Method for determining relative motion using a sequence of digital images

Abstract

A method for determining the relative motion of a device (10) and at least one object, the method being based on a sequence of digital images of the object detected from the position of the device, wherein the method comprises the following steps: calculating a plurality of optical flow fields (S2) from image pairs of the digital image sequence (S1), wherein the image pairs are formed by a latest image and an image of the image sequence having a time interval different from that of the latest image; locating at least one object in a sub-image region in each latest image (S3) and assigning the sub-image region to the object; forming a plurality of sub-optical flow fields (S4) from the plurality of optical flow fields, wherein a corresponding sub-flow field is generated by the intersection of a flow field of the plurality of flow fields with a sub-image region of the latest image of the image sequence; selecting at least one of the sub-flow fields from the plurality of sub-flow fields according to at least one criterion (S5) in order to facilitate the estimation of the scale change of the object; and estimating the scale change for the at least one object based on the selected at least one sub-flow field with the aid of the assigned sub-image region (S6).

Description

Technical Field

[0001] The invention relates to a method for determining a relative movement of a device and at least one object by means of a sequence of digital images.

[0002] Driver assistance systems detect scenes with the aid of electronic systems that record and process data, for example with the aid of radar, lidar, ultrasound or camera systems, in order to calculate parameters that characterize the relative movement of such a device relative to other objects. The processing of the data can be carried out (also depending on the electronic system used) based on a three-dimensional or two-dimensional model. In radar systems that detect distances in a systematic manner, a three-dimensional model is provided for analysis. Background Art

[0003] In a monocular digital camera system, the apparent motion of an object can be analyzed in an image sequence based on the two-dimensional image coordinates. The relative motion of the device and at least one object can be performed, for example, based on the scale change detected by the object or based on the scale change estimation based on the optical flow from the image sequence.

[0004] In this case, the estimation of the relative movement by means of a zoom change in the object detection is subject to a number of inaccuracies, such as the often incorrect assumption of a known and fixed object width or inaccuracies arising from changes in the extent of the object in successive camera images.

[0005] In the estimation of scale changes using optical flow, the optical flow belonging to the target object must be selected from the flow field, which can be done by motion segmentation. Motion segmentation of optical flow is a non-trivial and computationally intensive problem. Especially for embedded systems in the automotive field, the required computing power is often not available.

[0006] The results obtained are usually highly noisy, so that subsequent calculations have to deal with a large number of outliers.

[0007] From this scaling change, for example, the time to collision TTC (time-to-collision or time-to-contact) between the vehicles can then be derived as a characteristic parameter.

[0008] Here, the optical flow estimate is usually calculated at fixed time intervals over the entire image.

[0009] DE 10 2011 006 629 A1 describes a method for determining a time-to-collision of a vehicle equipped with a video system based on an evaluation of a video image of the vehicle's surroundings, wherein a collision parameter TTC is calculated to determine the time-to-collision of the vehicle with any part of the vehicle's surroundings. The calculation of the collision parameter TTC is performed by integrating the expansion rate of the optical flow field of the entire vehicle surroundings recorded by the video system in the video image. The invention also relates to a vehicle guidance system for carrying out the method according to the invention.

[0010] The object of the present invention is to specify a method for determining the relative movement of a device and at least one object with the aid of a digital image sequence, by means of which method a high accuracy in the estimation of the relative movement can be achieved with low computing power and the method also makes it possible to calculate other characterizing parameters of the relative movement from measures for the estimated relative movement. Summary of the invention

[0011] According to the invention, a method, a system, a computer program product and a computer-readable storage medium for determining the relative movement of a device and at least one object based on a sequence of digital images of the object detected from the position of the device according to the features of the independent claims are described, which at least partially have the mentioned effects. Advantageous embodiments are the subject matter of the dependent claims and the following description.

