 10 is equipped with an electronic program-controlled evaluation device (e.g. control device for image evaluation) to which the images recorded by the image recording device (e.g. front camera) are fed in the form of digital image data and which are based on predetermined parameters (algorithm parameters) working object recognition algorithm running in the evaluation device carries out an evaluation of the images in order to use the images to identify the objects 20th in the neighborhood 12th of the vehicle 10 to recognize.

2 FIG. 13 is a block diagram of one in the example of FIG 1 evaluation algorithm used 30th including the object recognition algorithm 32 , by means of which an image 34 corresponding image data PIC are processed based on the predetermined parameters PAR in order to accomplish the object recognition.

The object recognition algorithm delivers the result of the object recognition 32 corresponding object data OBJ, which in particular z. B. a list of the object recognition algorithm 32 detected objects 20th with respective specifications such as B. type of object, relative direction and / or relative distance of the respective object 20th regarding the vehicle 10 can contain.

If the vehicle 10 is conventionally controlled, the object data OBJ z. B. for comfort and / or safety functions such. B. an automated output of warnings to the user (driver) of the vehicle 10 and / or for automated interventions in the control of the vehicle 10 (e.g. automatic braking and / or forced steering etc.) can be used.

If the vehicle 10 is designed with the capability of autonomous driving, the provision of such object data OBJ in autonomous vehicle operation is essential in order to be able to carry out the necessary control processes such as steering, accelerating, decelerating, etc. based on the object data OBJ.

In a departure from the example described up to now, the could be used to carry out the object recognition algorithm 32 computing capacity or infrastructure (hardware) used instead of just on board the vehicle 10 (in-vehicle evaluation device) can also be implemented in a “distributed” manner, for example partially arranged in the vehicle (in-vehicle evaluation device) and partially in a stationary data center (including, for example, “cloud” infrastructure), on whose computing capacity the vehicle 10 or the vehicle's own evaluation device via a mobile radio connection or a mobile Internet connection of the vehicle 10 can fall back on.

A special feature of the in 2 The image evaluation method shown is that the (the object recognition algorithm 32 specified) parameter PAR as a function of a (current) geographical position POS (z. B. geographical coordinates) of the vehicle 10 can be specified.

In 2 is this peculiarity through a default means 40 symbolizes which based on the geographical position POS of the vehicle 10 the parameters PAR for the object recognition algorithm 32 generated. This allows when evaluating the image 34 by means of the object recognition algorithm 32 Advantageously, additional information, namely information which can be derived from the geographical position POS, is used in order to improve the quality of the object recognition.

The default means 40 can in particular z. B. comprise a server of the type already described above (z. B. formed inside and / or outside of the vehicle 10 ) to based on the vehicle's geographic position POS 10 first of all to determine relevant adaptation data, on which the parameters PAR to be specified for the object recognition algorithm are then based 32 be generated, be it z. B. also by the server, or by a device specially provided for this purpose (e.g. in the vehicle 10 , e.g. B. as part of a computer device arranged there or the evaluation device). In the context of the invention, it is also possible that the adaptation data already contain the parameters PAR, or there is even an identity between adaptation data and parameters PAR to be specified.

The object recognition algorithm 32 is in the example of 2 designed to create predetermined types of objects 20th assign respective probabilities of occurrence and for the detection of objects 20th to be taken into account, these probabilities of occurrence being determined by the specified parameters PAR.

3rd shows representative of such types of objects 20th in the 1 objects already shown 20-1 to 20-6 , where in 3rd however, to be understood only as examples, the following additional objects 20th or the corresponding object types are illustrated: Bicycle 20-7 , Scooter 20-8 , Scooter including person 20-9 , tall person (adult) including small person (child) 20-10 , Person with luggage (e.g. suitcase or bags or shopping bags etc.) 20-11 , and ball (e.g. soccer) 20-12 .

Each of the objects 20-1 to 20-12 can, for example, define a corresponding concrete “type of object” that is used when the object recognition algorithm is carried out 32 can be used. Alternatively or additionally, types of objects can also e.g. B. as generalizations and / or concretizations of such objects 20th such as B. the objects shown 20-1 to 20-12 defined and when performing the object recognition algorithm 32 must be taken into account.

