WO2016000704A1

WO2016000704A1 - Method and device for the adaptive quantisation of camera images

Info

Publication number: WO2016000704A1
Application number: PCT/DE2015/200331
Authority: WO
Inventors: Markus Friebe; Johannes PETZOLD
Original assignee: Conti Temic Microelectronic Gmbh
Priority date: 2014-07-03
Filing date: 2015-06-01
Publication date: 2016-01-07
Also published as: DE102014212886A1; DE112015001273A5

Abstract

The invention relates to a method and device for the adaptive quantisation of camera images (KB) produced by a camera (1) provided with a lens (2), wherein the image video data, generated by the camera (1), of the camera images (KB) have image data blocks (BB) that each undergo a discrete cosine transformation (DCT) to calculate DCT coefficients which are divided and then rounded by a quantisation matrix (QM) for the quantisation thereof, said quantisation matrix (QM) used for the quantisation of the DCT coefficients being adaptively adjusted to the lens (2) of the camera (1).

Description

METHOD AND DEVICE FOR THE ADAPTIVE QUANTIZATION OF CAMERA IMAGES

The invention relates to a method and apparatus for adaptively quantizing camera images provided by a camera equipped with a lens, and more particularly to adaptively quantizing camera images provided by a camera equipped with a fisheye lens.

Driver assistance systems such as lane departure warning or parkas ^¬ sistenten with 360 ° visibility are increasingly used in Fahrzeu ^¬ gen, in particular road vehicles. Especially for the visibility in the vicinity of wide-angle cameras are used herein ^¬ with fish-eye lenses having a field of view of up to 250 °. The images produced by the cameras ^¬ camera comprises image video data composed of image data blocks, and which can be transferred sequentially. The compression of the image data can have several processing steps. After a color space conversion low-pass filtering and sub-sampling of Farbabwei ^¬ monitoring signals carried. The image data blocks, combine the nxn image pixels around ^¬ are subjected to a discrete cosine transform. Each image data block is quantized using a quantization matrix.

A disadvantage of conventional methods is that the quantization matrix used does not take into account the intrinsic distortion properties of the camera lens used by the camera, and this particularly in camera ^¬ lenses with high radial distortion, especially cameras with fish-eye lenses, leads to a significant reduction in image quality. It is therefore an object of the present invention to provide a method and an apparatus for adaptive quantization of camera images, in which the image quality of the transmitted camera images is increased.

This object is achieved by a method with the features specified in claim 1.

The invention accordingly provides a method for adaptively quantizing camera images provided by a camera equipped with a lens,

wherein the image data of the video camera images image data blocks generated by the camera comprise, each subjected to a discrete cosine transform for the calculation of DCT coefficients divided to their quantization by a quantization and then rounded to ^¬,

wherein the quantization matrix used to quantize the DCT coefficients is adaptively adapted to the lens of the camera.

In a possible embodiment of the inventive method, the image data blocks of the Bildvi ^¬ deodaten nxn image pixels produced have on the quantization matrix corresponding to quantization nxn, where n is a number natuer ^¬ Liche.

In a further possible embodiment of the inventive method, the quantization values of the quantization ^¬ sierungsmatrix depending on the intrinsic characteristics of the lens distortion of the camera are adjusted adaptively. In a further possible embodiment of the inventive method, the quantization values of the quantization ^¬ sierungsmatrix be adjusted such that the DCT coefficients of each transformed image data block are quantized increasingly stronger the more the DCT coefficients are within the camera image.

In a further possible embodiment of the method according to the invention, a quantization matrix suitable for the respective image data block is read from a table of quantization matrices stored in a data memory.

The invention further provides an apparatus for the adaptive quantization of camera images with the features specified in claim 6.