[0012] The invention is based on the recognition that objects that are far away from the device are imaged smaller in images of an image sequence, for example, captured in perspective by means of a camera, than objects that are close. Therefore, according to the invention, an optical flow field is calculated from image pairs of such an image sequence with different time intervals and different sizes, wherein the following partial area on the image is taken into account, in which the object is imaged. This partial area - in which the object is imaged - can also be called a "bounding box" or "region of interest ROI". Subsequently, a suitable flow field is selected from a plurality of flow fields for determining the relative motion based on a criterion that improves the reliability and precision of the estimate.

[0013] In the method according to the invention, the relative movement of the device and at least one object is determined based on a sequence of digital images of the object detected from the position of the device. The device can be, for example, a motor vehicle, which itself has a movement relative to the ground. However, the device can also be stationary and the object (for example a motor vehicle) moves relative to the device.

[0014] For this determination, a plurality of optical flow fields are calculated from image pairs of a digital image sequence, wherein the image pair is formed by a most recent image and an image of the image sequence having a different time interval than the most recent image. In this case, the optical flow describes the movement of an image point recorded at time i relative to an image point recorded at time j in an image of the image sequence based on a flow vector which is assigned to the current image point and thus defines the optical flow field.

[0015] Furthermore, in the method according to the invention, at least one object, the relative movement of which is to be determined, is located in a sub-image area of ​​the digital image sequence in the respective most recent image, and the sub-image area is assigned to the object. Thus, for each image in the image sequence, the following area is identified in which the object is detected, and for further calculations it is sufficient to take into account these sub-image areas ROI (English: region of interest) that characterize the object. Usually, a rectangular area is assigned to the object for this purpose, which largely corresponds to the object. Alternatively, the sub-image areas can also be derived from so-called "semantischen Segmentierung". In another embodiment of the invention, the localization of at least one object can be performed only for selected images of the image sequence. If multiple objects are located in the respective most recent images, these objects can be tracked throughout the entire method by means of the assigned sub-image areas through the assignment of sub-image areas to objects.

[0016] A plurality of sub-optical flow fields are formed from the plurality of optical flow fields, wherein a corresponding sub-flow field is generated by the intersection of a flow field of the plurality of flow fields and a sub-image region of a latest image of the image sequence.

[0017] The flow field is calculated based on the latest image of the image sequence and the flow field is calculated with the aid of a further image of the image pair, which has a corresponding time interval with the current image. The object is located in a sub-image area in the current image, and the intersection of this sub-image area of ​​the image and the flow field calculated based on the latest image determines the sub-flow field assigned to the flow field.

[0018] At least one sub-flow field is selected from the plurality of sub-flow fields in accordance with at least one criterion to facilitate estimation of a scale variation of the object.

[0019] An estimation of the scale change for at least one object is performed based on at least one selected sub-flow field. For the estimation of the scale change, pairs of flow vectors are formed, with the aid of which the scaling change is calculated. Therefore, it is beneficial for the scale estimation if a suitable sub-flow field is selected, because the scale estimation is based on this sub-flow field.

[0020] If multiple objects are located in the image sequence, the method is performed for each object. That is to say, for each object, a suitable sub-flow field is selected based on a criterion. For this purpose, a unique and clear marking is performed for each object or for each associated sub-image area by allocation.

[0021] This method makes it possible to select the most suitable partial flow fields according to predefined criteria for estimating the size change of objects which are located at different distances from the device and are therefore imaged differently in the images of the image sequence.

[0022] Thus, the quality of the scale calculation is improved by selecting the most suitable sub-optical flow fields from different time intervals. For example, if the object is at a greater distance from the device, longer flow vectors can be used for the estimation by selecting, for example, a flow field with a greater time interval between the images of the corresponding image pair. Thus, for the estimation of the scale change, a different time interval can be used for the calculation of the optical flow at time i than at time j.

[0023] According to one embodiment of the invention, it is proposed that the criterion for selecting a sub-flow field takes into account the degree of overlap of the sub-image area with the calculated flow field. It is thus ensured that the object in the image area is within the calculated flow field and that a sufficient portion of the object represented by the sub-image area can be taken into account. This is important, for example, in particular when the flow field is not calculated for the entire area of ​​the images of the image sequence in order to reduce the computational overhead. This will be explained further below. Additionally or alternatively, the criterion may take into account the absolute size of the intersection of the sub-image area with the calculated flow field, since the overlapping area formed by this intersection area may be so large that the flow vectors within this coverage are sufficient for the estimation of the scale change.

