Method for detecting an event, computer program product, detection device and image sensor
The method improves event detection in image sensors by using tile-based gradient analysis and threshold verification to differentiate between significant events and ambient changes, reducing power consumption and false positives, thus optimizing power management.
Patent Information
- Application Number
- PCT/EP2025/051555
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2024-02-01
- Filing Date
- 2025-01-22
- Publication Date
- 2025-08-07
AI Technical Summary
Existing event detection methods in image sensors, such as those used in surveillance cameras, struggle with power consumption inefficiencies and inaccuracies due to false positives from changes in illumination and small movements, leading to unnecessary high-power mode activation.
A method involving frame division into tiles of varying sizes, gradient analysis, and threshold verification to confirm flagged tiles, reducing false positives and enabling efficient power management by switching between low-power and high-power modes based on significant event detection.
This approach enhances event detection accuracy while minimizing power consumption by reducing false detections and missed events, optimizing power usage by maintaining low-power mode until significant changes are confirmed.
Smart Images
Figure EP2025051555_07082025_PF_FP_ABST
Abstract
Description
[0001] METHOD FOR DETECTING AN EVENT , COMPUTER PROGRAM PRODUCT , DETECTION DEVICE AND IMAGE SENSOR
[0002] Image sensors e . g . for use in a surveillance camera, usually embed an event detection function that allows maintaining a sensor in a low-power consumption profile (with decreased videos resolution and frame rate ) until a signi ficant change is detected . When a signi ficant change is detected, the sensor is switched to a nominal profile consuming more power but acquiring a video with optimal quality to review and record the event .
[0003] Generally, attempts are being made in order to improve the event detection function .
[0004] It is an obj ect of the present invention to provide an improved method for detecting an event , an improved computer program product , an improved detection device and an improved sensor .
[0005] SUMMARY
[0006] According to embodiments , the above obj ect is achieved by the claimed matter according to the independent claims . Further developments are defined in the dependent claims .
[0007] Embodiments relate to a method for detecting an event in a sequence comprising at least a first frame and a second frame , each of the first and second frames comprising a plurality of tiles , each of the tiles comprising a plurality of pixels . The method comprises determining a change between the first and the second frame comprising determining at least one flagged tile , a first variation being associated to the flagged tile . The method further comprises determining, for each of flagged tiles , a set of surrounding gradients corresponding to a change of adj acent tiles surrounding the flagged tile between the first and the second frame , and determining, for each of the flagged tiles , a gradient module corresponding to a vector of di f ferences between the first variation and the set of surrounding gradients . The method further comprises confirming, for each of the flagged tiles , the flagged tile i f a norm of the gradient module is higher than a first threshold, and determining that an event is present based on a number of confirmed flagged tiles . For example , the first threshold may depend on the first variation . For example , the first threshold may be computed by multiplying the first variation with a coef ficient .
[0008] For example , determining at least one flagged tile may comprise dividing the first and the second frame into a plurality of first tiles having a first si ze , determining grid tile values based on a di f ference between tile values of the first tiles of the first frame and tile values of the first tiles of the second frame , and flagging those of the plurality of first tiles , for which the di f ference exceeds a second threshold .
[0009] For example , determining the grid tile values may comprise determining first grid tile values corresponding to a di f ference between first tile values of the first tiles of the first frame and first tile values of the first tiles of the second frame , determining second grid tile values corresponding to a di f ference between second tile values of second tiles of the first frame and second tile values of second tiles of the second frame , wherein the second tiles have a second si ze larger than the first si ze , and determining the grid tile values based on an average of the first grid tile values and the second grid tile values .
[0010] According to embodiments , determining the second grid tile values may comprise dividing the first and the second frame into a plurality of the second tiles having the second si ze , and determining a di f ference between tile values of the second tiles of the first frame and tile values of the second tiles of the second frame .
[0011] According to further embodiments , determining the second grid tile values may be performed on the basis of the first grid tile values .
[0012] According to embodiments , determining the grid tile value may comprise downsampling comprising determining an average of first grid tile values with a corresponding second grid tile value .
[0013] According to further embodiments , determining the grid tile value may comprise upsampling compri sing reducing a number of first grid tile values to determine a reduced grid tile value and determining an average of the reduced grid tile value with a corresponding second grid tile value .
[0014] For example , determining the first grid tile values may be performed parallel to determining the second grid tile values .
