EP4118620A1 - Procédé et dispositif de traitement d'images - Google Patents

Procédé et dispositif de traitement d'images

Info

Publication number
EP4118620A1
EP4118620A1 EP21709970.4A EP21709970A EP4118620A1 EP 4118620 A1 EP4118620 A1 EP 4118620A1 EP 21709970 A EP21709970 A EP 21709970A EP 4118620 A1 EP4118620 A1 EP 4118620A1
Authority
EP
European Patent Office
Prior art keywords
image
descriptor
preferred embodiments
further preferred
distance
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
EP21709970.4A
Other languages
German (de)
English (en)
Inventor
Stephan Simon
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Robert Bosch GmbH
Original Assignee
Robert Bosch GmbH
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Robert Bosch GmbH filed Critical Robert Bosch GmbH
Publication of EP4118620A1 publication Critical patent/EP4118620A1/fr
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
    • G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
    • G06V10/757—Matching configurations of points or features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30248—Vehicle exterior or interior
    • G06T2207/30252—Vehicle exterior; Vicinity of vehicle

Definitions

  • the disclosure relates to a method, in particular a computer-implemented one, for processing, in particular digital, images.
  • the disclosure also relates to a device for processing, in particular digital, images.
  • Preferred embodiments relate to a method, in particular a computer-implemented method, for processing, in particular digital, images, comprising the following steps: providing a first image and a second image, transforming the first image into a first descriptor image and the second image into a second Descriptor image, determining a distance image based on a comparison of the first descriptor image with the second descriptor image, and, optionally, forming a detection image based on the distance image.
  • This enables, for example, a particularly efficient detection of changes in relation to the images, e.g. detection of a movement of at least one object depicted on at least one of the images.
  • an image can be understood to be an encoded data record which describes or represents a representation of the image or an image representation.
  • a control rule for a display device can be determined by means of a computer from the data record representing the image in such a way that the display device can use the Control rule represents the coded image.
  • the display device can be, for example, a display unit or a projector unit.
  • the display device can, for example, be arranged on or in a driver's cab of a vehicle.
  • a transformation is used for the transformation into the descriptor images, which transforms the environment of a picture element ("pixel") of the first or second image in the respective image into a descriptor that describes this environment, preferably in a compact manner, that is to say with a few bits, in particular with fewer bits than corresponds to the information content of the area around the pixel.
  • the value of the descriptor is referred to as the signature.
  • the signature has a fixed length, in particular word length, of e.g. B. 8 bit to e.g. 32 bit, but in further preferred embodiments it can also be longer than 32 bit or shorter than 8 bit.
  • the transforming includes carrying out the transformation for a respective environment of a plurality of pixels, for example each pixel in the (first or second) image, so that as a result a respective (first or second) "image of descriptors" , i.e. the descriptor image or images already mentioned, is created.
  • a descriptor image can also be understood as a plurality of descriptor values or signatures, which are preferably organized in a matrix-like arrangement of rows and columns, corresponding to the position of the pixels evaluated for their formation or the respective surroundings of a pixel under consideration.
  • a descriptor image is e.g. comparable to the first or second image, but instead of the regular image information (such as brightness or intensity values of one or more gray scale or color channels), each image element of the descriptor image is assigned the respective signature.
  • pixels at the image edge of the first and / or second image can be disregarded for the transformation to the descriptor images, because there, for example, the frame for the transformation protrudes beyond the image, and the "pixel values" are therefore undefined.
  • pixels at the edge of the image of the first and / or second image are taken into account for the transformation to the descriptor images, with possibly missing or undefined pixel values at the edge being supplemented according to a predefinable rule, e.g. by copying the pixel values of existing pixels and / or setting the previously undefined values to a predefinable value.
  • At least one of the methods listed below can be used for the transformation: SIFT (scale-invariant feature transformation), SURF (Speeded Up Robust Features, for example according to Bay H., Tuytelaars T., Van Gool L. (2006) SURF: Speeded Up Robust Features. In: Leonardis A., Bischof H., Pinz A. (eds) Computer Vision - ECCV 2006. ECCV 2006. Lecture Notes in Computer Science, vol 3951. Springer, Berlin, Heidelberg, https: // doi. org / 10.1007 / 11744023_32), ORB (E. Rublee, V. Rabaud, K. Konolige and G.
  • a distance measure is available or can be specified for the descriptor selected for forming the descriptor images, which, for example, allows a difference between two descriptor values to be determined or assessed.
  • a similarity measure can also be used as an alternative to the distance measure, it being possible in particular for both measures to be convertible into one another. Therefore, in the following, only the distance measure is considered by way of example and without restricting the general validity, with all steps and embodiments described by way of example also correspondingly when using a Similarity measure - instead of a distance measure - can be used to evaluate a difference between the descriptor images, that is to say, for example, to characterize a result of the comparison of the first descriptor image with the second descriptor image.
  • a descriptor is used for the transformation, for which a distance measure and / or a similarity measure is definable and / or defined, in particular the comparison of the first descriptor image with the second descriptor image based on the distance measure and / or the similarity measure is carried out.
  • the Hamming distance or a distance measure based on the Hamming distance is used as the distance measure for the comparison, the Hamming distance in particular being compared with a predeterminable threshold value and, based on the comparison, a particularly binary one , Comparison value is determined.
  • a Hamming distance of approximately N / 2 can be expected on a statistical average. For a change detection in accordance with further preferred embodiments, this means that distance values close to 0 are to be expected in the case of essentially matching image areas - and for image areas that do not match, e.g. B. due to moving objects, distance values significantly greater than 0, and then z. B. be around N / 2.
  • the distance measure can optionally be further simplified, for example by binarization. For example, the Hamming distance can be compared to a threshold L and a binary one
  • the descriptor values can also be determined if necessary, e.g. "on the fly".
  • the transformation of the first image into the first descriptor image and / or the transformation of the second image into the second descriptor image can also be dynamic, that is to say, for example, as required and / or in real time, in particular, for example, also in direct temporal connection with the Determining the distance image, are carried out.
  • the transformation of the first image into the first descriptor image and / or the transformation of the second image into the second descriptor image can also be parallelized at least temporarily, e.g. if several computing cores are available to carry out the respective transformation (s).
  • the first descriptor image belonging to a first point in time is compared with a second descriptor image which belongs to a second, in particular earlier, point in time.
  • both the transformation to the descriptor images and the determination of the distance image can each be carried out, for example, only on a predeterminable sub-area of the first and second images or the descriptor images that can be derived therefrom.