[0024] According to another embodiment of the invention, the criterion for selecting the partial flow fields takes into account the magnitude of the signal-to-noise ratio of the flow vector of the partial flow field and additionally or alternatively a quality measure from the calculation of the flow vector In this case, the reliability in the quality of the sub-flow field is given either by the signal-to-noise ratio of the determined flow vector or also by the quality value determined in the calculation of the flow vector. Furthermore, the criterion for selecting the sub-flow field can additionally or alternatively take into account the number of flow vectors in the sub-flow field and / or the length of the flow vectors of the sub-flow field. In this case, if no other quality measure is available, the validity of the flow vector itself is essentially determined by means of the signal-to-noise ratio. The higher the number of such flow vectors, the more reliable and robust the estimation of the scale change. Accordingly, the length of the flow vector can improve the estimation of the scale change to a certain extent.

[0025] According to another embodiment of the invention, the criterion for selecting the sub-flow fields takes into account the detection quality of the localization of the object in the sub-image area of ​​the image of the image sequence. If the localization of the object in the sub-area of ​​the image is reliable, the correct flow vector is also selected in the sub-flow field, which improves the estimation.

[0026] Additionally or alternatively, the criterion may also take into account features of the image region in which the intersection of the sub-image region with the calculated flow field is located. Thus, properties of the images of the image sequence also enter into the selection of the sub-flow fields, which properties have a significant influence on the quality of the flow vectors, for example. This may involve contrast, homogeneity, gray value distribution, gradients or also more complex features, such as the histogram of oriented gradients ( H istogram of o Histogram of oriented gradients (HOG) (in this histogram of oriented gradients, the distribution of local intensities or the arrangement of edges is evaluated) and similar parameters that characterize the persuasiveness of an image.

[0027] In one embodiment of the method according to the invention, it is provided that the criterion for selecting the partial flow fields presupposes that the following inequality is fulfilled:

[0028] FV>alpha*(SZ / SZ MAX )*FV MAX

[0029] In addition, the size of the time intervals in which the sub-flow fields are calculated is taken into account. This criterion evaluates in particular the number of flow vectors in a sub-flow field compared to the number in other sub-flow fields. The derivation of this formula is carried out below in this description. The size of the time intervals is taken into account in the following way: from a plurality of sub-flow fields, preference is given to sub-flow fields calculated with larger time intervals.

[0030] According to one embodiment of the invention, it is provided that a plurality of optical flow fields are calculated from an image portion of an image of the image pair, wherein the image portion represents a partial region of the respective image.

[0031] Thus, in one application of the method, objects at a greater distance from the device (e.g. a vehicle moving on the ground relative to the device in front of the device) are shown in another image portion of an image of the image sequence compared to objects located in front of the device in a close range.

[0032] This makes it possible to already select the scene to be observed when calculating the optical flow field and thus save computing effort, since the flow field is calculated over a restricted image region. Thus, for example, the image of the sky in the image sequence can be hidden, or if an object to be estimated is expected to be in the close or long range of the device, the image portion can be selected accordingly.

[0033] According to one embodiment of the invention, it is provided that the calculation of a plurality of optical flow fields is carried out for at least two distinguishable image parts of an image of an image pair and at respectively different time intervals.

[0034] The assignment of the corresponding image parts to the time intervals for the calculation of the flow field makes it possible for the corresponding image regions in which objects from different distances in front of the device are imaged to take into account the time intervals that are relevant for this region and thus reduce the computational effort without a loss of quality. Since smaller sub-image regions of distant objects require longer time intervals for a robust estimation, short time intervals are required for a correspondingly timely reaction in the close range. Thus, not only is the computational effort saved, but also a matching calculation is performed.

[0035] According to a further embodiment of the invention, it is provided that at least two distinguishable image components are arranged correspondingly in terms of their arrangement and size on the images of the image pair in order to localize the object at different distances from the device.

