The invention relates to a method and apparatus for glare de ^¬ tection within an image provided by a camera, in particular by a camera of a driver assistance system.

In an automatic system that makes use of visual data, the im ^¬ age captured by a camera does not always correspond to the surrounding. This can occur due to media on the camera lens which obstruct the lens completely or at least partially or due to glare caused by sunlight or headlights of other ob- j ects .

A glare creates a halolike effect around a light source which prevents a camera from capturing what is behind the halo. Since an approaching vehicle, pedestrian or object may be invisible due to glare, the driver or a driver assistance sys ^¬ tem must be alert since some areas of the surrounding are not covered by the captured image. Further, glare regions in the captured image may cause subsequent processes of a driver as ^¬ sistance system to fail or to give false detection.

Figures 1A, IB show a situation with and without glare. The light source in the given example is the sun standing over a roof of a house. Figure 1A shows a situation without glare where the bright light source (the inner circle) does not ob ^¬ struct neighboring details. In Figure IB, a glare halo is shown around the light source obstructing details of the house .

In conventional methods, a glare is simply detected as image pixels that are maximally saturated. For example, with a cam ^¬ era that represents luminance with 8 bits, pixels having a luminance equal to 255 are maximally saturated and are de ^¬ tected to form a glare within the captured image. However, the conventional method is not very reliable, because a glare can also originate in an image area where the image pixels are not maximally saturated.

Accordingly, it is an object of the present invention to pro ^¬ vide a method and apparatus which can reliably detect a glare within a captured image.

This object is achieved by a glare detection method compris ^¬ ing the features of claim 1.

The invention provides according to a first aspect a glare detection method for detection of at least one glare region within an image,

said method comprising the steps of:

aggregating image pixels of the image having a high luminance intensity into bright image areas within said image,

calculating for image pixels around each bright image area gradients expected in case of a glare and actual gradients and

increasing a glare diameter of a glare region around the re ^¬ spective bright image area as long as the calculated actual gradients match the calculated expected gradients.

In a possible embodiment of the glare detection method ac ^¬ cording to the first aspect of the present invention, simi ^¬ larity metrics between the calculated actual gradients and the calculated expected gradients are computed.

In a further possible embodiment of the glare detection meth ^¬ od according to the first aspect of the present invention, an average similarity value of the computed similarity values for image pixels being located in a neighboring image area around the respective bright image area and being equidistant to the respective bright image area is calculated.

In a still further possible embodiment of the glare detection method according to the first aspect of the present inven ^¬ tion, an average similarity value is calculated with a step ^¬ wise increased distance between the equidistant image pixels of the neighboring image area and the respective bright image area until the respective calculated average similarity value becomes smaller than a predetermined threshold value, wherein the equidistant image pixels of the last incrementing step define the outer boundary of the glare region within said image .

In a further possible embodiment of the glare detection meth ^¬ od according to the first aspect of the present invention, the image forms part of an image sequence of images provided by a camera.

In a further possible embodiment of the glare detection meth ^¬ od according to the first aspect of the present invention, a glare is detected if a glare region is detected around a bright image area for a predetermined number of images in the image sequence provided by the camera.

In a still further possible embodiment of the glare detection method according to the first aspect of the present inven ^¬ tion, the image is generated by a camera of a vehicle.

In a further possible embodiment of the glare detection meth ^¬ od according to the first aspect of the present invention, the image provided by the camera is a digital image compris ^¬ ing a plurality of pixels each having a luminance intensity.

In a further possible embodiment of the glare detection meth ^¬ od according to the first aspect of the present invention, the image is downscaled and the downscaled image is scanned for image pixels having a high luminance intensity above a predetermined luminance intensity threshold.

In a still further possible embodiment of the glare detection method according to the first aspect of the present inven ^¬ tion, the image pixels of the downscaled image having a high luminance intensity above the predetermined luminance inten ^¬ sity threshold are aggregated into connected areas labelled as bright image areas of the image.

The invention further provides a glare detection apparatus comprising the features of claim 11.

The invention provides according to a second aspect a glare detection apparatus for detection of at least one glare re ^¬ gion within an image,

said apparatus comprising:

a processing unit being configured to aggregate image pixels of said image having a high luminance intensity to bright im ^¬ age areas within said image and to calculate for image pixels around each bright image area gradients expected in case of a glare and actual gradients and being configured to increase a glare diameter of a glare region around the respective bright image area as long as the calculated actual gradients match the calculated expected gradients. In a possible embodiment of the glare detection apparatus ac ^¬ cording to the second aspect of the present invention, the apparatus is further configured to compute similarity metrics between the calculated actual gradients and the calculated expected gradients and is configured to calculate an average similarity value of the computed similarity values for image pixels being located in a neighboring image area around the respective bright image area and being equidistant to the re ^¬ spective bright image area.