[0015] According to embodiments , a computer program product comprises instructions which, when the program is executed by a computer, cause the computer to carry out the method as explained above .
[0016] Further embodiments are directed to a detection unit for detecting an event in a sequence comprising at least a first frame and a second frame , each of the first and second frames comprising a plurality of tiles , each of the tiles comprising a plurality of pixels . The detection unit comprises a first detector configured to determine a change between the first and the second frame comprising determining at least one flagged tile , a first variation being associated to the flagged tile , and a gradient veri fication device configured to determine , for each of flagged tiles , a set of surrounding gradients corresponding to a change of adj acent tiles surrounding the flagged tile between the first and the second frame , determine , for each of the flagged tiles , a gradient module corresponding to a vector of di f ferences between the first variation and the set of surrounding gradients , and confirm, for each of the flagged tiles , the flagged tile i f a norm of the gradient module is higher than a first threshold . The detection unit further comprises a determination unit configured to determine that an event is present based on a number of confirmed flagged tiles . The first threshold may depend on the first variation .
[0017] The detection unit or components of the detection unit may be implemented as one or more computers carrying out the functionalities described . The first detector and the second detector may also be implemented as one computer carrying out the described functionality . For example , the first detector and the second detector may be components of an image sensor and may process data received from sensor elements that also form a component of the image sensor . According to further embodiments , the first detector and the second detector may be separate from the image sensor .
[0018] For example , the first detector may be configured to divide the first and the second frame into a plurality of first tiles having a first si ze , determine grid tile values based on a di f ference between tile values of the first tiles of the first frame and tile values of the first tiles of the second frame , and flag those of the plurality of first tiles , for which the di f ference exceeds a second threshold .
[0019] The detection unit may further comprise a second detector and an average determination unit , wherein the first detector is configured to determine first grid tile values corresponding to a di f ference between first tile values of the first tiles of the first frame and first tile values of the first tiles of the second frame . The second detector may be configured to determine second grid tile values corresponding to a di f ference between second tile values of second tiles of the first frame and second tile values of second tiles of the second frame , wherein the second tiles have a second si ze larger than the first si ze . The average determination unit may be configured to determine the grid values based on an average of the first grid tile values and the second grid tile values .
[0020] For example , the second detector may be configured to divide the first and the second frame into a plurality of the second tiles having the second si ze , and to determine a di f ference between tile values of the second tiles of the first frame and tile values of the second tiles of the second frame .
[0021] According to further embodiments , the second detector may be configured to determine the second grid tile values on the basis of the first grid tile values .
[0022] An image sensor may comprise the detection unit as described above .
[0023] A method for operating such an image sensor may comprise acquiring content with a first context , performing the method as explained above , switching the context to a second context i f an event is present , and acquiring content with the second context .
[0024] BRIEF DESCRIPTION OF THE DRAWINGS
[0025] The accompanying drawings are included to provide a further understanding of embodiments of the invention and are incorporated in and constitute a part of this specification. The drawings illustrate the embodiments of the present invention and together with the description serve to explain the principles. Other embodiments of the invention and many of the intended advantages will be readily appreciated, as they become better understood by reference to the following detailed description. The elements of the drawings are not necessarily to scale relative to each other. Like reference numbers designate corresponding similar parts.
[0026] Fig. 1A outlines a method according to embodiments.
[0027] Fig. IB outlines a method according to further embodiments.
[0028] Fig. 2A illustrates a method for determining a gradient.
[0029] Fig. 2B shows elements of a method according to embodiments.
[0030] Fig. 2C illustrates the determination of a gradient module.
[0031] Figs. 3A and 3B illustrate a frame divided according to different tile sizes.
[0032] Fig. 4A shows elements of a method according to embodiments.
[0033] Fig. 4B shows elements of methods according to further embodiments .
[0034] Fig. 4C shows elements of methods according to further embodiments .
[0035] Fig. 5A shows elements of a detection device according to embodiments . Fig. 5B shows elements of a detection device according to further embodiments .
[0036] Fig. 6A illustrates a sensor according to embodiments.
[0037] Fig. 6B illustrates a method performed by the sensor.