[0036] If an object moves in front of the device over a substantially flat area, the sub-image area of ​​the image in front of the device, in which the distant object is imaged, is significantly smaller than the sub-image area in the case of an object in the close range in front of the device. If these perspective conditions are taken into account, for example, the image portion for the more distant object can be smaller and placed further away in the direction of the upper edge of the image of the image sequence than the image portion for the object in the close range of the device.

[0037] According to one embodiment of the invention, at least two image portions are distinguishable by their area extension, the smaller image portion being arranged completely within the larger image portion and the smaller the image portion, the greater the time interval respectively allocated for calculating the flow field.

[0038] With this embodiment of the invention, the advantages resulting from perspective observation are combined with a longer time period which is advantageous for the estimation of the dimensional change.

[0039] According to one embodiment of the invention, a plurality of optical flow fields are calculated from image parts of an image pair, wherein the image parts and / or the assignment of the time intervals of the image parts to the image pairs are determined in each case after the acquisition of the corresponding image sequence.

[0040] It is thus possible that, during the running of the method, the parameters for calculating the flow field, such as the position, size and the assigned time interval, can be adapted, for example, based on information from the object positioning.

[0041] According to the present invention, a computer program product is also disclosed, which comprises instructions which, when the program is executed by a computer, cause the computer to carry out the above method.

[0042] According to the present invention, a computer-readable storage medium is also disclosed. The computer-readable storage medium includes instructions. When the program is executed by a computer, the instructions cause the computer to implement the above method.

[0043] The invention also describes a system for determining the relative movement of a system and at least one object. The system has a device for detecting a sequence of digital images of an object. Such a device for detecting a sequence of digital images can be, for example, a digital camera or a digital video camera. In addition, the system has an analysis unit, which is coupled to the device in such a way that the image sequence is transmitted to the analysis unit. This can be done, for example, by means of a bus system, which couples the device for detecting a sequence of digital images to the analysis unit. The analysis unit is configured to carry out the method according to the invention described above and to provide the result of the method at an output of the analysis unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The embodiments of the present invention are Figure 1 and 2 The accompanying drawings show:

[0045] Figure 1 A flow chart showing a method for determining relative motion of an apparatus and at least one object;

[0046] Figure 2 A system for determining a relative motion of a device and at least one object is shown. DETAILED DESCRIPTION

[0047] exist Figure 1 In the flow chart 10 , method steps for determining a relative movement of a device and at least one object are shown.

[0048] For the method, a sequence of digital images S1 is provided both for calculating the flow field {U(x, y, τ)} S2 and for localizing the object S3.

[0049] In step S2 of an embodiment of the method according to the invention, a plurality of optical flow fields {U(x, y, τ)} are calculated. In this case, each flow field {U(x, y, τ)} is formed by a plurality of flow vectors U, which are respectively assigned to the coordinates x and y of the image. The flow vectors U are calculated from image pairs of a sequence of digital images with a time interval τ. In this case, the individual flow fields {U(x, y, τ)} with a plurality of different time intervals τ: 1 ... n are calculated from the latest image and the images of the sequence of images.

[0050] Step S7 is an optional configuration of this embodiment. Step S7 is located before step S2, i.e., calculating a plurality of flow fields. Here, the calculation of a plurality of optical flow fields in S2 is based only on image portions of the images of the image pair, wherein the image portions thus represent partial areas of the respective images. Therefore, in step S7, a determination of a smaller image portion is provided for the calculation of a plurality of flow fields, such as in Figure 1 As outlined in step S7.

[0051] As described above, regions such as the sky, which are not of critical relevance for the estimation of scale variation, may therefore be hidden from the image, thereby reducing computational overhead.

[0052] In a further alternative embodiment of the invention, in step S7 , it is predefined for the calculation of a plurality of optical flow fields in step S2 that different time intervals are respectively predefined for the distinguishable image parts of the images of the image pair.