  • subareas can also be selected dynamically, that is to say at runtime of the method or a device executing the method, for example based on a current content of the images and / or previously recognized changes and / or objects.
  • the comparison of the first descriptor image with the second descriptor image takes place pixel by pixel, so one picture element of the first descriptor image is compared with a corresponding picture element of the second descriptor image and a corresponding value for the distance measure ("Distance value") obtained.
  • a descriptor value at a specific coordinate of the first descriptor image is compared with the descriptor value at the corresponding (same) coordinate of the second descriptor image.
  • the distance value determined in this way is entered in a distance image or in the distance image at the corresponding coordinate.
  • the distance image has the same size (number of pixels, e.g. characterizable by width and height) as the descriptor images to be compared.
  • the distance image i.e. each picture element
  • the distance image can have values in the range 0 to N, for example, which characterize the comparison result between the first descriptor image and the second descriptor image.
  • the method further comprises: filtering the distance image, as a result of which a filtered distance image is obtained, in particular the formation of the detection image taking place based on the filtered distance image.
  • the filtering is carried out in such a way that the distance image is converted into a filtered image which changes in indicates in a compact form what is useful, for example, for a functional interface according to further preferred embodiments.
  • the detection image has one or more contiguous regions, which are also referred to as "blobs" according to further preferred embodiments, for those image areas in which, in particular significant, changes between the (first and second) images or the herewith corresponding descriptor images are available.
  • an (at least partial) determination of the distance image and / or the detection image can also take place if necessary, for example "on the fly" .
  • the detection image can be converted into another form in an optional further or alternative step, e.g. B. to be able to transmit it more efficiently via an interface.
  • the contours of at least one blob are approximated, e.g. described as polygons.
  • the detection image can also be compressed, e.g. with a run length coding or another entropy coding, e.g. with a common coding for segment images.
  • the filtering includes the application of a majority filter and / or a threshold value filter.
  • the method further comprises: further processing of the detection image, in particular formation of output information, based on at least one of the following elements: a) detection image, b) first image, c) second image.
  • the output information has at least one of the following elements: a) acoustic signal, b) haptic signal, c) optical signal, d) image, in particular digital image, with at least one graphically highlighted image area, in particular one based on the detection image graphically highlighted image area.
  • the method further comprises: assigning an evaluation measure to at least one descriptor of the first descriptor image and / or the second descriptor image, and, optionally, taking the evaluation measure into account when determining the distance image, wherein in particular the evaluation measure is noise or characterizes a signal-to-noise ratio, in particular a region of the first image and / or of the second image associated with the respective descriptor.
  • the noise can thus also be taken into account when forming the descriptors or the descriptor images, for example according to the method described in DE 102017212 339.
  • each, formed descriptor (s) can be given an evaluation that is dependent on the strength of the noise or the signal-to-noise ratio (SNR): in some preferred embodiments, for example, in the form of a binary one Additional information (e.g. coded as an additional bit), which in further preferred embodiments can also be interpreted as the suitability of the descriptor for further processing.
  • SNR signal-to-noise ratio
  • the additional information allows, for example, to identify suitable descriptors less well (e.g. for subsequent further processing) due to the noise, e.g. B. as "unsuitable”, e.g. to treat them differently in the event of a change detection, than descriptors that are better suited for further processing, e.g. due to lower noise or higher SNR.
  • this additional information can be taken into account when determining the distance image ("distance calculation"), for example according to the following rule: If at least one of the descriptors to be compared is marked as "unsuitable”, the distance is not calculated according to the usual rule, but one Another rule is applied, according to which, for example, the actual distance is replaced by a specifiable, in particular fixed, value. In further preferred embodiments, in the example of the Hamming distance as a distance measure, the value can then be set to "0", for example, which means, for example, that image regions dominated by noise are treated as if they were motionless.
  • the Hamming distance can also be set to a different value, e.g. B. "1" or "2", so that, for example, areas excluded due to noise are not treated differently (e.g. better placed) than non-excluded unmoved areas.
  • this consideration can be particularly relevant due to an optional subsequent filtering of the distance image, in which, for example, a sliding window for the filtering can contain both excluded and non-excluded pixels at the same time.
  • the evaluation of the noise is not passed on as binary information, but rather more than two levels are provided, for example three or four (or more) levels. If, for example, three stages are provided according to further preferred embodiments, these could have the following meanings in further preferred embodiments: Level 0: The descriptor should not be used due to noise. Level 1: Due to noise, the descriptor is suitable for "Application A", but not for "Application B".
  • Level 2 The descriptor is suitable for "Applications A" and "B", so the noise is, for example, not significant.
  • Applications A and “B” can stand for two applications, for example: optical flow, change detection, correspondence formation, disparity estimation, tracking, etc.
  • the local signal-to-noise ratio of an image region under consideration can be attached to the respective descriptor as additional information (e.g. in the sense of a concatenation), e.g. B. as a number.
  • additional information e.g. in the sense of a concatenation
  • a distance calculation can evaluate and pass on this additional information, e.g. as a minimum or maximum or mean value of the two numbers of the descriptors involved in the distance calculation.
  • this additional information can, for example, also be further taken into account in a subsequent optional filtering step and, if necessary, also passed on, for example as the confidence of a decision made about the presence of an object in the first and / or second image.
  • the method further comprises: at least temporarily storing the first descriptor image and / or the second descriptor image, e.g. for subsequent use. For example, when two distance images are determined consecutively, a participating descriptor image can be used twice for the distance calculation, namely once in the sense of the first descriptor image and a second time in the sense of the second descriptor image.
  • the method further comprises: Compensating for a movement associated with the first image and / or the second image, in particular a proper movement a camera providing the first image and / or the second image, for at least one area, in particular a surface.
  • first image and the second image are each part of the same video data stream of at least one camera.
  • At least one further image is present in the video data stream between the first image and the second image.
  • a time interval between the first image and the second image can be changed, for example, by not using directly consecutive images, e.g. of the video data stream, as the first and second image, but rather that e.g. one or more images, e.g. of the Video data stream between the first image and the second image can be omitted.