The invention accordingly provides an apparatus for adaptively quantizing camera images originating from a camera equipped with a lens, wherein the image video data of the camera images generated by the camera comprise image data blocks, each by a transformation unit for calculating discrete DCT coefficients

Cosine transform are quantized, which are quantized for their quantization by a quantization unit by means of a quantization matrix, which is adaptively adaptable to the lens of the camera.

In one possible embodiment of the device according to the invention, the image data blocks of the generated image video data comprise n × n image pixels and the quantization matrix. Speaking nxn on quantization values, wherein n is a number natuer ^¬ Liche.

In one possible embodiment of the device according to the invention, the quantization values of the quantization matrix ^¬ depending on the intrinsic characteristics of the lens distortion of the camera are adjusted adaptively.

In a further possible embodiment of the device according to the invention this has a data memory which stores a table with different quantization matrices, wherein a form suitable for ei ^¬ NEN image data block of the transmitted camera image quantization matrix is read from the table by the quantization unit.

In a further possible embodiment of the device according to the invention, the quantization values of the read out from the data memory quantization matrices are set so that the DCT coefficients of the respective transfor ^¬-optimized image data block quantized increasingly stronger ^¬ the further they are the, inside the respective camera image.

The invention further provides a camera having the features specified in claim 11.

Accordingly, the invention provides a camera that is equipped with a Whether ^¬ objectively, and provides the camera images, which are quantized by a device for adaptive quantization, wherein in the camera adaptive to an apparatus for

Quantization of camera images is provided, the camera images, which come from a lens of the camera and image data blocks, by a transformation unit for calculating DCT coefficients of a discrete one

Cosine transformation undergoes which to her

Quantization by a quantization unit using

Quantization matrix are quantized, which adaptively the lens of the camera is adaptable.

In one possible embodiment of the invention mera lens is a fisheye lens with a Bil angle of more than 185 °.

In a further possible embodiment of the camera according to the invention the camera is a camera of a vehicle driving ^¬ tool.

In a further possible embodiment of the camera according to the invention, the vehicle camera is attached to a surround-view system of a driver assistance system of the vehicle ^¬ closed.

The invention also provides a medical device with the given in Pa ^¬ tentanspruch 15 characteristics.

The invention accordingly provides a medical device having at least one camera equipped with a lens, which delivers camera images that are quantized by an adaptive quantization device, wherein the image video data of the camera images generated by the camera comprise image data blocks, each by a transformation unit for calculating DCT Coefficients of a discrete

Cosine transform are quantized, which are quantized for their quantization by a quantization unit by means of a quantization matrix, which is adaptively adaptable to the lens of the camera of the medical device. In the following, possible embodiments of the device according to the invention and of the method according to the invention for the adaptive quantization of camera images will be explained in more detail with reference to the attached figures.

Show it:

FIG. 1 shows a schematic view of a coding and decoding process in which the method according to the invention for the adaptive quantization of camera images can be used;

Figure 2 shows a block diagram of a possible exporting ^¬ approximately embodiment of the device according to the invention for adaptive quantization of camera images;

Figure 3 shows schematically and exemplarily a conventional procedure for the quantization of a Ka ^¬ merabildes;

FIG. 4 shows schematically and by way of example the procedure for the quantization of the camera image in the method according to the invention.

As can be seen from FIG. 1, in one possible embodiment of the method according to the invention, a sequence of camera images KB is first generated by a camera. Each camera image KB includes image video data composed of image data blocks BB. Each image data block BB preferably comprises n × n image pixels, for example 4 × 4 image pixels or 8 × 8 image pixels. In a possible embodiment, a color space conversion can be carried out in a step S2. follow, for example, from a so-called RGB color space in a YCbCr color space.

In a further step S3, a low-pass filtering and a sub-sampling, for example by simple means ^¬ value formation, are performed.

In a further step S4, the image data blocks BB are subjected to a discrete cosine transformation DCT for calculating DCT coefficients. The DCT transformation can be implemented using the fast Fourier transform with relatively little computational effort. The DCT transform is an orthogonal transform with good energy compression properties.