[0038] DETAILED DESCRIPTION
[0039] In the following detailed description reference is made to the accompanying drawings, which form a part hereof and in which are illustrated by way of illustration specific embodiments in which the invention may be practiced. In this regard, directional terminology such as "top", "bottom", "front", "back", "over", "on", "above", "leading", "trailing" etc. is used with reference to the orientation of the Figures being described. Since components of embodiments of the invention can be positioned in a number of different orientations, the directional terminology is used for purposes of illustration and is in no way limiting. It is to be understood that other embodiments may be utilized and structural or logical changes may be made without departing from the scope defined by the claims.
[0040] The description of the embodiments is not limiting. In particular, elements of the embodiments described hereinafter may be combined with elements of different embodiments.
[0041] Fig. 1A outlines a method according to embodiments. Generally, when an event is to be detected, frames are taken at different times. As is illustrated in Fig. 1A, a first frame 100 may be taken at a first point of time, and a second frame 109 may be taken at a second point of time. According to all embodiments described herein, the method may be a computer-implemented method . Generally, for performing event detection, each of the frames 100, 109 is divided into a predetermined number of equally sized areas called tiles. Each of the tiles comprises a plurality of pixels 120. Each tile has an associated tile value, which corresponds e.g. to the average of all the pixel values in that tile. According to further embodiments, another statistical value such as the median, the maximum, the minimum etc. may be employed. For example, the pixel value may represent the intensity and / or color information of the pixel. For every new frame that is acquired, each tile value is calculated and compared to the corresponding tile value of the previous frame.
[0042] In step S100, a change between a first frame 100 and a second frame 109 is determined. This step comprises determining at least one flagged tile. A tile is flagged when it shows a significant change compared to the previous frame. In particular, a tile may be flagged when a first variation, i.e. the difference between the tile value of frame 100 and the tile value of frame 109 exceeds a predetermined threshold.
[0043] As is further illustrated in Fig. 1A, after determining a change between the first and the second frame 100, 109 and determining at least one flagged tile determining gradients, S110, may be performed. For example, determining gradients S110 may comprise assessing a temporary change of tiles adjacent to the flagged tile. When the tiles adjacent to the flagged tile are changed in a similar manner as the flagged tile, it may be determined that the change of the flagged tile is due to a global event, e.g. a global illumination variation that may be caused by clouds passing by, sunsets, sunrises or light source obstructions outside the field of view of the camera. In this case, when a similar change occurs to the neighboring tiles, it is possible that the flagged tile is not confirmed. Accordingly, the method further comprises determining a gradient module, S114, corresponding to a vector of differences between the first variation and the set of surrounding gradients. The method further comprises confirming, S117, the flagged tile, if a norm of the gradient module is higher than a first threshold. These processes of determining gradients S110, determining a gradient module S114 and confirming a flagged tile S117 are repeated for each of the flagged tiles. Finally, at S120, it is determined that an event is present, e.g. on a basis of a number of confirmed flagged tiles. For example, when the number of confirmed flagged tiles exceeds a predefined minimum amount of tiles, it may be determined that an event is present.
[0044] Fig. IB outlines a method according to further embodiments. In addition to elements described with reference to Fig. 1A, in step S100, a change is determined on the basis of a first tile size. Moreover, in the upper branch of Fig. IB, a change may be determined on the basis of a second tile size which may be larger than the first tile size. At S130, an average of the determined changes is assessed. On the basis of this average, flagged tiles are determined. Thereafter, at S110, gradients are determined in a similar manner as has been explained with reference to Fig. 1A. At S114, a gradient module is determined, and at S117 a flagged tile is confirmed. Further, at S120, it is determined whether an event is present. These processes are performed in a similar manner as has been explained above.
[0045] In the following, the process for determining a gradient S110 will be explained in more detail, while referring to Fig. 2A.
[0046] Fig. 2A shows an example of a first frame 100 and a second frame 109 which are taken at different times. In more detail, the temporal difference between the first frame 100 and the second frame 109 may be At. For example, each of the first and the second frames may be divided in a plurality of first tiles 121.
[0047] As is shown in Fig. 2B, a comparison between the tile values of the first tiles 121 may be performed between the first frame 100 and the second frame 109. The different hatchings illustrate a difference between the first frame 100 and the second frame 109. As is illustrated in Fig. 2B, a flagged tile 124 may be determined. For example, a flagged tile may be a tile wherein a difference between the tile values of the first frame 100 and the second frame 109, i.e. the first variation 123, exceeds a second threshold. The first variation 123 may be associated to the flagged tile 124. For each of the flagged tiles 124, a set of adjacent tiles 126 surrounding the flagged tile 124 may be determined. Further, a set of surrounding gradients 127 may be determined. The surrounding gradient or surrounding difference value 127 corresponds to a change of a tiles adjacent to the flagged tile 124 between the first and the second frame 100, 109. In more detail, for determining the surrounding gradient or surrounding difference 127, the tile value of each of the surrounding tiles is calculated and compared to the corresponding tile value of the previous frame.