  • detections eg in the form of the detection image
  • quick time sequence which is important for a graphical display for a user, for example can be (so that it does not jerk).
  • the method in particular at the same time, is performed on different Image pairs (first image, second image) of the same or the same video data stream is executed, a respective first image and a respective second image each having a different time interval from one another.
  • Further preferred embodiments relate to a device for processing, in particular digital, images, the device being designed to carry out the method according to the embodiments.
  • the device has: a computing device ("computer") having at least one computation core, a memory device assigned to the computing device for at least temporary storage of at least one of the following elements: a) data, b) computer program, in particular for execution of the method according to the embodiments.
  • a computing device having at least one computation core
  • a memory device assigned to the computing device for at least temporary storage of at least one of the following elements: a) data, b) computer program, in particular for execution of the method according to the embodiments.
  • the data DAT can at least temporarily and / or partially contain the at least one video data stream and / or the first image and / or the second image and / or data that can be derived therefrom, e.g. the first descriptor image and / or the second descriptor image or the Have distance image or the detection image or at least parts thereof.
  • the memory device has a volatile memory (e.g. main memory (RAM)) and / or a non-volatile memory (e.g. flash EEPROM).
  • a volatile memory e.g. main memory (RAM)
  • a non-volatile memory e.g. flash EEPROM
  • the computing device can also have at least one of the following elements: microprocessor (mR), microcontroller (pC), application-specific integrated circuit (ASIC), system on chip (SoC), programmable logic module (e.g. FPGA, field programmable gate array) , Hardware circuit, graphics processor (GPU, graphics processing unit), or any combination thereof.
  • microprocessor mR
  • microcontroller pC
  • ASIC application-specific integrated circuit
  • SoC system on chip
  • programmable logic module e.g. FPGA, field programmable gate array
  • Hardware circuit e.g. FPGA, field programmable gate array
  • GPU graphics processing unit
  • Further preferred embodiments relate to a computer-readable storage medium, comprising instructions which, when executed by a computer, cause the computer to carry out the method according to the embodiments. Further preferred embodiments relate to a computer program comprising instructions which, when the program is executed by a computer, cause the computer to execute the method according to the embodiments.
  • Further preferred embodiments relate to a data carrier signal that characterizes and / or transmits the computer program according to the embodiments.
  • the data carrier signal can be received, for example, via an optional data interface of the device.
  • FIG. 1 schematically shows a simplified block diagram according to preferred embodiments
  • 3A schematically shows an exemplary image according to further preferred embodiments
  • FIG. 3B schematically shows a descriptor image associated with the image according to FIG. 3A according to further preferred embodiments
  • FIG. 4B schematically shows a detection image associated with the distance image according to FIG. 4A according to further preferred embodiments
  • 6 shows schematically exemplary weightings for filtering according to further preferred embodiments
  • 7 schematically shows an exemplary image according to further preferred embodiments
  • FIG. 11 schematically shows a simplified block diagram of a device according to further preferred embodiments
  • FIG. 1 schematically shows a simplified block diagram of a system 10 for use with a method for processing images in accordance with preferred embodiments.
  • the system 10 is designed, for example, as a vehicle, in particular an industrial truck (e.g. forklift and / or forklift) and has at least one camera 12 that captures multiple images B1,
  • the system 10 can also have at least one further camera 12 ', which in turn can provide, for example, one or more images or a corresponding video data stream (not shown).
  • the system 10 can move by itself, for example in an environment U, for example on a reference surface RF such as a floor surface (e.g. a manufacturing facility).
  • a reference surface RF such as a floor surface (e.g. a manufacturing facility).
  • objects OBJ which in particular can also represent obstacles H for the system 10, can be present in the environment U at least temporarily.
  • the system 10 does not represent a vehicle or the system 10 does not have a vehicle, but for example the camera 12, which, as described above, delivers, for example, the video data stream VDS or the images B1, B2 or one containing the camera 12 stationary facility.
  • the camera 12 can be provided to observe a scene SZ in the environment U, for example it can be used as a surveillance camera.
  • FIG. 2A Further preferred embodiments relate to a method, in particular a computer-implemented method, see FIG. 2A, for processing, in particular digital, images B1, B2 (FIG. 1), having the following steps: providing 100 a first image B1 and a second image B2 (in particular in each case as a digital image), transforming 102 the first image B1 into a first descriptor image DB1 and the second image B2 into a second descriptor image DB2, determining 104 a distance image DISTB based on a comparison of the first descriptor image DB1 with the second descriptor image DB1, and , optionally, forming 106 a detection image DETB based on the distance image DISTB.
  • a method in particular a computer-implemented method, see FIG. 2A, for processing, in particular digital, images B1, B2 (FIG. 1), having the following steps: providing 100 a first image B1 and a second image B2 (in particular in each case as a digital image),
  • the distance image DISTB or the information contained in the distance image DISTB characterizes at least partial changes with respect to the images B1, B2 or the descriptor images DB1, DB2 that can be derived therefrom.
  • the distance image DISTB can be used as the detection image DETB, the optional step 106 in particular being omitted.
  • the detection image DETB can be formed based on the distance image DISTB, which can take place, for example, in the optional step 106.
  • FIGS. 14A to 14F show examples of different detection images, which are described in more detail below, as they are e.g. according to preferred embodiments, e.g. based on the exemplary sequence according to FIG.
  • At least one transformation is used for transforming 102 into descriptor images DB1, DB2, which transforms the environment of a picture element ("pixel") of the first or second image B1, B2 in the respective image into a descriptor that converts this environment , preferably in a compact manner, that is to say with a few bits, in particular with fewer bits than corresponds to the information content of the surroundings of the pixel in the image B1, B2 itself.
  • the value of the descriptor is referred to as the signature.
  • the signature has a fixed length, in particular word length, of e.g. B. 8 bit to e.g. 32 bit, but in further preferred embodiments it can also be longer than 32 bit or shorter than 8 bit.
  • the transforming 102 includes carrying out the transformation for a respective environment of a plurality of pixels, for example each pixel in the first image B1 or in the second image B2, so that the result is a respective first or second "image of descriptors ", ie the descriptor image or images DB1, DB2 already mentioned, is created.
  • a descriptor image DB1, DB2 can also be understood as a multiplicity of descriptor values or signatures, which are preferably organized in a matrix-like arrangement of rows and columns are, according to the position of the pixels evaluated for their formation or the respective surroundings of a pixel under consideration.
  • a descriptor image DB1, DB2 is e.g.
  • each image element of the descriptor image is assigned the respective signature is, therefore has information about the environment of the output image B1, B2, which was included in the formation of the descriptor or the signature.
