KR20100042632A - Video indexing method, and video indexing device - Google Patents

Abstract

The invention relates to a method and a device for indexing a coded video data stream. According to the invention,the video data stream comprises information relative to the location of regions of interest of each picture, said method comprises steps of: reception (T1) of coded video stream, recording the coded video stream on a recording support, decoding (T2) location information of regions of interest, selection (T3) of a region of interest per picture, decoding (T3) of video data, selecting (T4) a predetermined number of regions of interest for the video data stream from among the regions of interest selected per picture, recording (T6) of the selected regions of interest.

Description

Video indexing method, and video indexing device {VIDEO INDEXING METHOD, AND VIDEO INDEXING DEVICE}

The present invention relates to a video indexing method and a video indexing device.

Some picture processing applications use regions of interest (ROI) detection to improve picture quality. For example, coding applications often decode areas of interest and place more resources to code these areas.

Other methods enable detection of regions of interest in the image. In particular, methods based on the establishment of salience maps of an image or video taking into account visual parameters are known and enable the definition of areas where the human eye stays long when viewing an image or video.

Detection of regions of interest is used today in such a way as to privilege regions of interest during coding, primarily by giving more bandwidth to regions of interest before coding, for example by reducing quantization steps for these regions. do.

The advent of mobile terminals, such as mobile phones, PDAs, game consoles, portable DVD players, the development of display and screen technologies, and the emergence of new services all combine the display of video on a terminal with low display capacity. Made it necessary. For example, the possibility of receiving television on a mobile phone creates display problems for dense images on low dimension screens.

The present invention does not relate primarily to the detection of regions of interest, but rather to devices or applications that consider them for different applications and at least solve the video display problem on a terminal with low display capacity, whether mobile or not. Relates to the transmission of these areas of interest.

For this purpose, the present invention proposes a method for indexing coded video data streams. According to the invention, the video data stream comprises information regarding the location of the regions of interest of each image, the method comprising:

Receiving a coded video stream

Recording the coded video stream on a recording support

-Decoding the location information of the regions of interest

Selecting a region of interest for each image;

-Decoding the video data

Selecting a predetermined number of regions of interest for the video data stream from among the regions of interest selected per image;

Recording the selected regions of interest

It includes.

According to a preferred embodiment, during the recording step,

The selected regions of interest are recorded in temporary memory when they are selected and decoded,

When all of the selected regions of interest are recorded in the temporary memory, the selected regions of interest are transmitted to a permanent memory support 503.

Advantageously, prior to recording them the regions of interest are formatted to obtain a homogeneous size for all of the selected regions of interest.

Advantageously, the method comprises encrypting the location of said regions of interest with the aid of an encryption key.

Preferably, the method includes obtaining a decryption key in accordance with a payment by a user.

Advantageously, said video data stream is coded according to coding standard H.264 / AVC and said location information is included in a Supplemental Enhancement Information (SEI) type message.

According to a preferred embodiment, the SEI messages are encapsulated in real-time protocol (RTP) packets, and the RTP packets are encrypted.

Advantageously, said SEI type messages about regions of interest location information are inserted into coded data before or after each picture they refer to.

According to a preferred embodiment, the location information is:

The number of regions of interest of each image,

The coordinates of each region of interest for each image dimension,

The surface of each region of interest,

A weight relating to the importance of the region of interest with respect to other regions of interest of the image,

Information about the content of each region of interest,

And information selected from any combination of this information.

Preferably, selecting the region of interest for each image selects the region of interest according to the weights related to the importance of the region of interest.

Advantageously, the video coding standard uses flexible macro-bloc ordering (FMO), and the regions of interest are coded into slice groups, regardless of other image data, and location information of the regions of interest. Includes the slice group numbers in which the regions of interest are coded.

Advantageously, said SEI message includes an identifier for each slice group indicating whether it relates to one region of interest.

Advantageously, the method comprises the further step of reading said SEI messages and said step of decoding said video data decodes only slice groups comprising said region of interest.

The invention also relates to a device for indexing a coded video data stream. According to the invention, the video data stream comprises information relating to the location of the regions of interest of each image, the device comprising:

Means for receiving the coded video stream,

Means for recording the coded video stream on a recording support device 503,

Means (501) for decoding the location information of the regions of interest,

Means (501) for decoding the video data,

Means 502 for selecting a region of interest per image,

Means 502 for selecting a predetermined number of regions of interest for the video data stream from among the regions of interest selected per image,

Means 503 for recording the selected regions of interest

.