For example, in generalizing the object 20-1 (Buildings) Object types such as “stationary building” or “stationary building outside the roadway area” can be defined. In generalizing the object 20-2 (Lane marking) could e.g. B. an object type "stationary traffic guidance sign" can be defined, under which then z. B. the traffic sign 20-3 fell. In generalizing the object 20-4 (oncoming car) could e.g. B. Object types such. B. "oncoming two-lane vehicle", "two-lane vehicle", "vehicle" or "mobile object" can be defined, etc.

For example, in the specification of the object 20-1 (Building) Object types such as “school”, “kindergarten” or “shopping center” are defined become. In the specification of the object 20-2 (Lane marking) could e.g. B. Object types “lane markings”, “arrow markings” and “restricted area markings” can be defined. In the specification of the object 20-3 (Traffic sign) could e.g. B. Object types such. B. "Traffic sign relating to moving traffic", "Traffic sign relating to stationary traffic" or "Traffic sign on vertical post" can be defined, etc.

Alternatively or additionally, z. For example, object types defined by combinations of individual objects (or individual object types) can also be used, such as for the in 3rd objects shown 20-9 (Scooter including person), 20-10 (tall person including small person, alternatively e.g. "group of people") or 20-11 (Person with luggage), etc.

In the following description of further exemplary embodiments, the same reference symbols are used for components with the same effect. Essentially, only the differences from the exemplary embodiment (s) already described will be discussed, and reference is hereby expressly made to the description of previous exemplary embodiments.

4th Figure 3 shows an embodiment of a system 100 for performing an image evaluation method with an object recognition algorithm of the type described here.

The system 100 includes, for example, one or more vehicles, here z. B. at least that in 1 shown vehicle 10 or one on board the vehicle 10 trained image evaluation device 50 with an image recording device connected to it for recording the image of the surroundings 12th of the vehicle10, here z. B. a camera 60 (e.g. video camera).

The image evaluation device 50 is designed to generate the image using the object recognition algorithm operating on the basis of the specified parameters PAR 32 evaluate to objects 20th in the neighborhood 12th of the vehicle 10 to recognize, and also designed to set the parameters PAR for the object recognition algorithm 32 adjust based on adjustment data ADJ.

In this example, they are completely on board the vehicle 10 trained image evaluation device 50 (which alternatively, as already mentioned, could also be implemented in a distributed manner) includes an object recognition algorithm for this purpose 32 assigned parameter adjustment algorithm (or a corresponding hardware component within the image evaluation device 50 ) 36 , by means of which, based on the supplied adjustment data ADJ, the object recognition algorithm 32 specified parameters PAR can be generated (or adapted / modified as required).

The system 100 further comprises a server 42 , the one from the image evaluation device 50 of the vehicle 10 provides adjustment data ADJ to be received.

The server 42 is in the example shown as a stationary component of the system 100 For example, implemented as a central data center (or a cloud-based data center) and designed to store position-dependent environmental data ENV in a database DB for a specific geographical area (e.g. Bavaria, Germany, Europe, etc.) and for all geographical positions POS to determine respective adjustment data ADJ within the geographical area.

The system 100 further comprises a communication system such. B. a cellular network or the like, which is designed to be used by the server 42 adjustment data ADJ determined by the server for a respective instantaneous geographical position POS of the vehicle 42 to the image evaluation device 50 of the vehicle 10 to submit.

It goes without saying that the system 100 in practice it can also be used to display respective images of respective surroundings of a variety of vehicles (including vehicle 10 ) and to use the respective images to identify respective objects in the respective surroundings of these vehicles. In such an embodiment, the server 42 used to determine respective individual adjustment data for the (multitude of) geographical positions of all vehicles, which are then transmitted individually via the communication system, ie to the respective image evaluation devices of the respective vehicles.