  • FIG. 3A shows an exemplary image or output image B1a
  • FIG. 3A shows an associated descriptor image DB1a which can be derived therefrom by means of the transformation 102 (FIG. 2A).
  • 3A, B illustrate by way of example how the environment U1 around the foot of the guide post of the output image B1a is converted into a descriptor D, see arrow A1.
  • the associated descriptor image DB1a (Fig. 3B), which no longer looks like the original image B1a (and is usually not intended for viewing), has information about the individual pixels of the original image B1a and their respective surroundings ("context").
  • descriptor image DB1a pixel-by-pixel descriptor values are stored which in the present case correspond, for example, to intensity values (black / white or gray levels).
  • the descriptor values see e.g. the designated descriptor D, describe a respective environment U of the output image B1a at a corresponding point in a more compact form (compared to the environment U of the output image B1a).
  • the descriptor image DB1a can also be formed “on the fly”, that is to say, for example, only when required.
  • pixels at the image edge of the first and / or second image B1, B2 for the transformation 102 (FIG. 2A) to the descriptor images DB1, DB2 can be ignored because there, for example, the frame for the transformation 102 over the image B1, B2 protrudes, the "pixel values" are therefore undefined.
  • pixels at the image edge of the first and / or second image B1, B2 can be taken into account for the transformation 102 to the descriptor images DB1, DB2, with missing or undefined pixel values at the edge possibly being supplemented according to a prescribable rule, for example by Copying the pixel values (for example neighboring) existing pixels and / or setting the previously undefined values to a prescribable value or values.
  • At least one of the methods listed below or parts thereof or combinations thereof can be used for transforming 102: SIFT (scale-invariant feature transformation, see e.g. US Pat. No. 6,711,293), SURF (Speeded Up Robust Features, e.g. according to Bay H., Tuytelaars T., Van Gool L. (2006) SURF: Speeded Up Robust Features. In: Leonardis A., Bischof H., Pinz A. (eds) Computer Vision - ECCV 2006. ECCV 2006. Lecture Notes in Computer Science, vol 3951 . Springer, Berlin, Heidelberg, https://doi.org/10.1007/11744023_32), ORB (E. Rublee, V.
  • a distance measure is present or can be specified which, for example, allows a difference between two descriptor values to be determined or assessed.
  • a similarity measure can also be used as an alternative to the distance measure, it being possible in particular for both measures to be convertible into one another. Therefore, in the following, only the distance measure is considered as an example and without limiting the general validity, with all steps and embodiments described by way of example also when using a similarity measure - instead of a distance measure - to evaluate a difference in the descriptor images DB1, DB2, e.g. to characterize a result of the comparison 104 of the first descriptor image DB1 with the second descriptor image DB2 can be used.
  • a descriptor is used for the transformation, for which a distance measure and / or a similarity measure is definable and / or defined, in particular the comparison 104 of the first descriptor image DB1 with the second descriptor image DB2 based on the distance measure and / or the similarity measure is carried out.
  • the Hamming distance or a distance measure based on the Hamming distance is used as the distance measure for the comparison 104, wherein in particular the Hamming distance is compared with a predeterminable threshold value and based on the comparison, in particular binary, comparison value is determined.
  • a Hamming distance of approximately N / 2 can be expected on a statistical average.
  • distance values close to 0 are to be expected in the case of essentially matching image areas - and for non-matching image areas, e.g. B. due to moving objects, distance values significantly greater than 0, and then z. B. be around N / 2.
  • the distance measure can optionally be further simplified, for example by binarization.
  • the Hamming distance can be compared to a threshold L and a binary one
  • FIG. 4A shows, by way of example, a distance image DISTB1 according to further preferred embodiments, which has been obtained, for example, using the Hamming distance with subsequent binarization.
  • FIGS. 15A to 15D show, by way of example, images from the camera 12 (FIG. 1) in which an industrial truck FFF and a person (not designated) have been successfully detected according to further preferred embodiments, and in which corresponding areas are graphically highlighted according to further preferred embodiments , so for example a change with respect to the images B1, B2 has been recognized. Furthermore, FIGS. 15A to 15D show an effect of a variation of the threshold L described above in the case of the optional binarization of the distance image DISTB (FIG. 2A) on a detection image DETB that can be derived therefrom, cf. the expansion of the blobs or highlights H1, H2, H3 ( Area of the person), H4 (area of the industrial truck FFF).
  • the images in FIGS. 15A, 15B, 15C, 15D each correspond to a detection image DETB which can be determined in accordance with further preferred embodiments and which is essentially based, for example, on the first image B1 and / or the second image B2, and in which the mentioned emphasis H1, H2, H3, H4 based on a distance image binarized according to the threshold L are included.
  • a threshold L 1 for the detection image DETBb according to FIG. 15B
  • a threshold L 2 for the detection image DETBc according to FIG. 15C
  • a threshold L 4 for the detection image DETBd according to FIG. 15D.
  • the descriptor values can also be determined if necessary, for example “on the fly”.
  • the transformation 102 of the first image B1 into the first descriptor image DB1 and / or the transformation 102 of the second image B2 into the second descriptor image DB2 can also be dynamic, that is to say for example when required and / or in real time, in particular for example also in direct temporal connection with the determination 104 of the distance image DISTB.
  • the transformation 102 of the first image B1 into the first descriptor image DB1 and / or the transformation 102 of the second image B2 into the second descriptor image DB2 can also be parallelized at least temporarily, e.g. if several computing cores 202a of a device 200 (see below on FIG 11) are available for executing the respective transformation (s) 102.
  • the first descriptor image DB1 belonging to a first point in time is compared with a second descriptor image DB2 which belongs to a second, in particular earlier, point in time.
  • both the transformation 102 to the descriptor images DB1, DB2 and the determination 104 of the distance image DISTB can be carried out, for example, only on a predeterminable sub-area of the first and second images B1, B2 or the descriptor images DB1, DB2 that can be derived therefrom.
  • subareas can also be selected dynamically, that is to say at the runtime of the method or a device 200 executing the method (FIG. 11), for example based on a current content of the images B1, B2 and / or (previously ) recognized changes and / or objects and / or a state of the system 10, for example its current direction of travel.
  • the comparison 104 (FIG. 2A) of the first descriptor image DB1 with the second descriptor image DB2 takes place, that is to say that Formation of the distance measure, pixel by pixel, so in each case a picture element of the first descriptor image DB1 is compared with a corresponding picture element of the second descriptor image DB2 and a corresponding value for the distance measure ("distance value") is obtained.
  • a descriptor value at a specific coordinate of the first descriptor image DB1 is compared with the descriptor value at the corresponding (same) coordinate of the second descriptor image DB2.
  • the distance value determined in this way is entered in a or the distance image DISTB at the corresponding coordinate.
  • the distance image DISTB has the same size (number of pixels, for example can be characterized by width and height) as the descriptor images DB1, DB2 to be compared.