In a further possible embodiment of the glare detection appa ^¬ ratus according to the second aspect of the present inven ^¬ tion, the processing unit is configured to calculate an aver ^¬ age similarity value with a stepwise increased distance be ^¬ tween the equidistant image pixels of the neighboring image area and the respective bright image area until the respec ^¬ tive calculated average similarity value becomes smaller than a predetermined threshold value, wherein the equidistant im ^¬ age pixels of the last incrementing step define the outer boundary of the glare region within said image.

In a further possible embodiment of the glare detection appa ^¬ ratus according to the second aspect of the present inven ^¬ tion, the apparatus further comprises a downscaling unit adapted to downscale the image,

wherein said downscaled image is scanned by the processing unit for image pixels having a high luminance intensity above a predetermined luminance intensity threshold, wherein image pixels of the downscaled image having a high luminance inten ^¬ sity above the predetermined luminance intensity are aggre ^¬ gated by the processing unit into connected areas labelled as bright image areas of the image. The invention further provides a driver assistance system of a vehicle comprising the features of claim 15.

The invention provides according to a third aspect a driver assistance system of a vehicle comprising:

at least one camera configured to provide images of the vehi ^¬ cle's surrounding and

a glare detection apparatus according to the second aspect of the present invention.

In the following, possible embodiments of the different as ^¬ pects of the present invention are described in more detail with reference to the enclosed figures.

Figures 1A, IB show diagrams illustrating a problem underlying the present invention;

Figure 2 shows a flowchart of a possible exemplary em ^¬ bodiment of a glare detection method accord ^¬ ing to the first aspect of the present inven ^¬ tion;

Figure 3 shows a further flowchart of a possible em ^¬ bodiment of a glare detection method accord ^¬ ing to the first aspect of the present inven ^¬ tion;

Figure 4 shows a block diagram of a possible exemplary embodiment of a glare detection apparatus ac ^¬ cording to the second aspect of the present invention ; Figure 5 shows a diagram for illustrating the opera ^¬ tion of a method and apparatus according to the present invention;

Figure 6 shows a similarity metric as a function of an angle difference illustrating a possible im ^¬ plementation of a method and apparatus ac ^¬ cording to the present invention; shows a diagram for illustrating a possible exemplary implementation of the method and apparatus according to the present invention;

Figure 8 shows a further diagram for illustrating a possible exemplary implementation of a method and apparatus according to the present inven ^¬ tion.

As can be seen in Figure 2, the glare detection method ac ^¬ cording to the first aspect of the present invention can com ^¬ prise several steps SA to SC as illustrated in Figure 2. The glare detection method according to the present invention as shown in Figure 2 is provided for detection of at least one glare region within an image. The image can be a digital im ^¬ age captured by a camera of a system. In a possible embodi ^¬ ment, the image is a digital image captured by a vehicle cam ^¬ era of a driver assistance system. In a possible embodiment, the camera can be a fisheye camera of a surround view system of a vehicle. The image can be taken by the camera from the front, back, left or right side of the vehicle. In a first step SA, image pixels of the captured image having a high luminance intensity are aggregated into bright image areas BIA within said image.

In a further step SB, for image pixels around each bright im ^¬ age area BIA gradients expected in case of a glare and actual gradients are calculated.

In a further step SC, a glare diameter of a glare region around the respective bright image area BIA is increased as long as the calculated actual gradients match the calculated expected gradients.

In a possible embodiment, of the glare detection method simi ^¬ larity metrics between the calculated actual gradients and the calculated expected gradients are computed in step SC. Further, an average similarity value of the computed similar ^¬ ity values for image pixels being located in a neighboring image area NIA around the respective bright image area BIA and being equidistant to the respective bright image area BIA can be calculated. The average similarity value is calculated with a stepwise increasing distance between the equidistant image pixels IP of the neighboring image area NIA and the re ^¬ spective bright image area BIA until the respective calculat ^¬ ed average similarity value becomes smaller than a predeter ^¬ mined threshold value. The equidistant image pixels IP of the last incrementing step define the outer boundary of the glare region within the captured image.

Figure 3 shows a flowchart of a possible exemplary embodiment of a glare detection method according to the first aspect of the present invention. In a first step SI, an image is captured by a camera. The im ^¬ age may comprise for instance 1280x800 image pixels.