Detection of regions of interest of the image is generally done before coding. This data is then used to facilitate encoding. The inventors have realized that the location of the region of interest may also be important during decoding of the image and especially during display on devices with limited display capacity. In fact, the receiving terminal may choose to display only the regions of interest in nature, which may have better visibility of these regions as compared to the display of the complete image.

The invention will be further understood and illustrated by embodiments and implementations, which are by no means intended to be limiting with reference to the accompanying drawings.
1 illustrates a coding device according to a preferred embodiment of the present invention.
2 illustrates a coding method according to a preferred embodiment of the present invention.
3 illustrates a decoding device according to a preferred embodiment of the present invention.
4 illustrates a decoding method according to another embodiment of the present invention.
5 illustrates a personal recording type device according to another embodiment of the present invention.
FIG. 6 illustrates an indexing method in a personal recording type device implementing one embodiment of the present invention. FIG.

1 shows a coding device according to the coding standard H.264 / AVC, which implements a preferred embodiment of the invention. In this preferred embodiment, the video stream is coded.

The current frame F _n is provided at the coder input to be coded by it. This frame is coded in the form of slices, ie it is divided into sub-units each containing a certain number of macroblocks corresponding to groups of 16x16 pixels. Each macroblock is coded in intra or inter mode. Whether in intra mode or inter mode, the macroblock is coded by based on a reconstructed frame. The module 109 determines, in the intra mode of the current image, the coding mode according to the content of the image. In the intra mode, P (shown in FIG. 1) includes samples of the current frame Fn that was previously coded, decoded, and reconstructed (uF'n in FIG. 1, u means unfiltered). In inter mode, P is constructed from motion estimation based on one or more F ′ _n−1 frames.

The motion estimation module 101 establishes an estimate of the motion between the current frame Fn and at least one preceding frame F′n-1. From this motion estimation, motion compensation module 102 generates frame P when the current picture Fn must be coded in inter mode.

The subtractor 103 generates a signal Dn, which is the difference between the image Fn and the image P to be coded. This image is then converted by DCT conversion in module 104. The converted image is then quantized by quantization module 105. The images are then reorganized by the module 111. Context-based Adaptive Binary Arithmetic Coding (CABAC) type entropy coding module 112 then codes each image.

Modules 106 and 107 of quantization and inverse transformation, respectively, allow the difference D'n to be reconstructed after inverse quantization and inverse transformation after transform and quantization.

When an image is coded in intra mode, according to module 109, intra prediction module 108 codes the image. The uF'n image is obtained at the adder output 114 as the sum of the D'n signal and the P signal. This module 108 also receives a reconstructed unfiltered F'n image at the input.

The filter module 110 may obtain a reconstructed and filtered F′n image from the uF′n image.

Entropy decoding module 112 transmits the coded slices encapsulated in NAL type units. NALs contain information about the headers, as well as slices, for example. NAL type units are sent to module 113.

Module 116 allows areas of interest to be determined. Some methods now allow regions of interest to be placed in the image. In particular, methods based on the establishment of protrusion maps are known.

For example, the patent application WO2006 / 07263, filed on January 10, 2006 and published on July 13, 2006, discloses an effective method for establishing protrusion maps.

The means 116 then establishes a salient map for each image of the video. In order to establish this protrusion map, the parameters entered by the user can also be considered. For example, according to the event with which the video is related, it is possible to define certain important objects of the shot scene and in particular for sporting events it is possible to specify that it is about a soccer game. Advantageously, this allows a salience map to be obtained that weights salience zones according to the event. In a football game, it would be desirable to focus on the ball rather than the terraces.

The region of interest module thus allows one or more protruding zones, also referred to as regions of interest, to be extracted. These regions of interest are then geographically placed on the image.

They are identified by their coordinates along the height and width of the image. Their size can also be extracted for each region of interest. It is also possible to associate them with elements of semantic information. In fact, for a football game, if the user can select areas of interest to be displayed from the selection of several areas of interest to be displayed, information about the area of interest may be requested.

Module 115 may receive information about areas of interest to code them into a "Supplemental Enhancement Information" (SEI) type message.

SEI messages are coded as indicated in the table below:

uuid_iso_iec_11578 : A single word of 128 bits to indicate our message type to the decoder.

user_data_payload_byte : 8 bits containing part of the SEI message.