The server 42 determines the respective adjustment data ADJ for the geographical positions POS within the geographical area based on the respective geographical position POS and based on the data in the server 42 stored position-dependent environmental data ENV.

• - Durchführen eines basierend auf vorgegebenen Parametern PAR arbeitenden Objekterkennungsalgorithmus 32, um anhand des Bildes 34 Objekte 20 in der Umgebung 12 des Fahrzeuges 10 zu erkennen,
• - Übermitteln von Anpassungsdaten ADJ von einem Server 42 an eine den Objekterkennungsalgorithmus 32 durchführende (z. B. in dem Fahrzeug 10 angeordnete) Auswerteeinrichtung 50, wobei diese Anpassungsdaten ADJ von dem Server 42 basierend auf der geografischen Position POS des Fahrzeuges 10 (bzw. eines betreffenden Fahrzeuges) und basierend auf in dem Server 42 für ein bestimmtes geografisches Gebiet gespeicherten positionsabhängigen Umgebungsdaten ENV ermittelt werden,
• - Anpassen der Parameter PAR für den Objekterkennungsalgorithmus 32 basierend auf den von dem Server 42 an die Bildauswerteeinrichtung 50 übermittelten Anpassungsdaten ADJ.

That by means of the system 100 procedures carried out (see e.g. 2 ) to evaluate an image 34 an environment 12th of a vehicle 10 thus comprises the following steps:

• - Carrying out an object recognition algorithm that works on the basis of predetermined parameters PAR 32 to use the picture 34 Objects 20th in the neighborhood 12th of the vehicle 10 to recognize
• - Transmission of adjustment data ADJ from a server 42 to the object recognition algorithm 32 performing (e.g. in the vehicle 10 arranged) evaluation device 50 , this adjustment data ADJ from the server 42 based on the vehicle's geographical position POS 10 (or a relevant vehicle) and based on in the server 42 position-dependent environmental data ENV stored for a specific geographical area are determined,
• - Adjustment of the parameters PAR for the object recognition algorithm 32 based on the from the server 42 to the image evaluation device 50 transmitted adjustment data ADJ.

The ones in the server 42 In the example shown, stored position-dependent environmental data ENV represent features in an area around the relevant geographic position POS and / or the predetermined types of objects 20th Probabilities of occurrence to be assigned in this area.

The features can be in particular z. B. relate to the presence (or non-existence) of predetermined types of structures, in particular buildings, in the area around the relevant position POS, these types of structures each z. B. can be defined by a specific purpose of the building (e.g. use as a school, kindergarten, shopping center, museum, bicycle parking garage, etc.). Alternatively or additionally, the features can be, for. B. refer to the (current) presence or absence of predetermined types of temporary events, such as e.g. B. events or construction work in the area around the relevant position POS, these types each z. B. can be defined by a specific purpose of the event (e.g. construction of a building, renovation of a street, sporting event, demonstration parade, emergency response at an accident site, etc.).

In order to be able to take into account changes in such features and in particular also temporary events in the image evaluation method, the in the server 42 stored environmental data ENV through corresponding input data INP ( 4th ) continuously, e.g. B. updated at predetermined time intervals of preferably less than 1 hour each. With such an update, time-dependent (always correct) adaptation data ADJ are consequently also advantageously generated. The input data INP can e.g. B. by authorities (e.g. building office, surveying office, etc.) or other organizations (e.g. regulatory office, police, fire brigade, etc.).