In a further step S5, the DCT coefficients formed are divided by a quantization matrix QM for their quantization and then rounded. In the he ^¬ inventive method, the quantization matrix QM used for quantizing the DCT coefficients, it is adaptive to the camera lens, which genes ^¬ riert the camera image KB adjusted.

In a further step S6, a coding, in particular an entropy coding, can take place by means of coding tables. Subsequently, a decoding on the receiver side in the reverse direction, i. E. in step S7 an entropy decoding and in step S8 a De- or

Requantization in step S9 an inverse discrete

Cosine transformation IDCT, in step S10 an oversampling and low-pass filtering of the color difference signals, in step S11 a color space conversion in a target color space and

Finally, in step S12, a provision of the camera of the KB, in particular for further data processing by a data processing unit of a surround view system. As can be seen in FIG. 1, in one possible embodiment the quantization tables used become

Dequantization provided in step S8. Further, the encoding tables used in step S9 may be provided for decoding the transmitted data.

In the inventive method for adaptive

Quantization of the camera images KB, the DCT coefficients are divided for their quantization by a quantization matrix QM and then rounded. These

Quantization is done in step S4. For this purpose, for each image data block BB of the camera image KB to be transmitted, a quantization matrix QM is used which is adaptively adapted to the lens of the camera which supplies the camera image KB. For this purpose, the quantization matrix QM used for this purpose is preferably read from a look-up table which is stored in a data memory provided for this purpose. The quantization values q a quantization matrix QM are adaptively adjusted in dependence on the intrinsic properties of the respective distortion ^¬ lens of the camera or set. The quantization values Q of the Quan ^¬ tisierungsmatrix QM are preferably adjusted such that the DCT coefficients of each transformed image data block BB are increasingly more quantized, the more the DCT coefficients lie in the interior of the camera image KB. By adapting and / or selecting the quantization matrix QM, the intrinsic distortion properties of the respective camera or the camera lens contained therein can be taken into account. In this way, ei ^¬ ne better image quality can be achieved when the same bit rate is used, or the image quality may at low ger bitrate to be kept at the same level of quality. When the lens is a fisheye lens, the quantization matrix QM may be designed to quantize the discrete cosine transform coefficients of the small fimed circle inner portions of the fisheye circle rather than the larger radius farther away from the center of the circle lie. In this way, one obtains a larger image quality or it is sufficient to have a low data transmission rate at the same ^¬ image quality.

FIG. 2 shows a block diagram for illustrating an ^exemplary embodiment of an arrangement according to the invention. As you can see in Figure 2, a camera 1 provides with an included camera lens 2 camera images KB of a transformation unit 3. The camera images KB have Bildvi ^¬ deodaten, which are composed of image data blocks BB. By the transformation unit 3, the image data blocks ^¬ BB be that best ^¬ hen subjected example of nxn image pixels for the computation of DCT coefficients of a discrete cosine transform. The calculated DCT coefficients are then quantized to their quantization by a quantization unit 4 by a quantization matrix QM ^¬ approximately, which is adaptively matched to the lens 2 of the camera. 1 To this end, the quantization matrix QM ^¬ is read out in a possible embodiment of a data storage. 5 In the data memory 5, there is preferably a table in which different Quan ^¬ tisierungsmatrizen QM are saved or stored, wherein the quantization unit 4 for the DCT coefficients of the various image data blocks BB a respectively suitable Quan ^¬ tisierungsmatrix QM stored in the look-up Can read table. In one possible embodiment form the Quantization unit 4 and its data memory 5 together a device 6 for adaptive quantization of the camera images KB. In one possible embodiment, the quantum tisierten data is then transmitted through an encoder 7 ko ^¬ diert and to a receiving unit for decoding.