In a further step S2, the image is downscaled and the

downscaled image is then scanned for image pixels having a high luminance intensity above a predetermined luminance in ^¬ tensity threshold. This is done to reduce computational com ^¬ plexity and to reduce the effect of image noise. For example, the captured image can be reduced by a factor 15 in each di ^¬ rection, i.e. the downscaled image comprises a size being smaller by a factor of 15 when compared to the original cap ^¬ tured image in each direction. If the original captured image comprises a size of 1280x800 image pixels the downscaled im ^¬ age comprises a size of 85x53 image pixels as illustrated in Figure 3.

The downscaled image is then scanned in step S3 for pixels that have a very high intensity value, for example 255 in an 8 bit luminance image starting at zero.

These high intensity image pixels are then aggregated into bright image areas BIA within said image. In a possible em ^¬ bodiment, the image pixels of the downscaled image having a high luminance intensity above a predetermined luminance in ^¬ tensity threshold are aggregated into connected areas la ^¬ belled as bright image areas of the image. For each labelled bright image area BIA, the central pixel can be found in step S4, for example through a simple average of the position of the pixels in the respective area.

Since the gradient direction within glare regions is the same as an angle formed by the mean pixel of the area and the cur ^¬ rent image pixel as illustrated by the arrows in Figure IB, a set of expected gradient angles is computed in step S5 in the neighboring image area NIA of the bright image source. As can be seen in Figure 3, in another branch of the illustrated flowchart the downscaled image is loaded and an actual gradi ^¬ ent that is seen on the image is computed. Finally, given the gradient that is expected in case of a glare and the actual gradient in the captured image, a metric of similarity be ^¬ tween the two values can be calculated for circles that are progressively further away from the bright light source in step S5. Accordingly, with the method according to the pre ^¬ sent invention in the illustrated embodiment, for image pix ^¬ els IP around each bright image area BIA, gradients expected in case of a glare and actual gradients are calculated in calculating step S5.

Finally, a glare diameter of a glare region around the re ^¬ spective bright image area is increased as long as the calcu ^¬ lated actual gradients match the calculated expected gradi ^¬ ents in step S6. For a circular glare, for each radius, pro ^¬ gressively further away from the center of the bright light source a number given by the average similarity metrics of the image pixels in the respective circle is computed. If the similarity metric becomes too low, it is deemed that the oth ^¬ er boundary of the glare region has been reached. The average similarity value is calculated with a stepwise increased dis ^¬ tance between the equidistant image pixels IP of the neigh ^¬ boring image area NIA and the respective bright image area BIA until the respective calculated average similarity value becomes smaller than a predetermined threshold value wherein the equidistant image pixels IP of the last incrementing step define the outer boundary of the glare region within the re ^¬ spective image. In more detail, if a pixel (xo, yo) is the computed center of a high luminance intensity area and (x, y) is a pixel in the neighborhood as illustrated in Figure 5, a gradient that is expected for a glare pixel in (x, y) is

0 _O = tan ^-1 (y - y ₀ , x -x ₀ ) (D

The actual gradient of the captured image can be calculated by :

_ ₁ dl dl

^dy ' dx^

The computation of the similarity metric between the expected angle and the actual angle can comprise several substeps. In a first substep, the difference between the angles is comput ^¬ ed .

0 _A = min(|0-0o ⁷ r-|0-0 _o l)- (3)

Then, the angle difference can be used to compute a similari ^¬ ty metric ε [-1, 1], where 1 means that the values are iden ^¬ tical and -1 that the values are conflicting, as follows: e = max(l-^,-l). (4) A visual representation of the possible values of the simi ^¬ larity metric ε as a function of the difference between the angles is shown in Figure 6.

As illustrated in Figure 6, only gradients that are aligned with the respective gradient contribute positively to an as ^¬ sessment of whether the image pixels of a particular radius from the center of the bright image area constitute glare.