• - eine Vielzahl von Fahrzeugen wie z. B. das dargestellte Fahrzeug 10, die jeweils aufweisen: eine Bildaufnahmeeinrichtung 60 zum Aufnehmen eines Bildes 34 einer Umgebung 12 des Fahrzeuges; eine Bildauswerteeinrichtung 50, die dazu ausgebildet ist, das durch Bilddaten PIC repräsentierte Bild unter Verwendung eines basierend auf vorgegebenen Parametern PAR arbeitenden Objekterkennungsalgorithmus 32 auszuwerten, um Objekte 20 in der Umgebung 12 des Fahrzeuges zu erkennen, und ferner dazu ausgebildet ist, die Parameter PAR für den Objekterkennungsalgorithmus basierend auf zugeführten Anpassungsdaten ADJ anzupassen; und eine Ortungseinrichtung GPS zum Erfassen der geografischen Position POS des Fahrzeuges, wie insbesondere eine Empfangseinrichtung zum Empfangen und Auswerten von Satellitensignalen zur Ortung des Fahrzeuges,
• - einen Server 42, der dazu ausgebildet ist, für ein bestimmtes geografisches Gebiet positionsabhängige Umgebungsdaten ENV zu speichern und für geografische Positionen innerhalb des geografischen Gebiets jeweilige Anpassungsdaten ADJ zu ermitteln,
• - ein Kommunikationssystem (z. B. Mobilfunknetz), das dazu verwendet wird, um die von dem Server 42 für die jeweiligen geografischen Positionen POS der Fahrzeuge jeweils ermittelten Anpassungsdaten ADJ an die betreffenden Bildauswerteeinrichtungen 50 der jeweiligen Fahrzeuge zu übermitteln, und um die von den Ortungseinrichtungen GPS jeweils erfassten geografischen Positionen POS der Fahrzeuge an den Server 42 zu übermitteln.

5 Figure 3 shows another embodiment of a system 100 for carrying out a method for image evaluation with an object recognition algorithm 32 , full:

• - a variety of vehicles such as B. the vehicle shown 10 each comprising: an image pickup device 60 to take a picture 34 an environment 12th of the vehicle; an image evaluation device 50 , which is designed to generate the image represented by image data PIC using an object recognition algorithm that operates on the basis of predetermined parameters PAR 32 evaluate to objects 20th in the neighborhood 12th of the vehicle, and is further designed to adapt the parameters PAR for the object recognition algorithm based on supplied adaptation data ADJ; and a location device GPS for detecting the geographical position POS of the vehicle, such as in particular a receiving device for receiving and evaluating satellite signals for locating the vehicle,
• - a server 42 which is designed to store position-dependent environment data ENV for a specific geographical area and to determine respective adaptation data ADJ for geographical positions within the geographical area,
• - A communication system (e.g. cellular network) that is used to receive the from the server 42 adaptation data ADJ determined in each case for the respective geographical positions POS of the vehicles to the relevant image evaluation devices 50 of the respective vehicles, and to transmit the geographic positions POS of the vehicles respectively recorded by the positioning devices GPS to the server 42 to submit.

• - eine Vielzahl von Fahrzeugen 10-1, 10-2, 10-3 ... (z. B. mehr als 1000 Fahrzeuge), die jeweils aufweisen: eine Bildaufnahmeeinrichtung zum Aufnehmen eines Bildes einer jeweiligen Umgebung des betreffenden Fahrzeuges; eine Bildauswerteeinrichtung, die dazu ausgebildet ist, das jeweilige Bild unter Verwendung eines basierend auf vorgegebenen Parametern arbeitenden Objekterkennungsalgorithmus auszuwerten, um Objekte in der jeweiligen Umgebung zu erkennen, und ferner dazu ausgebildet ist, die Parameter für den Objekterkennungsalgorithmus basierend auf jeweils zugeführten Anpassungsdaten ADJ₁, ADJ₂, ADJ₃ ... anzupassen; und (optional) z. B. eine Ortungseinrichtung zum Erfassen der geografischen Position des betreffenden Fahrzeuges,
• - einen Server 42, der dazu ausgebildet ist, für ein bestimmtes geografisches Gebiet positionsabhängige Umgebungsdaten ENV zu speichern und für geografische Positionen innerhalb des geografischen Gebiets jeweilige Anpassungsdaten zu ermitteln,
• - ein unter Nutzung eines Mobilfunknetzes arbeitendes Kommunikationssystem 70, das als ein Geocasting-System eingesetzt wird, um die von dem Server 42 für die jeweiligen geografischen Positionen POS₁, POS₂, POS₃ ... der Fahrzeuge 10-1, 10-2, 10-3 ... jeweils ermittelten Anpassungsdaten ADJ₁, ADJ₂, ADJ₃... an die betreffenden Bildauswerteeinrichtungen der jeweiligen Fahrzeuge 10-1, 10-2, 10-3 ... zu übermitteln, und gegebenenfalls auch die von den jeweiligen Ortungseinrichtungen erfassten geografischen Positionen POS₁, POS₂, POS₃ ... der Fahrzeuge 10-1, 10-2, 10-3 ... an den Server 42 zu übermitteln.