The camera 1 shown in FIG. 2 may, for example, in one possible embodiment, be a vehicle camera of a vehicle. In one possible ^embodiment , the objective 2 of the camera 1 is a fisheye lens with an angle of view of more than 185 °. In one possible embodiment, the on-vehicle camera 1 is connected to a surround-view system of a driver assistance system of the driving ^¬ tool which receives the encoded data by the encoding unit 7.

The inventive device 6 for adaptive

Quantization of camera images is particularly suitable for the quantization of camera images KB, which originate from a fisheye lens. The fisheye lens, it may be a fisheye lens of a given type, in particular an angle-abiding, a linear split, a sur fa ^¬ chentreues or orthographically fisheye lens. In one possible embodiment, the quantization matrix QM selected for the quantization is selected as a function of the lens type of the objective 2. In one possible embodiment, different tables of quantization matrices QM within the data memory 5 are configured or stored for different types of lenses. The configura ^¬ tion or setting of the quantization matrices QM depending on the camera type of the camera 1 and / or type of the objective lens 2 is performed in one possible embodiment of a data interface of the Vorrich- shown in Figure 2 In an alternative embodiment, the set of quantization matrices QM suitable for the camera 1 or the object type of the objective 2 can be read from a data memory of the camera 1 via an interface provided therefor and transmitted to the quantization unit 4 of the ^device 6 become. In this embodiment, the data memory for storing at least one table of quantization matrices QM in a data memory of the camera 1. In a further alternative embodiment, the camera 1 via an interface the Kame ^¬ Ratyp and / or the object type used to the device 6 for adaptive Quantization report, which is a suitable set of quantization matrices QM in the

Quantization used.

FIG. 3 schematically shows the quantization of image data blocks BB using a conventional procedure. The camera image KB generated by the camera 1 has an image width B and an image height H. The camera image KB consists of a plurality of image data blocks BB, each comprising n × n image pixels. In the conventional procedure shown schematically in Figure 3, the quantization is uniform for the entire camera image. In this case, the same quantization matrix QM with the quantization matrix number 1 is always used.

In contrast, Figure 4 illustrates the inventive pre ^¬ hens, in the quantization of the image data blocks BB. In the illustrated example in Figure 4, the generated Ka ^¬ merabild KB is using quantified by five different quantization matrices ^¬ QM1 to QM5. In this case, the ^¬ located in corners of the camera image KB four image blocks with a quantization QM1 be quantified example, which can be read from a table of the data memory 5. The image data blocks BB located in the center or in the camera image center are quantified in the illustrated example with a quantization matrix QM5. The quantization ^¬ rungsmatrizen QM preferably be sequentially read out from the space A, since ^¬ 5 for quantizing the calculated DCT coefficients. These DCT coefficients are divided by the information contained in the read quantization matrix QM quantization values q and then Gerun ^¬ det. In the example shown in Figure 4, the quantization unit 4 receives, for example, initially three Bildda ^¬ tenblöcke BB (in the first row), which are quantified in each case with the Quan ^¬ tisierungsmatrix QM1. For which the quantization QM2 is used then fol ^¬ gen three image data blocks. Subsequently, four image data blocks BB are the quantization QM1 quantified (the last three of the first row and the first image data block in the second row), followed by two image data blocks which are quantified with the quantization QM2 and then three image data blocks obtained by the quantization ^¬ approximate matrix be quantified QM3 etc. Each of the various five quantization QM1, QM2, QM3, QM4, QM5 environmentally summarizes nxn quantization values q, which are individually Fit to ^¬ bar. In one possible embodiment, the quantization values q the quantization matrix QM within the table depending on the intrinsic distortion characteristic of the lens 2 of the camera 1 may be adaptively inserted ^¬ represents.

In one possible embodiment, the order of the image data processing of the different image data blocks BB can be changed. For example, in one possible embodiment, the images in the middle may first be displayed. data blocks, for example with the quantization matrix QM5 and a quantization matrix QM4, and then the peripheral image data blocks BB of the camera image KB.