  • the distance image DISTB (that is to say each picture element) can have values in the range 0 to N, for example, which characterize the comparison result between the first descriptor image DB1 and the second descriptor image DB2.
  • the method further comprises: filtering 105 the distance image DISTB, whereby a filtered distance image DISTB 'is obtained, in particular the formation 106 of the detection image DETB taking place based on the filtered distance image DISTB' .
  • the sequence according to Fig. 2B can, for example, be the sequence according to Fig.
  • the filtering 105 (Fig. 2B) is carried out in such a way that the distance image DISTB is converted into a filtered image DISTB 'which shows changes in a compact form, which is useful, for example, for a functional interface according to further preferred embodiments.
  • FIG. 4B shows, by way of example, a filtered distance image DISTBT, as it has been obtained by means of the filtering 105 based on the distance image DISTB1 according to FIG. 4A.
  • the Information from the filtered distance image DISTBT can be used, for example, to graphically highlight corresponding regions of the underlying image.
  • a person P1 with a goods carrier P1a is identified by the highlighting HP1.
  • two persons P1, P2 are identified in FIG. 14B by means of corresponding highlighting HP1, HP2, which are based, for example, on information from a filtered distance image similar to the filtered distance image DISTB1 'according to FIG. 4B.
  • FIG. 14C a person P1
  • FIG. 14D a vehicle F1
  • FIG. 14E a cyclist R1
  • FIG. 14F a person P1 and a forklift G1
  • the detection image has one or more contiguous regions, which are also referred to as "blobs" according to further preferred embodiments, for those image areas in which, in particular significant, changes between the (first and second) images or the herewith corresponding descriptor images are available.
  • the blobs can, for example, be determined based on the filtered distance image DISTBT, such as is obtained, for example, in step 105 according to FIG. 2B.
  • an (at least partial) determination of the (filtered) distance image DISTB, DISTB' and / or of the detection image DETB if necessary, eg "on-the-fly".
  • the detection image DETB (FIGS. 2A, 2B) can be converted into another form in an optional further or alternative step, e.g. B. to be able to transmit it more efficiently via an interface.
  • the contours of at least one blob are approximated, for example described as polygons.
  • the detection image can also be compressed, e.g. with a run length coding or another entropy coding, e.g. with a common coding for segment images.
  • the filtering 105 includes the application of a majority filter and / or a threshold value filter.
  • the distance image DISTB can in itself have a high level of detail and therefore, for example, not well suited to be passed on to a functional interface and / or transmitted in any other way.
  • the optional filtering step 105 (FIG. 2B) already described above is provided, which in further preferred embodiments "condenses" the information contained in the distance image DISTB, DISTB1, and thus compresses the distance image, for example for forwarding For example, to an optional subsequent function (e.g. determination of the detection image DETB) suitably prepared.
  • a distance image DISTB1 is shown by way of example in FIG. 4A as an input image for an optional filtering, and in FIG. 4B the filtered distance image DISTBT as an output image of the filtering step 105 (FIG. 2B).
  • both the input and the output data are binary images DISTB1, DISTBT, which are shown in black and white by way of example. Black stands for “changed” or “moved”, white for the opposite.
  • a majority filter is used for the filtering 105, the mode of operation of which can be easily understood with the aid of the exemplary illustration according to FIG.
  • Reference symbol BA1 from FIG. 5 denotes an exemplary part of the distance image DISTB1, for example according to FIG. 4A
  • reference symbol BA2 from FIG. 5 denotes an exemplary part of the filtered distance image DISTBT according to FIG. 4B.
  • the majority filter is used, for example, to determine in the sliding window GF, which in the present case covers 3 x 3 pixels of the distance image DISTB1, whether the black or white pixels in the window GF are in the majority. Since there is an odd number of pixels in the window GF, the result of the majority filter is unambiguous.
  • the output pixel AP which corresponds to the window GF at the current position in the distance image DISTB1, is given the color of the majority, here for example black, because in the window GF the result is "5: 4" in favor of black.
  • a comparison can also be made with another threshold.
  • a (filtered) result image DISTB1 '(FIG. 4B) with an edge length of the filter of 2R + 1 is by 2R many pixels shorter (narrower or lower) than the input image DISTB1 (FIG. 4A).
  • the filter for the filtering 105 (FIG. 2B), which is a two-dimensional filter, operates on a square window GF with an edge length of 3 ⁇ 3 (FIG. 5).
  • the filter window GF can, however, also be rectangular and not square, or have a different shape (for example circle or polygon).
  • the filter radius R is larger than shown in FIG. 5 for better clarity, e.g. with values between 2 and 30.
  • the voice weight SGW can have a maximum in the middle of the window GF and, for example, decrease towards the edge of the window GF, which is shown by way of example in FIG. 6 for a dimension along the coordinate axis x, see curve K1.
  • curve K2 corresponds, for example, to filtering with constant voice weight.
  • curve K1 can also be referred to as a "triangular filter” with the filter radius R, which effects center-weighted weighting
  • curve K2 can also be referred to as a "rectangular filter”.
  • the distribution of the voice weights along the other dimension not shown in Fig. 5 can be identical to the distribution along the dimension or axis x.
  • the resulting voice weight can then result, for example, as a product or as the sum of the voice weights of the first dimension and the second dimension.
  • integral filters or integral images or a representation can be used to carry out the filtering 105 (FIG. 2B) of the triangular filter K1 can be used as a convolution of two rectangular filters or a suitable series connection of several integral filters.
  • a suitable decision threshold is established, particularly in the case of center-weighted weighting (curve K1 according to FIG. 6).
  • the majority filter according to further preferred embodiments: Imagine that in the window with an odd edge length either the black or the white elements are just one element in the majority, and that the colors are evenly distributed are (e.g. alternating in a checkerboard pattern pixel by pixel).
  • the threshold is then to be selected in such a way that this simple majority just overturns the decision.
  • FIG. 7 shows an example of a binary-valued pattern (here a chessboard as an arbitrarily chosen example), which in FIG. 7 is increasingly noisy from top to bottom with binary noise (so-called salt and pepper noise).
  • the signal-to-noise ratio varies from infinity to 1/3, for example.
  • the left half of the image LH according to FIG. 7 can correspond to the distance image DISTB according to FIG. 2B
  • the right half of the image RH then corresponds, for example, to the filtered distance image DISTB '(Fig. 2B), which in further preferred embodiments can be used directly as a detection image DETB, for example, or on the basis of which the detection image DETB can be determined, see step 106 from FIG.
  • step 105 (FIG. 2B) of the filtering leads to a degree of (not too high) detailing that is appropriate for many applications in accordance with further preferred embodiments.
  • FIGS. 17A to 17D show the influence of the choice of the radius R for the optional filtering 105 (FIG. 2B) on the degree of detail obtained for a real one Example.
  • two people P1, P2 walk through the image of the camera of a forklift truck.
  • the output information AI has at least one of the following elements: a) acoustic signal, b) haptic signal, c) optical signal, d) image, in particular digital image, with at least one graphically highlighted image area, in particular one based image area graphically highlighted on the detection image (or the filtered) distance image DISTB, DISTB '), cf., for example, the highlighting HP1, HP2, F1, R1, P1, G1. 14 and / or the emphasis H1, H2 like. 15 and / or the highlighting H1, H2, H12 according to FIG. 17.