[0048] Fig. 2B shows four surrounding tiles 126 adjacent to the flagged tile 124. The hatching of these four surrounding tiles 126 represents a variation or a surrounding gradient between the first frame 100 and the second frame 109. According to implementations, more than four or less than four surrounding tiles 126 may be taken into account.
[0049] Thereafter, referring to Fig. 2C, a gradient module is determined (S114 ) . The gradient module corresponds to a vector of differences between the first variation and the vector of surrounding gradients. Thereafter, a norm of the gradient module may be determined. For example, an 1-norm may be taken. The norm is compared with a first threshold. If the norm of the gradient is higher than a first threshold, the corresponding flagged tile is confirmed (S117 ) .
[0050] As a result of the method explained with reference to Figs. 2A to 2C, the number of flagged tiles may be reduced. Thereafter, at S120, it may be determined that an event is present based on a number of confirmed flagged tiles.
[0051] Due to the employment of the above-described gradient verification, for example, the effect of global illumination variation may be cancelled. Accordingly, using the gradient verification will make the event detection algorithm more robust with respect to a change of brightness, e.g. caused by a change of an ambient light level.
[0052] According to further embodiments, attempts are made to decrease the influence of, for example, regular movement of small objects or other events that may be not relevant. In particular, attempts are made to detect a higher number of true events while at the same time avoid detecting false events, e.g. tree leaves moving, shadows or water. This approach may be based on the idea that tiles with different sizes may be sensitive to the detection of events with different scales.
[0053] Figs. 3A and 3B show the same frame 100 having different tile sizes. For example, Fig. 3A shows a frame 100 wherein a size of the tiles is approximately 150 x 160 pixels. When e.g. the person illustrated in the lower left portion of the frame moves within one tile, no change is detected. According to Fig. 3B, when employing e.g. a smaller tile size, e.g. approximately 75 x 80 pixels, a movement of the person may be detected when the person moves from one small tile to an other, both being in the same big tile illustrated in Fig. 3A.
[0054] On the other hand, when employing the large tile size as e.g. illustrated in Fig. 3A, a small movement of the tree in the left lower portion is not detected, whereas such a movement may be detected when employing a smaller tile size as is illustrated in Fig. 3B. Accordingly, a movement of a tree which is more likely to be detected by the small tile size will result in a detection of a false event.
[0055] For combining grid values of different sizes, different methods may be employed. For example, for upsampling, the 4 values of the small tiles are reduced to obtain reduced grid tile values. This reduction may be performed using any kind of reduction function, e.g. determining the mean value or maximum or another suitable value. The reduced grid tile value is then averaged with the big tile value. For downsampling, each value of the small tile is used and is averaged with the big tile value. In this case one value is kept for each of the small tiles.
[0056] Fig. 4A illustrates an example of a method of combining tile values of different tile sizes using upsampling. In the upper portion of Fig. 4A, a first frame 100 and a second frame 109 are divided into a plurality of second tile 122 having a second size. Moreover, in the lower portion of Fig. 4A, the first frame 100 and the second frame 109 may be divided in a plurality of first tiles 121 having a first size smaller than the second size. For example, the size of a second tile 122 may correspond to the size of four first tiles 121.
[0057] At S200, differences between the respective tiles are computed.
[0058] As a result, second grid tile values 152 may be determined for the upper portion of Fig. 4A, and first grid tile values 151 may be determined for the lower portion of Fig. 4A. At S206, the first grid tile values 151 may be reduced, e.g. by determining a maximum value or an average value. As a result, a reduced grid tile value 153 is obtained. Thereafter, at step S207, an average between the second grid tile value 152 and the reduced first grid tile value 153 may be determined. As a result, a grid tile value 150 may be output.