[0053] The calculation S2 of a plurality of flow fields can be performed with the aid of any number of image parts and associated time intervals. As shown above, the following can thus be achieved: objects and their sub-image areas that are shown small at a distance for perspective reasons can be fully detected by smaller image parts during the flow field calculation. The computational overhead for calculating the flow field is thus reduced, and the accuracy of the estimation of the scale change is improved by the possible allocation of longer time intervals for more distant objects. In the close range, where the sub-image areas of the objects are shown larger, the time intervals can be selected to be smaller in order to improve the reaction time.

[0054] If the image parts are arranged in such a way that a sub-image region associated with the localized object covers a plurality of image parts, an estimate from one image part can also be compared with an estimate from another image part.

[0055] In typical applications in motor vehicles, for example, three image portions of different sizes can be arranged centered and nested within one another, with the smallest image portion being assigned the longest time interval, the largest image portion being assigned the smallest time interval, and the medium image portion being assigned a time interval that lies between the other two time intervals.

[0056] In this way, the method according to the present invention can achieve a robust and accurate estimation of scale changes for both objects in a short-distance range and objects in a long-distance range, and the scale changes can also be matched to other given conditions.

[0057] In step S3, at least one object is located in a sub-image region of an image of the image sequence and the corresponding object is assigned to a sub-image region. The result of positioning S3 is a sub-image region of an image of the image sequence in which the object has been located and to which the object has been assigned.

[0058] In step S4, each sub-flow field is formed from the intersection of the calculated flow field {U(x, y, τ)} and at least one sub-image region, whereby a plurality of sub-flow fields are formed with the aid of a plurality of calculated flow fields {U(x, y, τ)}. Each sub-flow field is characterized by a time interval for determining the flow field {U(x, y, τ)}.

[0059] In step S5 , a sub-flow field is selected from a plurality of sub-flow fields in accordance with a criterion, wherein the estimation of the scale variation of the object is facilitated by the selection according to a suitable criterion.

[0060] The criteria for selecting at least one suitable sub-flow field from the plurality of sub-flow fields are based on necessary conditions for estimating the scale change from the sub-flow fields on the one hand and on quality considerations of the possibility for a robust and accurate estimation of the scale change on the other hand.

[0061] A necessary condition for the scale estimation with the aid of the sub-flow fields is a minimum size of the intersection between the calculated flow field {U(x,y,τ)} and the sub-image region in which the object, for example a motor vehicle, has been located. This means that the intersection between the sub-image region and the image part must have a minimum size, determined, for example, empirically, for a robust estimation of the scale change. In the empirical determination, a trade-off between the robustness of the estimate and the required computing time can also be taken into account.

[0062] Furthermore, for the estimation of the scale change it is necessary that the intersection between the calculated flow field {U(x,y,τ)} and the sub-image region has a sufficient number of flow vectors for the estimation of the scale change, with the aid of which the estimation of the scale change is performed. In determining the sufficient number of flow vectors, quality criteria such as the signal-to-noise ratio or a quality factor from the flow vector calculation can also be taken into account.

[0063] The quality criteria for selecting the sub-flow fields can relate to the flow vectors of the sub-flow fields, features of the object detection in the sub-image regions and features in the image parts.

[0064] The quality metric for selecting a sub-flow field relates in particular to the number of flow vectors in the sub-flow field under consideration or the density of the sub-flow field occupied by flow vectors and the quality parameter of the flow vectors. Provided that the quality measure is determined and provided by the calculation algorithm of the flow vector, the quality magnitude of the flow vector can be provided by an optional data exchange V1 between the calculation of the plurality of flow fields in step S2 and the selection in step S5.

[0065] Furthermore, the quality criterion can relate to a feature of the partial image region which results from the localization of at least one object. This includes the detection quality or the reliability or probability of the localization.

[0066] In order to take into account the quality criteria from the object detection in the selection in step S5 , an optional data exchange V2 can take place between the object detection in step S3 and the selection of the partial flow fields in step S5 .

[0067] Features from the overlapping area between the sub-image area and the calculated flow field, such as contrast, gray value distribution, gradients, the presence of specific features, or also properties resulting from the histogram of oriented gradients HOG, i.e. taking into account the distribution of local intensities or the arrangement of edges, can optionally be included in the quality considerations.