Generally in these cases:

paloadSize = 17(바이트들)이고 따라서 UUID에 대해서는 16이고 소유 데이터(proprietary data)에 대해서는 1이다.

paloadSize = 17 (bytes) and thus 16 for UUID and 1 for proprietary data.

user_data_payload_byte:

user_data_payload_byte:

From here:

number_of_ROI: 영상(또는 다음의 영상들)에 존재하는 관심 영역들의 수.

number_of_ROI: The number of regions of interest present in the image (or following images).

roi_x_16: 16 픽셀들의 배수인, 영상에서의 관심 영역의 위치 X.

roi_x_16: The position of the region of interest in the image, which is a multiple of 16 pixels.

roi_y_16: 16 픽셀들의 배수인, 영상에서의 관심 영역의 위치 Y.

roi_y_16: The position Y of the region of interest in the image, which is a multiple of 16 pixels.

roi_w_16: 16 픽셀들의 배수인, 영상에서의 관심 영역의 폭 W.

roi_w_16: The width W of the region of interest in the image, which is a multiple of 16 pixels.

roi_h_16: 16 픽셀들의 배수인, 영상에서의 관심 영역의 높이 H.

roi_h_16: The height H of the region of interest in the image, which is a multiple of 16 pixels.

semantic_information: 관심 영역을 특징짓는 타이틀.

semantic_information: The title that characterizes the region of interest.

Relative weights: 원칙적으로 가장 큰 관심을 갖는 관심 영역이 어느 것인지를 아는 방식으로 영상의 각 관심 영역의 가중치를 제공.

Relative weights: In principle, we provide the weights of each region of interest in the image in such a way as to know which region of interest is of greatest interest.

Macroblock_alignment: 관심 영역이 발견되는 시작 매크로블록의 넘버뿐만 아니라, 매크로블록들의 수로, 폭으로 및 높이로 관심 영역의 크기를 제공.

Macroblock_alignment: Provides the size of the region of interest in number, width and height of the macroblocks, as well as the number of the starting macroblock in which the region of interest is found.

When regions of interest are detected using the salient map, the rate of salience is obtained for each region of interest, and the regions are higher than a certain threshold whose salience is predetermined by the method for obtaining salient maps. Are classified as facet protrusions. Therefore, in SEI messages, regions of interest are sorted in ascending order of protrusion for all regions where the protrusion is higher than a fixed threshold.

Module 113 inserts the SEI message into the data stream and accordingly sends the coded video stream to the transmitting network.

The SEI message is sent before each image it references.

In other embodiments, it is also possible to transmit the SEI message only when the location of the at least one region of interest varies between two or more images. Therefore, during decoding, the decoder considers the last SEI message received, whether it is just before the picture to be decoded or if it is about a previously received picture unless the current picture is preceded by such an SEI message. Consider the last SEI message received.

2 illustrates a coding method according to the coding standard H.264 / AVC, which implements a preferred embodiment of the present invention.

During step E1, a salient map associated with the video to be broadcast is determined. To determine this salient map representing regions of interest, information about the video content may also be received to take this information into account during the establishment of the salient map. In particular, during a sporting event, the location of the ball may be considered to correspond to the region of interest for the user, in which case the zones of the image where the ball is located are privileged. When the video corresponds to the broadcast of the sidewalk where it is broadcast, it can also be assumed that the newscaster corresponds to the region of interest, in which case, using known image processing techniques, for example, by detecting a face Determine areas of interest by giving privileges to zones that contain newscasters.

At the end of the E1 step, one or more regions of interest regarding the video content are thus obtained.

During step E2, the coordinates of the regions of interest in the images are determined. The size of the regions of interest can also be determined in pixels and semantic information on the content can be associated with each region of interest.

At the same time, during step E3, the video stream is coded according to coding standard H.264. During coding, zones detected as regions of interest are privileged. In order to privilege regions of interest at the coding level, lower quantization steps are applied to them.

Following step E2, during step E4, an SEI message is generated from the position and semantic information associated with the regions of interest. The SEI message generated according to this follows the SEI message described above in Tables 1 and 2.

During step E5, the stream is constructed by inserting SEI messages into the stream to obtain a coded stream according to the H.264 standard.

The coded video stream is thus transmitted to the decoding devices in real time or in a delayed manner during step E6, which may be local or remote.

3 shows a preferred embodiment of the decoding device according to the invention, according to the coding standard H.264 / AVC.

The 209 module receives SEI messages on input. It extracts different SEI messages. NALs of useful data are sent to the entropy decoding module 201.

SEI messages are analyzed by module 210. This module enables decoding of the content of SEI messages representing regions of interest. The regions of interest of each image are thus identified in a simple manner at the level of the decoding device and before decoding of each image using the information contained in the field macroblock_alignment.