[0059] According to further embodiments, the process illustrated in Fig. 4A may be further simplified. The lower portion of Fig. 4B corresponds to the lower portion of Fig. 4A. However, differing from embodiments described with reference to Fig. 4A, the second grid values 152 are not determined from the second tiles 122. Instead, the second grid values 152 are determined based on the first grid tile values 151 of the first tiles 121. In more detail, the first grid tile values 151 are determined based on the first tiles 121, in a similar manner as has been described with reference to Fig. 4A. Then, at S207, an average of the first grid tile values 151 illustrated in the lower portion of Fig. 4B is determined. As a result, a second grid tile value 152 is determined. Using these computationally determined values, grid tile values 150 are determined in a corresponding manner as has been described with reference to Fig. 4A.
[0060] For example, the determination of an average at S207 may be different from the reduction processing at S206, which may e.g. be a determination of a maximum. These different functions introduce a non-linearity.
[0061] As is clearly to be understood, the tile values of different tile sizes may as well be determined using downsampling.
[0062] According to methods described with reference to Fig. 4B, the number of operations and thus the computational footprint of the multi-tile sizes approach may be reduced. Fig. 4C illustrates an example of a method of combining tile values of different tile sizes using downsampling. The upper portion corresponds to the processing that has been explained with reference to Fig. 4A. Moreover, in the lower portion of Fig. 4C, the processing to obtain the first grid tile values 151 may be similar to the processing described with reference to Fig. 4A. At step S207, each of the first grid tile values 151 is averaged with the second grid tile value 152. As a result, 4 grid tile values 150 are obtained.
[0063] As is clearly to be understood, downsampling may as well applied to the processing described with reference to Fig. 4B.
[0064] Fig. 5A outlines elements of a detection unit 15 according to embodiments. As is shown, a detection unit 15 according to embodiments comprises a first detector 110. The first detector 110 receives a first frame 100 and a second frame 109. The first detector 110 is configured to determine a change between the first and the second frame 100, 109. This may comprise determining at least one flagged tile. A first variation may be associated to the flagged tile. The detection unit 15 further comprises a gradient verification device 111 that is configured to confirm, for each of the flagged tiles, the flagged tile, if a norm of a gradient module is higher than a first threshold. For example, this may comprise determining, for each of the flagged tiles, a set of surrounding gradients and a gradient module corresponding to a vector of differences between the first variation and the set of surrounding gradients. For example, the first threshold may depend on the first variation. For example, the first threshold may be obtained by multiplying the first variation by a coefficient. The detection unit 15 further comprises a determination unit 115 that is configured to determine for example that an event is present based on a number of confirmed flagged tiles.
[0065] Fig. 5B shows a detection unit 15 according to further embodiments. In addition to elements shown in Fig. 5A, the detection unit 15 may comprise a second detector 112. The second detector 112 may be operable to determine flagged tiles based on a second tile size that is larger than the first tile size.
[0066] Due to the use of different sizes of tiles small noise movements such as leaves of a tree or non significant moves may be prevented from being detected as false events.
[0067] Fig. 6A shows an example of an image sensor 10 which comprises the detection unit 15 that has been explained above. For example, the image sensor may be a component of a surveillance camera, a doorbell or a traffic monitoring device.
[0068] Fig. 6B illustrates a portion of a method performed by an image sensor, for example, the image sensor 10 illustrated in Fig. 6A. At step S210, content is acquired with a context A. The context A may, for example, mean that content is acquired at a low frame rate wherein event detection is enabled. For example, the event detection function may be regularly triggered to look for some events in the captured low resolution, low frame rate images. For example, the full region of interest may be supervised and a subsampling at a ratio of e.g. 1 to 8 may be performed in both directions. The frames are acquired, e.g. at a decreased video resolution and a low frame rate resulting in a low power consumption. At S211 it is determined whether an event has been detected. When no event has been detected, the process returns to S210. If an event has been detected, S211, the sensor switches the context to context B and content is acquired with content B, S212. Context B may e.g. mean that a high frame rate is employed and a high resolution . Such a context may consume more power . Further, a video may be acquired with optimum quality to review and record the event . When no more event is detected, the sensor switches the context to context A and returns to S210 .
[0069] As is to be clearly understood, the described method and detection unit may be integrated in an image sensor in an alternative manner . For example , the image sensor may be configured to noti fy a host in case of detection of an event . Further, the host may decide i f the high power consumption mode , e . g . context B, shall be switched on or not . Further, the return to the low power mode can also be left at the host ' s initiative .