6th Figure 3 shows another embodiment of a system 100 for carrying out a method for image evaluation with an object recognition algorithm 32 , full:

• - a wide variety of vehicles 10-1 , 10-2 , 10-3 ... (e.g. more than 1000 vehicles), each having: an image recording device for recording an image of a respective environment of the vehicle in question; an image evaluation device which is designed to evaluate the respective image using an object recognition algorithm that works on the basis of predetermined parameters in order to recognize objects in the respective environment, and is also designed to evaluate the parameters for the object recognition algorithm based on the respectively supplied adaptation data ADJ ₁ , ADJ ₂ , ADJ ₃ ... adapt; and (optionally) e.g. B. a locating device to record the geographical position of the vehicle concerned,
• - a server 42 which is designed to store position-dependent environmental data ENV for a specific geographical area and to determine respective adaptation data for geographical positions within the geographical area,
• - a communication system that works using a cellular network 70 that is being used as a geocasting system to view those from the server 42 for the respective geographical positions POS ₁ , POS ₂ , POS ₃ ... of the vehicles 10-1 , 10-2 , 10-3 ... in each case determined adaptation data ADJ ₁ , ADJ ₂ , ADJ ₃ ... to the relevant image evaluation devices of the respective vehicles 10-1 , 10-2 , 10-3 ... to be transmitted, and possibly also the geographical positions POS ₁ , POS ₂ , POS ₃ ... of the vehicles recorded by the respective location devices 10-1 , 10-2 , 10-3 ... to the server 42 to submit.

For the design of the server 42 and the communication system 70 there are many options. To achieve this, z. B. a communication system such as "eHorizon" from Continental can be used. In such a communication system, current information on the state of road sections in a predetermined geographical area can be kept, with vehicles integrated in the system themselves acting as sensors to provide the required data (“big data”) and z. B. such data to a server ("backbone") of the z. B. cloud-based systems and at the same time receive corresponding individually relevant data with information about the route sections ahead of them from the server. Thus, such a communication system within the scope of the invention and z. B. the embodiment described above can be used in particular to transfer the determined adaptation data ADJ ₁ , ADJ ₂ , ADJ ₃ ... to the respective vehicles 10-1 , 10-2 , 10-3 ... to transmit.

In addition, in a system according to the invention it is also possible, using such a (“bidirectional”) communication system, as an alternative or in addition to the above-mentioned provision of the input data INP by authorities or other organizations, to provide the input data INP at least partially with the help of the vehicles 10-1 , 10-2 , 10-3 ... to be realized by z. B. in the example of 6th the vehicles 10-1 , 10-2 , 10-3 ... or their image evaluation devices 50 the respective object data OBJ and / or respective derived therefrom, e.g. B. the information content of the object data OBJ summarizing vehicle environment data via the communication system in question 70 to the server 42 transmitted, with the server 42 is designed to generate corresponding input data INP for improving and / or updating the environment data ENV stored in the database DB from the plurality of object data and / or vehicle environment data obtained in this way.

Such vehicle environment data transmitted to the server can in particular contain information about one or more probabilities of occurrence of the predetermined types of objects (at the respective positions of the respective vehicles) so that the server advantageously updates the position-dependent environment data stored in the server based on this information can.

7th shows an embodiment of an object recognition algorithm that can be used within the scope of the invention for image evaluation 32 , having a first algorithm part 32a , by means of which based on an image 34 (entered image data) a preliminary recognition of objects ( 20th ) in the neighborhood ( 12th ) of the vehicle concerned ( 10 ) and a second part of the algorithm 32b , by means of which, based on predetermined parameters PAR, the result of the preliminary detection is corrected and (corrected) object data OBJ are supplied as the end result.