[0068] In addition to such simple heuristics, more complex methods such as automated classifiers or learning classifiers, which are determined, for example, by “machine learning” (eg decision trees) or “deep learning”, for example “offline”, can also be used to select the most suitable partial flow fields.

[0069] Depending on the application of the method according to the invention and the available computational budget, individual or also all of these criteria can be used to select the most suitable partial flow field.

[0070] An example of selecting at least one sub-flow field from a plurality of sub-flow fields for estimating the scale change of the object and the criteria used are shown below.

[0071] Each sub-flow field is taken into consideration for the selection process for which the ratio of the overlapping area SZ of the intersection of the sub-image area with the image part to the entire area of ​​the sub-image area is above a certain threshold value, for example 0.75, or for which the overlapping area SZ (i.e. the intersection of the sub-image area with the calculated flow field) lies above a certain minimum size.

[0072] For the sub-flow fields thus selected, the number of flow vectors FV is then determined from the overlap area SZ. In addition, the maximum overlap area SZ is also determined over all considered sub-flow fields: MAX and the maximum number FV of flow vectors in the corresponding overlapping area SZ of one of the selected sub-flow fields MAX .

[0073] Then, the sub-flow field with the maximum time interval between the image pairs on which the calculation is based is selected, which satisfies the following conditions:

[0074] FV>alpha*(SZ / SZ MAX )*FV MAX

[0075] Here, alpha is a weighting factor to be selected, which can be determined empirically, for example.

[0076] Therefore, it is compared whether the number of flow vectors contained in the sub-flow field under consideration is greater than the value given by the overlap area SZ and the maximum overlap area SZ MAX The product of the ratio of and the maximum number of flow vectors in one of the subflow fields is multiplied by alpha.

[0077] If multiple objects are identified, the most suitable flow field is selected individually for each overlapping area.

[0078] In step S6, after selecting the most suitable sub-flow region, an estimation of the scale variation is performed. To estimate the scale variation S6, pairs of flow vectors are formed from the overlapping regions so that the relative motion can be quantified when the object moves along the optical axis of the image detection system. Typically, in the case of a scale variation, one pixel in one image is imaged on multiple pixels in the other image and vice versa.

[0079] From this scaling change, in a further step, in particular the collision time TTC (time to collision or time to contact) between the vehicles can then be derived as a characteristic variable for the relative movement.

[0080] The method shown can be implemented by means of a computer program product, which includes instructions which, when executed by a computer, cause the computer to carry out the method with all its optional salient features. The computer program product can also be stored on a computer-readable storage medium.

[0081] Figure 2 A system 20 for determining a relative movement of the system 20 and at least one object is shown. The system has a device 1 for capturing, for example, a sequence of digital images of an object. Such a device for capturing a sequence of digital images can be, for example, a digital camera or a digital video camera. In addition, the system 20 has an evaluation unit 2, which is coupled to the device 1 in such a way that the sequence of images is transmitted to the evaluation unit 2. This can be done, for example, by means of a bus system 1a, which couples the device 1 for capturing a sequence of digital images to the evaluation unit 2. The evaluation unit 2 is configured to carry out the above-described method according to the invention and to provide the results of the method at an output 3 of the evaluation unit.