To assess if the image pixels IP at a particular radius r constitute glare, the following steps can be performed for a radius becoming progressively larger starting at 1. First, the image pixels IP at a particular radius from the center of the bright image area (xo, yo) are identified. Figure 7 il ^¬ lustrates such image pixels for a radius r=5 around the cen ^¬ ter pixel (x ₀ , yo) ·

To compute a similarity between the direction of the gradi ^¬ ents of the image pixels IP at that radius r and the expected gradient all similarity metrics of such image pixels can be added and divided by the number of pixels, i.e. a mean simi ^¬ larity metric for that radius r is calculated. If the mean similarity metric is above a certain threshold TH it is con ^¬ sidered that there is glare and the radius r is incremented and the next radius is checked. In a possible implementation, a similarity threshold of 0.5 can be used. If the mean simi ^¬ larity metric is below this threshold the limit or outer boundary of the glare region has been reached and the process stops. Because there is no information propagated along time, the detection of the glare according to the method of the present invention occurs instantaneously. In a possible em ^¬ bodiment, a glare is only detected if a glare region is de ^¬ tected around a bright image area BIA for a predetermined number N of images in an image sequence provided by a camera. In this embodiment, by requiring that glare detections are consistent along time, a glare is detected only if it is vis ^¬ ible around the same area for N frames.

With the glare detection method according to the present in ^¬ vention it is possible to detect circular shaped glare re ^¬ gions. In cases where the glare is not circular, an equiva ^¬ lent yet adapted process can be used. For example, if the glare exhibits a vertical streak the computation of the cen ^¬ ter (xo, yo) of the bright image area BIA and how equidistant points are found can be adapted. If Ώ is a set of very bright points or pixels, for any pixel (x, y) one can find a pixel

(xo, yo) ε Ώ that is closest to the respective pixel (x, y) . Starting from here, the process can proceed as described above using equation (1) to compute the expected angles, equation (2) to compute the actual angles and then equations

(3) and (4) to compute similarity metrics. At the end of the process instead of collecting the similarity metrics at a given radius from the center of the bright area, these simi ^¬ larity metrics are collected for pixels whose distance at any point in set Ώ is given by a radius r.

Figure 8 illustrates such equidistant image points IP of the set Ώ at a distance r=5. If the similarity metric for a given radius r is above a threshold the radius r is incremented and the next set of pixels is checked. This process continues un ^¬ til the similarity becomes smaller than a predetermined threshold value.

Figure 4 shows a block diagram of a possible exemplary embod ^¬ iment of a glare detection apparatus 1 according to a further aspect of the present invention. The glare detection appa- ratus 1 according to the present invention comprises in the illustrated embodiment a downscaling unit 2 and a processing unit 3. The downscaling unit 2 is adapted to downscale an im ^¬ age received from at least one camera 4 as shown in Figure 4. The camera 4 can be a camera of a surround view system of a vehicle. The camera 4 can be at the front, back, left or right side of the respective vehicle VEH. The downscaling unit 2 is adapted to downscale the received digital image. The glare detection apparatus 1 further comprises a pro ^¬ cessing unit 3 configured to aggregate image pixels of the image having a high luminance intensity to bright image areas BIA within said image. The processing unit 3 is further adapted to calculate for image pixels around each bright im ^¬ age area BIA gradients expected in case of a glare and actual gradients. The processing unit 3 is further configured to in ^¬ crease a glare diameter of a glare region around the respec ^¬ tive bright image area BIA as long as the calculated actual gradients match the calculated expected gradients. In a pos ^¬ sible embodiment, the processing unit 3 is configured to com ^¬ pute similarity metrics between the calculated actual gradi ^¬ ents and the calculated expected gradients and is configured to calculate an average similarity value of the computed sim ^¬ ilarity values for image pixels being located in a neighbor ^¬ ing image area NIA around the respective bright image area BIA and being equidistant to the respective bright image area BIA. In a possible embodiment, the processing unit 3 of the apparatus 1 is configured to calculate an average similarity value with a stepwise increased distance between the equidis ^¬ tant image pixels IP of the neighboring image area NIA and the respective bright image BIA area until the respective calculated average similarity value becomes smaller than a predetermined threshold value. The equidistant image pixels IP of the last incrementing step define then the outer bound ^¬ ary of the glare region within the received image.

In a possible embodiment, the processing unit 3 further pro ^¬ cesses the found glare region within the captured image. In a possible embodiment, the found glare region is filtered. In a further embodiment, another application program is informed about the detected glare region. In a still further possible embodiment, the processing unit 3 outputs a warning if a glare region in the captured image is detected.

The glare detection apparatus 1 as shown in Figure 4 can form part of a driver assistance system of a vehicle VEH. The driver assistance system comprises at least one camera 4 con ^¬ figured to provide an image of the vehicle's VEH surrounding and the glare detection apparatus 1 as shown in Figure 4.

The method and apparatus can also be employed in other sys ^¬ tems, in particular surveillance systems, manufacturing sys ^¬ tems and consumer electronic devices.