The result of the preliminary detection can in particular contain preliminary object data, which in particular z. B. a list of the object recognition algorithm 32 objects recognized with a certain probability (e.g. above a predetermined threshold) 20th with respective specifications such as B. Type of object, relative direction and / or relative distance of the respective object ( 20th ) regarding the vehicle ( 10 ) together with this probability.

The correction can in this case, for. B. consist in correcting the individual probabilities in each case by means of a correction value (e.g. to multiply by a correction factor) in order to obtain corrected probabilities for each provisionally recognized object and to determine the final result (object data OBJ) based on this. The correction can, for. B. the probabilities of those objects are increased that belong to an object type for which a relatively high probability of occurrence is determined by the specified parameters PAR. Conversely, when correcting z. B. the probabilities of those objects that belong to an object type with a relatively low probability of occurrence are reduced.

In the example of 7th is the first part of the algorithm 32a designed as an AI module (e.g. artificial neural network) that operates based on AI parameters determined by training or has at least one such AI module.

In the context of the invention, an KI module can be designed in such a way that its KI parameters, which determine the way in which input data is converted, here e.g. B. the image data PIC, in output data, here z. B. in the object data OBJ, can be changed by self-learning and / or training in such a way that an improvement in the algorithm quality (with regard to the correctness of the output object data OBJ) can be achieved. In self-learning or training, experience data or training data consisting of (a large number of) matching input data and output data are used (that is, each containing "correct" output data) and, in the event of a discrepancy between the output data actually supplied for this input data by the KI module, and the "correct" output data varies the KI parameters in such a way that the deviation is reduced (in particular optimized).

8th shows a further embodiment of an object recognition algorithm that can be used within the scope of the invention for image evaluation 32 .

The object recognition algorithm 32 has an KI module operating based on KI parameters KI-PAR determined by training, these KI parameters KI-PAR as a component of the object recognition algorithm 32 The parameters to be specified are specified as a function of the geographical position POS of the vehicle in question, in that depending on the geographical position POS of the vehicle, these KI parameters KI-PAR as a KI parameter set KI-PAR assigned to the geographical position POS of the vehicle from a large number is selected from KI parameter sets KI-PAR ₁ , KI-PAR ₂ , ..., KI-PAR _N , which are empirically determined in advance and for a certain geographical area as position-dependent KI parameter sets KI-PAR ₁ , KI-PAR ₂ , ..., KI-PAR _N (e.g. in a server as already described).

In summary, the invention proposes a “situational weighting” of recognition probabilities within the object recognition algorithm in the context of an object recognition implemented by means of an image evaluation in the surroundings of a vehicle. In this way, the probability of correct object recognition can advantageously be increased. Essential for this is the use of additional input information (parameter PAR), which is formed as a function of the (current) geographical position of the vehicle and accordingly influences the operation of the object recognition algorithm used.

Claims (13)