Claims

1. A method for determining the relative motion of a device (10) and at least one object, the method being based on a sequence of digital images of the object detected from the position of the device, wherein: The method comprises the following steps: Calculating a plurality of optical flow fields (S2) from image pairs of the digital image sequence (S1), wherein the image pairs are formed by a latest image and an image of the image sequence having a different time interval than the latest image; locating at least one object in a sub-image area in each latest image (S3) and assigning the sub-image area to the object; forming a plurality of sub-optical flow fields from the plurality of optical flow fields (S4), wherein a corresponding sub-flow field is generated by the intersection of a flow field of the plurality of flow fields and a sub-image region of a latest image of the image sequence; selecting at least one of the sub-flow fields from the plurality of sub-flow fields in accordance with at least one criterion (S5) so as to facilitate estimation of the scale variation of the object; and Based on the selected at least one sub-flow field, a scale change is estimated for the at least one object by means of the allocated sub-image area (S6), Wherein, for objects in a long distance range, a longer time interval is allocated for calculating the optical flow field, and for objects in a short distance range, a smaller time interval is allocated for calculating the optical flow field, The criteria for selecting the sub-flow field (S5) are based on the premise of satisfying the following inequality: FV>alpha*(SZ / SZ MAX )*FV MAX , and additionally taking into account the size of the time interval over which the sub-flow fields are calculated, in, FV: the number of flow vectors in the corresponding sub-flow field, alpha: the weighting factor to be selected, SZ: an overlapping area / intersection between one of the flow fields of the plurality of flow fields and a sub-image area of ​​the latest image of the image sequence, SZ MAX : The maximum overlapping area / intersection of all considered sub-flow fields, FV MAX： The maximum number of flow vectors in the corresponding overlap region SZ of one of the selected sub-flow fields.

2. The method (10) according to claim 1, characterized in that: The criterion for selecting a sub-flow field (S5) takes into account the degree of overlap of the sub-image area with the calculated flow field and / or the absolute size of the intersection of the sub-image area with the calculated flow field.

3. The method (10) according to claim 1 or 2, characterized in that: The criteria for selecting a sub-flow field (S5) consider the size of the signal-to-noise ratio of the flow vectors of the sub-flow field and / or a quality measure from the calculation of the flow vectors and / or the number of flow vectors in the sub-flow field and / or the length of the flow vectors of the sub-flow field.

4. The method (10) according to claim 1 or 2, characterized in that: The criterion for selecting the sub-flow field (S5) takes into account the detection quality of locating the object (S3) in the sub-image region of the image of the image sequence and / or the characteristics of the image region in the intersection of the sub-image region and the calculated flow field.

5. The method (10) according to claim 1 or 2, characterized in that: The plurality of optical flow fields (S2) are calculated from image portions of images of the image pair, wherein the image portions represent partial regions of the respective images.

6. The method (10) according to claim 1 or 2, characterized in that: The calculation of the plurality of optical flow fields is performed for at least two distinguishable image parts of the images of the image pair and at respectively different time intervals (S2).

7. The method (10) according to claim 6, characterized in that: The at least two distinguishable image components are arranged in their arrangement and size on the images of the image pair in accordance with the perspective in order to localize the object at different distances from the device.

8. The method (10) according to claim 6, characterized in that: The at least two image portions can be distinguished by their area extent, the smaller image portion being arranged completely within the larger image portion and the smaller the image portion, the greater the respective time interval allocated for calculating the flow field.

9. The method (10) according to claim 6, characterized in that: A plurality of optical flow fields are calculated ( S2 ) from the image parts of the image pairs, wherein the image parts and / or the assignment of the image parts of the image pairs to time intervals are determined in each case after the acquisition of the corresponding image sequence.

10. The method (10) according to claim 7, characterized in that A plurality of optical flow fields are calculated ( S2 ) from the image parts of the image pairs, wherein the image parts and / or the assignment of the image parts of the image pairs to time intervals are determined in each case after the acquisition of the corresponding image sequence.

11. The method (10) according to claim 8, characterized in that: A plurality of optical flow fields are calculated ( S2 ) from the image parts of the image pairs, wherein the image parts and / or the assignment of the image parts of the image pairs to time intervals are determined in each case after the acquisition of the corresponding image sequence.

12. A computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method (10) according to any one of claims 1 to 11.

13. A computer-readable storage medium comprising instructions which, when executed by a computer, cause the computer to implement the method (10) according to any one of claims 1 to 11.

14. A system (20) for determining relative motion between the system and at least one object, characterized in that: The system comprises a device (1) for detecting a sequence of digital images of the at least one object and an analysis unit (2), which is coupled to the device (1) in such a way that the sequence of images is transmitted to the analysis unit (2), and the analysis unit (2) is configured to execute a method (10) according to any one of claims 1 to 11 and to provide a result of the method at an output (3) of the analysis unit.

Images