  • a change in the first and second images B1, B2 detected by means of the method cf., for example, FIG.
  • 2A can be efficiently brought to the attention of a user , e.g. a driver of a forklift truck.
  • a user e.g. a driver of a forklift truck.
  • the driver can reliably and easily perceive information about changes in the surroundings of his vehicle, for example , whereby, for example, accidents with people approaching the vehicle can be avoided.
  • the assessment measure BM when determining the distance image DISTB, the assessment measure BM in particular being a noise or a signal-to-noise ratio, in particular an area of the first image B1 and / or the second image associated with the respective descriptor B2, characterized.
  • the noise can thus also be taken into account when forming the descriptors or descriptor images DB1, DB2, for example according to the method described in DE 102017212 339.
  • each, formed descriptor (s) can be given an evaluation that is dependent on the strength of the noise or the signal-to-noise ratio (SNR): in some preferred embodiments, for example, in the form of a binary one Additional information (for example coded as an additional bit) which, in further preferred embodiments, can also be interpreted as the suitability of the descriptor for further processing.
  • SNR signal-to-noise ratio
  • the additional information allows, for example, to identify suitable descriptors less well (e.g. for subsequent further processing) due to the noise, e.g. B. as "unsuitable”, e.g. to treat them differently in the event of a change detection, than descriptors that are better suited for further processing, e.g. due to lower noise or higher SNR.
  • this additional information can be taken into account when determining the distance image ("distance calculation") DISTB (FIG. 2A), for example according to the following rule: Is at least one of the closed If comparative descriptors are identified as “unsuitable”, the distance is not formed according to the usual rule, but a different rule is applied, according to which, for example, the actual distance is replaced by a specifiable, in particular fixed, value.
  • the value can then be set to "0", for example, which means, for example, that image regions dominated by noise are treated as if they were motionless.
  • the Hamming distance can also be set to a different value, e.g. B. "1" or "2", so that, for example, areas excluded due to noise are not treated differently (e.g. better placed) than non-excluded unmoved areas.
  • this consideration can be particularly relevant due to an optional subsequent filtering of the distance image, in which, for example, a sliding window for the filtering can contain both excluded and non-excluded pixels at the same time.
  • the evaluation of the noise is not passed on as binary information, but rather more than two levels are provided, for example three or four (or more) levels. If, for example, three stages are provided according to further preferred embodiments, these could have the following meanings in further preferred embodiments:
  • Level 0 The descriptor should not be used due to noise.
  • Level 1 Due to noise, the descriptor is suitable for "Application A”, but not for "Application B”.
  • Level 2 The descriptor is suitable for "Applications A" and "B", so the noise is, for example, not significant.
  • “Applications A” and “B” can stand for two applications, for example: optical flow, change detection, correspondence formation, disparity estimation, tracking, etc.
  • further refinements that deviate therefrom are also conceivable.
  • the local signal-to-noise ratio of an image region under consideration e.g. associated with a descriptor, i.e. taken into account in the formation of the descriptor
  • additional information e.g. in the sense of a concatenation
  • a distance calculation can evaluate and pass on this additional information, for example as a minimum or maximum or mean value of the two numbers of the descriptors involved in the distance calculation.
  • this additional information can, for example, also be further taken into account in a subsequent optional filtering step 105 (FIG. 2B) and possibly also passed on, for example as the confidence of a decision made about the presence of an object OBJ1, OBJ2 (FIG. 4B) in the first and / or second image.
  • a participating descriptor image can be used twice for the distance calculation, namely once in the sense of the first descriptor image and a second time in the sense of the second descriptor image.
  • the method further comprises: Compensating 116 a movement associated with the first image B1 (FIG. 1) and / or the second image B2, in particular an intrinsic movement of the first image B1 and / or the second image B2 providing camera 12, for at least one surface RF, in particular surface, for example a homography compensation with respect to the ground plane RF.
  • FIG. 8 schematically shows a simplified block diagram according to further preferred embodiments.
  • the identifiers B101 to B503 have the following meaning:
  • B101 first camera image, see also image B1 according to FIG. 2A
  • B105 second camera image, see also image B2 according to FIG. 2A, image B105 being captured at an earlier point in time than the first camera image B101,
  • B201 first descriptor image, see also reference symbol DB1 according to FIGS. 2A, 2B, B205: second descriptor image, see also reference symbol DB2 according to FIGS. 2A, 2B, B300: step of comparison (see also reference symbol 104 according to FIG. 2A,
  • step 105 optional filtering step (see also step 105 as per Fig. 2B), e.g. in order to obtain a few (r) contiguous regions in the case of moving objects OBJ (Fig. 1),
  • B503 further processed form of the detection image, which is e.g. suitable for transmission via an interface, e.g. B. to a warning system or an actuator system (not shown), or for output on a display device (not shown), e.g. for the driver of the vehicle 10 (Fig. 1).
  • an interface e.g. B. to a warning system or an actuator system (not shown), or for output on a display device (not shown), e.g. for the driver of the vehicle 10 (Fig. 1).
  • FIG. 9 schematically shows a simplified block diagram according to further preferred embodiments. In comparison to Fig. 8, the following elements are added:
  • B250 Buffer for descriptor images. This provides e.g. older descriptor images B205 and stores them until they are no longer needed. This saves recalculations. For example, one transformation B200 may be sufficient for each new input image B101, in contrast to twice as many in FIG. 8, in the configuration of which the optional buffer B250 is not provided. With the optional memory B250 according to FIG. 9, a memory (not shown here) for storing camera images B101, e.g. until they are needed, B105, can also be dispensed with in further preferred embodiments.
  • FIG. 10 schematically shows a simplified block diagram according to further preferred embodiments.
  • the following elements are added: B150 optional step of compensating for a proper movement in an image for a surface RF (FIG. 1). For example a homography compensation with respect to the ground plane RF,
  • the compensation B150 according to further preferred embodiments only takes place, for example, in the lower branch Z2, i.e. for the second camera image B105, then this is compensated, for example, in such a way that it matches the first and thus newest camera image B101 (with regard to the selected surface RF).
  • this variant is usually the preferred one, since the resulting detection image B403 is then also available in the coordinates of the most recent camera image B101. This is particularly advantageous for real-time visualization in accordance with further preferred embodiments.
  • FIG. 11 relate to a device 200 for processing, in particular digital, images B1, B2, the device 200 being designed to carry out the method according to the embodiments (see, e.g., FIG. 2).
  • the device 200 has: a computing device 202 ("computer") having at least one computing core 202a, a memory device 204 assigned to the computing device 202 for at least temporary storage of at least one of the following elements: a) data DAT, b ) Computer program PRG, in particular for carrying out the method according to the embodiments.