If the lens 2 of the camera 1 is replaced, the quantization matrices QM in the data memory 5 can be changed or reconfigured in the device 1 according to the invention. The quantization values q of quantization ^¬ rungsmatrizen QM are set such that the DCT coefficients of each transformed image data block are increasingly stronger quantitated the more the DCT coefficients are located in the interior of the respective camera image KB. The arrangement shown in Figure 2 may be used, for example, in a surround-view system of a driver assistance ^¬ tenzsystems.

Claims

1. A method for adaptive quantization of camera images (KB) supplied by a camera (1) equipped with a lens (2),

wherein the image video data of the camera images (KB) generated by the camera (1) comprises image data blocks (BB) each subjected to a discrete cosine transformation (DCT) for calculating DCT coefficients, which are quantized by a quantization matrix

(QM) divided and then rounded,

wherein the quantization matrix (QM) used to quantize the DCT coefficients is adaptive to the objective

(2) the camera (1) is adjusted.

2. The method according to claim 1,

wherein the image data blocks (BB) of the Bildvideoda ^¬ th nxn image pixels produced have, and

wherein the quantization matrix (QM) has n x n quantization values (q),

where n is a natural number.

3. The method according to claim 2,

wherein the quantization values (q) of the Quantisierungsmat ^¬ rix (QM) as a function of the intrinsic distortion characteristics of the lens (2) of the camera (1) can be set ^¬ adap tive.

4. The method according to claim 3,

wherein the quantization values (q) of the Quantisierungsmat ^¬ rix (QM) can be adjusted such that the DCT coefficients of each transformed image data block are quantized increasingly stronger, the further the DCT coefficients are inside the camera image (KB).

5. The method according to any one of the preceding claims 1 to 4, wherein one of the respective image data block (BB) appropriate quantization matrix (QM) from a data in a memory (5) stored table of Quantisierungsmat ^¬ Rizen is read out.

Have 6. A device for adaptive quantization of camera images (KB), which are equipped by one with a lens (2) camera (1) originate, said by the camera (1) erzeug ^¬ th frame video data of the camera images (KB) image data blocks (BB) which are in each case subjected by a transformation ^¬ unit (3) for the computation of DCT coefficients of a discrete cosine transform (DCT), which are to their quantization by a quantization ^¬ unit (4) by means of a quantization matrix (QM) quantum tisiert which adaptively adaptable to the lens (2) of the camera (1).

7. Apparatus according to claim 6,

wherein the quantization matrix (QM) has n x n quantization values (q),

where n is a natural number.

8. Apparatus according to claim 7,

wherein the quantization values (q) of the Quantisierungsmat ^¬ rix (QM) as a function of the intrinsic distortion characteristics of the lens (2) of the camera (1) are adap tive ^¬ adjustable.

9. Device according to one of claims 6 to 8 with a data memory (5) which stores in a table different quantization matrices (QM) which are readable by the quantization unit (4).

10. Apparatus according to claim 9,

wherein the quantization values (q) from the Datenspei ^¬ cher readable quantization matrices (QM) are set such that the DCT coefficients of each transformed image data block increasingly more quantum tisiert be farther they (KB) are in the interior of the respective camera image.

A camera (1) equipped with a lens (2) and providing camera images (KB) quantized by an adaptive quantization apparatus according to any one of the preceding claims 6 to 10.

12. Camera according to claim 11,

wherein the lens (2) is a fisheye lens with an angle of view of more than 185 °.

13. Camera according to claim 11 or 12,

wherein the camera (1) is a vehicle camera of a vehicle.

14. Camera according to claim 13,

wherein the vehicle ^camera is connected to a surround view system ei ^¬ nes vehicle assistance system of the vehicle. Medical device with at least one with a lens (2) equipped camera (1), which camera images (KB) be ^¬ fert, which by a device for adaptive

Quantization according to one of the preceding claims 6 to 10 are quantized.