Method for evaluating an image (34) of an environment (12) of a vehicle (10) using an object recognition algorithm (32) that operates on the basis of predetermined parameters (PAR) in order to identify objects (20) in the environment (12 ) of the vehicle (10), the parameters (PAR) being specified as a function of a geographical position (POS) of the vehicle (10). Procedure according to Claim 1 , the object recognition algorithm (32) being designed to assign respective occurrence probabilities to predetermined types of objects (20) and to take them into account for the detection of objects (20), these occurrence probabilities being determined by the predetermined parameters (PAR). Procedure according to Claim 1 or 2 , the method further comprising the following steps: - Transmission of adjustment data (ADJ) from a server (42) to an evaluation device performing the object recognition algorithm (32), the adjustment data (ADJ) from the server (42) based on the geographical position ( POS) of the vehicle (10) and based on position-dependent environmental data (ENV) stored in the server (42) for a specific geographical area, - adapting the parameters (PAR) for the object recognition algorithm (32) based on the data from the server ( 42) adaptation data (ADJ) transmitted to the evaluation device. Procedure according to Claim 3 , the method further comprising the following steps: - detecting the geographical position (POS) of the vehicle (10), - transmitting the detected geographical position (POS) of the vehicle (10) to the server (42). Procedure according to Claim 3 or 4th At least one component of the evaluation device is arranged in the vehicle (10) and the adaptation data (ADJ) are transmitted from the server (42) to this component of the evaluation device by geocasting the adaptation data (ADJ) to vehicles located within the geographical area . Method according to one of the Claims 3 to 5 , wherein the position-dependent environmental data (ENV) stored in the server (42) features in an area around the relevant geographic Position and / or the predetermined types of objects (20) in this area represent probabilities of occurrence to be assigned. Procedure according to Claim 6 , wherein the features include the presence of predetermined types of structures, in particular buildings, in the area around the geographic position in question, in particular these types being defined in each case by a specific purpose of use of the structure or building. Procedure according to Claim 6 or 7th , wherein the features include the presence of predetermined types of temporary events, in particular events, in the area around the geographic position in question, in particular these types each being defined by a specific purpose of the event. Method according to one of the Claims 6 to 8th wherein the probabilities of occurrence to be assigned to the predetermined types of objects (20) in the area around the geographic position in question have been determined empirically. Method according to one of the preceding claims, wherein the object recognition algorithm (32) comprises: - A first algorithm part (32a) which is designed to carry out a preliminary detection of the objects (20) in the surroundings (12) of the vehicle (10) on the basis of the image (34), - A second algorithm part (32b) which is designed to correct the result of the preliminary detection of the objects (20) in the vicinity (12) of the vehicle (10) based on the predetermined parameters (PAR). Procedure according to Claim 10 , wherein the first algorithm part (32a) has an KI module operating on the basis of KI parameters (KI-PAR) determined by training. Method according to one of the Claims 1 to 11 , wherein the object recognition algorithm (32) has an KI module operating on the basis of KI parameters (KI-PAR) determined by training, these KI parameters (KI-PAR) as part of the parameters (PAR ) can be specified as a function of the geographical position (POS) of the vehicle (10) by using these KI parameters (KI-PAR) as one of the geographical (POS) position as a function of the geographical position (POS) of the vehicle (10) of the vehicle (10) assigned KI parameter set is selected from a large number of KI parameter sets (KI-PAR ₁ , KI-PAR ₂ , KI-PAR ₃ ...), which are empirically determined in advance and as position-dependent for a specific geographical area KI parameter sets (KI-PAR ₁ , KI-PAR ₂ , KI-PAR ₃ ...) have been saved. System for evaluating respective images of respective surroundings of a large number of vehicles (10-1, 10-2, 10-3 ...) in order to use the respective images to identify respective objects (20) in the respective surroundings (12) of the vehicles (10 -1, 10-2, 10-3 ...), comprising: a multiplicity of vehicles (10-1, 10-2, 10-3 ...), each of which has: an image recording device for recording an image (34) of an environment (12) of the vehicle (10-1, 10-2, 10-3 ...), - an evaluation device (50) which is designed to generate the image (34) using a based to evaluate the object recognition algorithm (32) operating on predetermined parameters (PAR) in order to recognize objects (20) in the surroundings (12) of the vehicle (10-1, 10-2, 10-3 ...), and is also designed to do so to adapt the parameters (PAR) for the object recognition algorithm (32) based on adaptation data (ADJ), - a server (42) which is designed to be position-dependent for a specific geographical area e to store environmental data (ENV) and to determine respective adaptation data (ADJ) for geographical positions within the geographical area, - a communication system which is designed to receive the data from the server (42) for the respective geographical positions (POS ₁ , POS ₂ , POS ₃ ...) of the vehicles (10-1, 10-2, 10-3 ...) each determined adaptation data (ADJ ₁ , ADJ ₂ , ADJ ₃ ...) from the server (42) to the image evaluation devices (50) of the respective vehicles (10-1, 10-2, 10-3 ...).

Images