Training of a classifier

#The present application claims priority to and the benefit of German patent application no. 10 2018 207 923.4, which was filed in Germany on May 18, 2018, the disclosure which is incorporated herein by reference.

#The present invention relates generally to the field of artificial intelligence. In particular, the present invention relates to a method for training an artificial intelligence module, AI module, and a device for detecting an object and/or a situation.

#There are numerous technical applications in which objects and/or situations are to be detected automatically, that is to say, in particular, in a manner that is automated by a data processing device. This frequently occurs by way of feature detection or pattern detection, which is of great importance in particular in the field of artificial intelligence. Exemplary applications are found e.g. in the keyword-like search for images in Internet search engines, in the use of robots or also in automotive technology, where computer-assisted driver assistance systems or autonomously driving vehicles require reliable object detection so as to be able to move the vehicle safely through traffic.

#FIG. 1 a device for detecting an object according to an exemplary embodiment of the invention,

#FIG. 2 a block diagram of a device for detecting an object according to an exemplary embodiment of the invention,

#FIG. 3 a block diagram of a device for detecting an object according to an another exemplary embodiment of the invention,

#FIG. 1 shows a device 100 for detecting an object 200 according to one exemplary embodiment, device 100 here being integrated into a vehicle 300 by way of example. In exemplary fashion, vehicle 300 is here developed as a motor vehicle in the form of a passenger car.

#Device 100 has a data processing device 110 in the form of an artificial intelligence module having a processor 111 and a memory 112 for storing a program code to be processed by processor 111, the program code comprising at least one first classifier 113. Device 100 furthermore comprises a first image recording device 120 in the form of a camera, which is situated in a first position for recording object 200 from a first perspective 121 and is coupled to data processing device 110. The first position is here by way of example approximately at the height of a headlight of motor vehicle 300 so that first perspective 121 records object 200 frontally in relation to motor vehicle 300. It should be noted that first classifier 113 was trained from first perspective 121, that is, it is already set up to detect object 200 from this first perspective 121. Classifier 113 is therefore able to classify object 200, that is, to assign it to a particular class of objects.

#Furthermore, device 100 also has a second image recording device 130 that is situated in a second position for recording the same object 200 from a second perspective 131 differing from first perspective 121 and that is coupled to data processing device 110. In this exemplary embodiment, the second position is located in the area of a vehicle roof of motor vehicle 300, that is, in the direction of a vertical axis of the vehicle, above the first position of first image recording device 120. Accordingly, second image recording device 130 records object 200 from a different height as well as at a different angle, which in this exemplary embodiment corresponds to a recording at an angle from above. By contrast, only a comparatively small section of object 200 is recorded frontally so that the second image recording device is here by way of example not able to provide an image of object 200 that would suffice for a reliable detection of object 200. As indicated in FIG. 1, however, first perspective 121 and second perspective 131 have a common overlapping region 140, which is situated here in exemplary fashion at a front of object 200 relative to motor vehicle 300. With reference to FIG. 2, which shows device 100 as a block diagram, the functional interaction of first image recording device 120 and second image recording device 130 by using first classifier 113 is now functionally explained.