  • a computing device 202 (“computer") having at least one computing core 202a, a memory device 204 assigned to the computing device 202 for at least temporary storage of at least one of the following elements: a) data DAT, b ) Computer program PRG, in particular for carrying out the method according to the embodiments.
  • the data DAT can at least temporarily and / or partially the at least one video data stream VDS (or a part thereof) and / or the first image B1 and / or the second image B2 and / or data that can be derived therefrom, for example the first Descriptor image DB1 and / or the second descriptor image DB2 or the distance image DISTB (or DISTB ') or the detection image DETB or at least parts thereof.
  • the memory device 204 has a volatile memory 204a (e.g. working memory (RAM)) and / or a non-volatile memory 204b (e.g. flash EEPROM).
  • volatile memory 204a e.g. working memory (RAM)
  • non-volatile memory 204b e.g. flash EEPROM
  • the computing device 202 can also have at least one of the following elements: microprocessor (mR), microcontroller (pC), application-specific integrated circuit (ASIC), system on chip (SoC), programmable logic module (e.g. FPGA, field programmable gate array ), Hardware circuitry, graphics processing unit (GPU), or any combination thereof.
  • microprocessor microcontroller
  • pC microcontroller
  • ASIC application-specific integrated circuit
  • SoC system on chip
  • programmable logic module e.g. FPGA, field programmable gate array
  • Hardware circuitry e.g. FPGA, field programmable gate array
  • GPU graphics processing unit
  • a data carrier signal DCS which the computer program PRG characterizes and / or transmits according to the embodiments.
  • the data carrier signal DCS can be received via an optional data interface 206, 208 of the device 200, for example.
  • there are also the video data stream VDS or the images B1, B2 can be received via an optional data interface 206.
  • the device 200 or components 202, PRG thereof can also be designed, for example, to implement the configuration according to FIG. 8 and / or 9 and / or 10 at least temporarily.
  • FIG. 12 relate to a use 300 of the method according to the embodiments and / or the device 200 according to the embodiments and / or the computer-readable storage medium SM according to the embodiments and / or the computer program PRG according to the embodiments and / or / or the data carrier signal DCS according to the embodiments for at least one of the following elements: a) Detection 302 a1) an environment U (FIG. 1) of a system 10, in particular a vehicle, in particular an industrial truck such as for example forklifts and / or forklifts, and / or a2) a scene SZ, b)
  • robust change detection 304 means, for example, no or a number of false detections (false alarms) that is below a predeterminable threshold value, e.g. B. because the lighting suddenly changes (e.g. through flashing lights, warning lights, light cones passing by, etc.), but at the same time reliable detection of relevant cases, especially all relevant cases.
  • a predeterminable threshold value e.g. B. because the lighting suddenly changes (e.g. through flashing lights, warning lights, light cones passing by, etc.), but at the same time reliable detection of relevant cases, especially all relevant cases.
  • FIG. 13 shows schematically exemplary image sequences BF1, BF2 according to further preferred embodiments, such as can be obtained, for example, by means of the camera 12 (FIG. 1) or based on the video data stream VDS.
  • an image B1, B2, B3, ... is periodically provided with the period duration AT - 1t, where t characterizes a time interval between two consecutive images, e.g. 1/30 second.
  • the first image B1 and the then already present second image B2 according to FIGS. 2A, 2B etc. can be evaluated, e.g. in order to determine a first detection image DETB.
  • the image B2 and then also the image B3 already present according to FIGS. 2A, 2B (in the sense of the images B1, B2) etc. can be evaluated, e.g. to determine a second detection image DETB, etc.
  • a time interval AT between the first image B1 and the second image B2 can be changed, for example, by not using consecutive images, e.g. of the video data stream, between the first and second images, but instead, for example, one or several images B ', for example of the video data stream between the first image and the second image, are omitted, see arrow B12, for which the following applies: AT-4T.
  • This is symbolized in FIG. 13 for the image sequence BF2 by the arrows not designated (with the exception of arrow B12), an origin of an arrow corresponding to the respective first image B1 of a pair of images being viewed, and a tip of the same arrow pointing to the respective second Image of the pair of images being viewed.
  • the first pair of images thus corresponds, for example, to the arrow B12, and the second and all further pairs of images to the arrows that are not individually designated in FIG. 13 in the image sequence BF2.
  • a quick response can be made and, in particular, detections (e.g. in the form of the detection image DETB) can also be output in quick time sequence, which is e.g. can be important for a graphical display for a user (e.g. so that it does not jerk).
  • FIG. 16 shows different detection images that have been obtained based on camera images each with a different time interval & T, where the following applies to FIGS. 16A & T-1t, DT-2T for FIG. 16B, DT-4T for FIG 16D & T-8T, see also Fig. 13.
  • the different time intervals between the camera images lead to different detection results.
  • the time interval DG between the two camera images B1, B1 can have a considerable influence on the detection result. If the time interval is small (FIG. 16A), only comparatively fast movements are detected, see the feet PF of person P. If it is large, see FIG. 16D, the detection is possibly too sensitive.
  • a suitable time interval & T which in further preferred embodiments can also be adjusted dynamically if necessary. That includes in further preferred embodiments, for example, also the possibility of a triggered camera recording, in which the camera does not record or provide the images in a fixed time grid, but in a controllable variable time interval.
  • the time interval AT can be specified as follows, for example: a) Selection of the time interval t between successive images B1, B2; B2, B3, ..., b) skipping pictures.
  • skipping by omitting images it could be the case, for example, that the process or system reacts too slowly to changes.
  • a solution for skipping without leaving out images according to further preferred embodiments is therefore shown on the basis of the image sequence BF2 according to FIG. 13.
  • Each newest image B101 is used here as an example and the change detection is carried out, for example, with the fourth from last image B105.
  • the possible disadvantage of a sluggish reaction is thus eliminated.
  • the time interval AT is sufficiently large (here e.g. 4/30 s) to be able to reliably detect even slowly moving objects.
  • the detection results can be combined or fused.
  • the method is carried out, in particular in parallel, on different images of the same or the same video data stream, cf. also step 118 according to FIG second image each have a different time interval from one another.
  • the principle according to preferred embodiments enables, among other things, a particularly robust change detection in camera images B1, B2 with comparatively little computing effort.
  • Robust means in particular that it does not provide false detections (false alarms), e.g. B. because the lighting suddenly changes (through flashing lights, warning lights, light cones passing by, etc.), but at the same time reliably detects relevant cases.
  • the detections can be supplied in a compactness that is appropriate for the subsequent function. This means, for example, if a person moves through the image, according to further preferred embodiments, if possible, an object should be connected to an interface (e.g. a system for visualization, i.e. e.g. output of the images B1, B2, possibly with highlighting based on the detected objects OBJ , OBJ1, OBJ2).
  • an interface e.g. a system for visualization, i.e. e.g. output of the images B1, B2, possibly with highlighting based on the detected objects OBJ , OBJ1, OBJ2.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Databases & Information Systems (AREA)
  • Software Systems (AREA)
  • Medical Informatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Image Analysis (AREA)