[0070] According to further implementations , the detection unit 15 may be a component of the host . The image sensor 10 may be operable to output low resolution, low frame rate images to the host . The event detection function is periodically triggered . When an event is detected, the host may reconfigure the image sensor to produce higher resolution and higher frame rate images . The host may also be responsible for switching the image sensor back to the low power mode .
[0071] Generally, any of the gradient veri fication device 111 or the second detector 120 enabling a multi-resolution analysis , may be activated or switched of f depending on the computational constraints .
[0072] Due to the functionality of the method for detecting an event , power consumption may be reduced while improving the detection accuracy . In particular, when the method described above is employed, the number of false detections and the number of missed signi ficant events may be largely decreased . As a consequence , power may be saved since the system exits the low-power mode only when a significant event is detected. With decreasing number of false detected events, power consumption may be optimized.
[0073] As is described, a trade-off is found to obtain the best results in terms of event detection while keeping the algorithm complexity and memory consumption at the lowest level. In particular, the described method is a low complex method that is more robust to non-signif leant change of the image (e.g. small moves, illumination changes and others) , thus improving its overall accuracy.
[0074] The described detection unit has an ultra-low footprint, e.g. a need for only small memory and computational resources. Accordingly, the detection unit may be easily embedded in a sensor. The sensor costs may be kept at a low level and its power consumption may be minimized. Further, the detection unit may be implemented in hardware. Further, there is no need to basically change the sensor architecture, and the detection unit may be easily integrated in an image sensor.
[0075] While embodiments of the invention have been described above, it is obvious that further embodiments may be implemented. For example, further embodiments may comprise any subcombination of features recited in the claims or any subcombination of elements described in the examples given above. Accordingly, this spirit and scope of the appended claims should not be limited to the description of the embodiments contained herein. LIST OF REFERENCES
[0076] 10 image sensor
[0077] 15 detection unit
[0078] 100 first frame
[0079] 105 result
[0080] 109 second frame
[0081] 110 first detector
[0082] 111 gradient veri fication device
[0083] 112 second detector
[0084] 113 average determination unit
[0085] 115 determination unit
[0086] 121 first tile
[0087] 122 second tile
[0088] 123 first variation
[0089] 124 flagged tile
[0090] 125 tile value
[0091] 126 surrounding tile
[0092] 127 surrounding gradient
[0093] 150 grid tile value
[0094] 151 first grid tile value
[0095] 152 second grid tile value
[0096] 153 reduced grid tile value
[0097] S 100 determining a change
[0098] S 110 determining gradients
[0099] S 114 determining a gradient module
[0100] S 117 confirming a flagged tile
[0101] S 120 determining that an event is present
[0102] S 130 determining an average
[0103] S200 calculation of di f ference
[0104] 5206 reduction processing
[0105] 5207 determination of average
[0106] 5210 content acquisition with context A
[0107] 5211 determination whether event has been detected
[0108] 5212 context switch and content acquisition with context B
Claims
CLAIMS1. A method for detecting an event in a sequence comprising at least a first frame (100) and a second frame (109) , each of the first and second frames (100, 109) comprising a plurality of tiles (121, 122) , each of the tiles (121, 122) comprising a plurality of pixels (120) , the method comprising : determining a change (S100) between the first and the second frame (100, 109) comprising determining at least one flagged tile (124) , a first variation (123) being associated to the flagged tile (124) ; determining (S110) , for each of flagged tiles (124) , a set of surrounding gradients (127) corresponding to a change of adjacent tiles (126) surrounding the flagged tile (124) between the first and the second frame (100, 109) ; determining (S114) , for each of the flagged tiles (124) , a gradient module corresponding to a vector of differences between the first variation (123) and the set of surrounding gradients (127) ; confirming (S117) , for each of the flagged tiles (124) , the flagged tile (124) if a norm of the gradient module is higher than a first threshold; and determining that an event is present based on a number of confirmed flagged tiles (124) .
2. The method according to claim 1, wherein determining at least one flagged tile (124) comprises: dividing the first and the second frame (100, 109) into a plurality of first tiles (121) having a first size; determining grid tile values (150) based on a difference between tile values of the first tiles (121) of the first frame (100) and tile values of the first tiles (121) of the second frame (109) , andflagging those of the plurality of first tiles (121) , for which the difference exceeds a second threshold.
3. The method according to claim 2, wherein determining the grid tile values (150) comprises: determining first grid tile values (151) corresponding to a difference between first tile values of the first tiles(121) of the first frame (100) and first tile values of the first tiles (121) of the second frame (109) ; determining second grid tile values (152) corresponding to a difference between second tile values of second tiles(122) of the first frame (100) and second tile values of second tiles (122) of the second frame (109) , wherein the second tiles (122) have a second size larger than the first size, and determining the grid tile values based on an average of the first grid tile values and the second grid tile values.
4. The method according to claim 3, wherein determining the second grid tile values (152) comprises: dividing the first and the second frame (100, 109) into a plurality of the second tiles (122) having the second size, and determining a difference between tile values of the second tiles (122) of the first frame (100) and tile values of the second tiles (122) of the second frame (109) .
5. The method according to claim 3, wherein determining the second grid tile values (152) is performed on the basis of the first grid tile values (151) .
6. The method according to any of claims 3 to 5, wherein determining the grid tile value comprises downsamplingcomprising determining an average of first grid tile values(151) with a corresponding second grid tile value (152) .
7. The method according to any of claims 3 to 5, wherein determining the grid tile value comprises upsampling comprising reducing a number of first grid tile values (151) to determine a reduced grid tile value (153) and determining an average of the reduced grid tile value (153) with a corresponding second grid tile value (152) .
8. The method according to any of claims 3 to 7, wherein determining the first grid tile values (151) is performed parallel to determining the second grid tile values (152) .
9. A computer program product comprising instructions which, when the program is executed by a computer, cause the computer to carry out the method according to any of the preceding claims.
10. A detection unit (15) for detecting an event in a sequence comprising at least a first frame and a second frame (100, 109) , each of the first and second frames (100, 109) comprising a plurality of tiles (121, 122) , each of the tiles (121, 122) comprising a plurality of pixels (120) , the detection unit (15) comprising: a first detector (100) configured to determine a change between the first and the second frame (100, 109) comprising determining at least one flagged tile (124) , a first variation (123) being associated to the flagged tile (124) ; a gradient verification device (111) configured to: determine, for each of flagged tiles (124) , a set of surrounding gradients (127) corresponding to a change of adjacent tiles (126) surrounding the flagged tile (124) between the first and the second frame (100, 109) ;determine, for each of the flagged tiles (124) , a gradient module corresponding to a vector of differences between the first variation (123) and the set of surrounding gradients (127) ; to confirm, for each of the flagged tiles (124) , the flagged tile (124) if a norm of the gradient module is higher than a first threshold; and a determination unit (115) configured to determine that an event is present based on a number of confirmed flagged tiles (124) .
11. The detection unit (15) according to claim 10, wherein the first detector is configured to: divide the first and the second frame (100, 109) into a plurality of first tiles (121) having a first size; determine grid tile values based on a difference between tile values of the first tiles (121) of the first frame (100) and tile values of the first tiles (121) of the second frame (109) , and flag those of the plurality of first tiles (121) , for which the difference exceeds a second threshold.
12. The detection unit (15) according to claim 11, further comprising a second detector (112) and an average determination unit (113) , wherein the first detector (110) is configured to determine first grid tile values (151) corresponding to a difference between first tile values of the first tiles (121) of the first frame (100) and first tile values of the first tiles (121) of the second frame (109) ; the second detector (112) is configured to determine second grid tile values (152) corresponding to a difference between second tile values of second tiles (122) of the first frame (100) and second tile values of second tiles (122) ofthe second frame (109) , wherein the second tiles (122) have a second size larger than the first size, and the average determination unit (113) is configured to determine the grid values based on an average of the first grid tile values (151) and the second grid tile values (152) .
13. The detection unit (15) according to claim 12, wherein the second detector (112) is configured to: divide the first and the second frame (100, 109) into a plurality of the second tiles (122) having the second size, and determine a difference between tile values of the second tiles (122) of the first frame (100) and tile values of the second tiles (122) of the second frame (109) .
14. The detection unit (15) according to claim 12, wherein the second detector (112) is configured to determine the second grid tile values (152) on the basis of the first grid tile values (151) .
15. An image sensor (10) comprising the detection unit (15) according to any of claims 10 to 14.
16. A method for operating the image sensor (10) according to claim 15 comprising: acquiring content (S210) with a first context; performing the method according to any of claims 1 to 9; switching the context to a second context if an event is present; and acquiring content (S212) with the second context.