Abstract

L'invention concerne un procédé, en particulier un procédé mis en œuvre par ordinateur, permettant de traiter des images, en particulier des images numériques, ledit procédé comprenant les étapes suivantes consistant à : fournir une première image et une seconde image ; transformer la première image en une première image de descripteur et la seconde image en une seconde image de descripteur ; déterminer une image de distance sur la base d'une comparaison de la première image de descripteur avec la seconde image de descripteur ; et former une image de détection sur la base de l'image de distance.
EP21709970.4A 2020-03-09 2021-03-03 Procédé et dispositif de traitement d'images Pending EP4118620A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
DE102020202973.3A DE102020202973A1 (de) 2020-03-09 2020-03-09 Verfahren und Vorrichtung zum Verarbeiten von Bildern
PCT/EP2021/055366 WO2021180547A1 (fr) 2020-03-09 2021-03-03 Procédé et dispositif de traitement d'images

Publications (1)

Publication Number Publication Date
EP4118620A1 true EP4118620A1 (fr) 2023-01-18

Family

ID=74858447

Family Applications (1)

Application Number Title Priority Date Filing Date
EP21709970.4A Pending EP4118620A1 (fr) 2020-03-09 2021-03-03 Procédé et dispositif de traitement d'images

Country Status (3)

Country Link
EP (1) EP4118620A1 (fr)
DE (1) DE102020202973A1 (fr)
WO (1) WO2021180547A1 (fr)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE102022130297A1 (de) 2022-11-16 2024-05-16 Still Gesellschaft Mit Beschränkter Haftung Flurförderzeug und Verfahren zum Betreiben eines Flurförderzeugs

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6711293B1 (en) 1999-03-08 2004-03-23 The University Of British Columbia Method and apparatus for identifying scale invariant features in an image and use of same for locating an object in an image
US8928815B1 (en) * 2013-03-13 2015-01-06 Hrl Laboratories, Llc System and method for outdoor scene change detection
DE102017212339A1 (de) 2017-07-19 2019-01-24 Robert Bosch Gmbh Verfahren und Vorrichtung zur Bewertung von Bildausschnitten für eine Korrespondenzbildung

Also Published As

Publication number Publication date
WO2021180547A1 (fr) 2021-09-16
DE102020202973A1 (de) 2021-09-09

Similar Documents

Publication Publication Date Title
EP2368216B1 (fr) Procédé et dispositif pour l'analyse d'objets environnants et/ou de scènes environnantes ainsi que pour la segmentation en classe d'objets et de scènes
DE102015121339B4 (de) Systeme und verfahren zum ermitteln eines zustands einer fahrbahn
DE112017001311T5 (de) System und Verfahren zum Trainieren eines Objektklassifikators durch maschinelles Lernen
DE102017220307B4 (de) Vorrichtung und Verfahren zum Erkennen von Verkehrszeichen
DE102015207676A1 (de) Verfahren und Vorrichtung zur Hinderniserkennung auf Basis einer monokularen Kamera
DE102009012441B4 (de) Verfahren zur Reduktion des Speicherbedarfs bei der Bestimmung von Disparitätswerten für mindestens zwei stereoskopisch aufgenommene Bilder
DE102011106072A1 (de) Schattenentfernung in einem durch eine fahrzeugbasierte kamera erfassten bild unter verwendung einer optimierten ausgerichteten linearen achse
DE102008006709A1 (de) Videobasiertes Überwachungssystem und -verfahren
WO2018158020A1 (fr) Procédé et dispositif de détermination d'une trajectoire dans des scénarios hors route
DE102019131971A1 (de) Ein Bildverarbeitungsmodul
DE102019214402A1 (de) Verfahren und vorrichtung zum verarbeiten von daten mittels eines neuronalen konvolutionsnetzwerks
EP4118620A1 (fr) Procédé et dispositif de traitement d'images
DE102008036219A1 (de) Verfahren zur Erkennung von Objekten im Umfeld eines Fahrzeugs
DE102009048117A1 (de) Verfahren und Vorrichtung zum Erkennen einer Fehldetektion eines Objektss in einem Bild
EP2037407B1 (fr) Procédé de détection d'objet
EP2359308B1 (fr) Dispositif de production et/ou de traitement d'une signature d'objet, dispositif de contrôle, procédé et produit-programme
DE102022204722A1 (de) Verfahren zum Trainieren eines neuronalen Konvolutionsnetzwerks
DE102018201909A1 (de) Verfahren und Vorrichtung zur Objekterkennung
DE102020129164A1 (de) Verfahren und vorrichtung zur unterscheidung von verschiedenen konfigurationszuständen eines objekts auf der grundlage einer bildlichen darstellung des objekts
DE102020208080A1 (de) Erkennung von Objekten in Bildern unter Äquivarianz oder Invarianz gegenüber der Objektgröße
EP1998272A2 (fr) Dispositif de détermination de la vraisemblance d'un objet et/ou de l'existence d'un objet de recherche dans une fenêtre de lecture d'une image, procédé et programme informatique
WO2021063572A1 (fr) Dispositif et procédé de traitement de données à partir d'un réseau neuronal
WO2019072451A1 (fr) Procédé de traitement d'images
DE102019129029A1 (de) System und verfahren zur objektdetektion
DE102018010099A1 (de) Vermeiden der Blendung von Personen durch eine Lichtquelle

Legal Events

Date Code Title Description
STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: UNKNOWN

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE INTERNATIONAL PUBLICATION HAS BEEN MADE

PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: REQUEST FOR EXAMINATION WAS MADE

17P Request for examination filed

Effective date: 20221010

AK Designated contracting states

Kind code of ref document: A1

Designated state(s): AL AT BE BG CH CY CZ DE DK EE ES FI FR GB GR HR HU IE IS IT LI LT LU LV MC MK MT NL NO PL PT RO RS SE SI SK SM TR

DAV Request for validation of the european patent (deleted)
DAX Request for extension of the european patent (deleted)