Macroblocks are sent to the re-ordering module 202 to obtain a set of coefficients. These coefficients undergo inverse quantization at module 203 and inverse DCT transform at module 204, and D'n macroblocks are obtained at the output of module 204, where D'n is a modified version of Dn. In order to reconstruct the macroblock uF'n, the adder 205 adds the prediction block P to D'n. Block P is a macroblock, performed by module 208 of the preceding decoded frame while coding in inter mode, by module 207 after motion compensation or when coded in intra mode. Obtained after intra prediction of (uF'n). Filter 206 is applied to signal uF'n to reduce the effects of distortion and a reconstructed frame F'n is generated from successive macroblocks.

Using information about the regions of interest included in the SEI messages, before blocks representing the regions of interest are detected in the stream and displayed, these blocks are identified and cropped at the user's choice, such as a PDA or mobile phone. May be sent for display to a device such as a.

For example, by entering semantic information, it is also possible to entrust the user to select a macroblock that the user wants to display. He enters "ball", for example, in which case the areas of interest containing the ball are displayed. If no region of interest is relevant to this meaning, all of the regions of interest may be displayed. Different regions of interest may be displayed in the form of a mosaic on the screen. When only one region of interest is displayed, this region of interest is enlarged and displayed on the screen to occupy the entire screen.

The decoding device thus decodes only macroblocks that are likely to contain information of interest to the user. In this way decoding is faster and requires less resources at the level of the decoding device and thus in reception. This is particularly useful when the receiving device is a mobile terminal with limited processing capacity.

4 illustrates a decoding method according to coding standard H.264 / AVC, which implements a preferred embodiment of the present invention.

Such a method can be implemented in a mobile terminal with limited display capacity.

During step S1, the type of display requested is selected. The selection is made by a user interface provided on the mobile terminal. It is determined to function in full picture mode and in this case the integrity of the video stream is displayed when the video stream is transmitted by the transmitter. Or it is determined to display only the regions of interest of the image. This unusual mode constitutes the specificity of the present invention. When it is determined to display the regions of interest, it proceeds to step S2, otherwise it proceeds to step S8. It is natural that different types of SEI messages may be inserted into the video stream in the case of other applications, in which case there may be a step of analyzing the SEI message before step S8 or during step S8.

During step S2, the user selects the usage he wants to do for the areas of interest. In particular, he can choose:

The maximum number of areas of interest he wants to display,

-In the form of a mosaic, for example, the way he wants to display various areas of interest on the screen,

The degree of zoom in the areas of interest he desires,

― Using keywords. The "semantic information" field of the regions of interest includes a keyword. In this case, for each image it is necessary to specify whether it is desired to display only one region of interest per image containing the keyword (and in this case the ones with the largest protrusions) or several regions of interest containing the keyword. It is also possible.

During step S3, SEI messages present in the stream are analyzed when they are being received. The SEI message is used to code the location of the regions of interest of the image when they are detected prior to image coding. Thus, for each image, there may be one or more regions of interest according to the visual characteristics of the image, according to the image content, or both. SEI messages are coded according to Tables 1 and 2 above. Information about SEI messages is temporarily recorded up to the display of the corresponding picture.

During step S4, the images are all decoded according to the decoding standard.

During step S5, the decoded ROIs are processed according to the user's selections during the S2 step. If the user selects a zoom of the main region of interest of the image, during step S6, the zone is enlarged to reach the maximum size of the display. If the user has selected a mosaic of regions of interest, the image is recreated into regions of interest, each enlarged according to the screen size and the number of regions of interest selected for display. If the user has specified a keyword, the regions of interest containing the keyword are displayed and enlarged.

During step S7, the regions of interest are displayed on the screen of the mobile terminal, in accordance with the desire of the user.

During step S8, after non-selection by the user displaying only regions of interest, the entire video stream is decoded for display.

5 shows a video indexing application of the present invention.

5 shows, in part, a personal recorder (PVR) type device 500. PVR 500 receives a compressed video stream at its input. According to the described embodiment, this video data stream is in accordance with coding standard H.264. The compressed video stream contains in particular SEI messages as described above in Tables 1 and 2.

This video data stream is partly sent to the recording support device 503. The recording support device may be understood as a hard disk, holographic support, memory card or "Blu-ray" disk. This recording support device may be remote in other embodiments.

The video data stream is sent to the decoder 501 to be decoded in real time in another part, which is displayed for example in a television set. In known devices, the stream is sent to the decoder 501 when the user wants to watch it in real time. Otherwise it is not decoded but simply recorded when recording is requested.

According to this aspect, the present invention proposes to decode a portion of the video data stream even when viewing in real time is not required. For a portion of a video stream it is understood in particular to areas of interest or specific areas of interest.

When the decoder 501 receives a video stream for which recording is requested, data is transmitted to the recording support apparatus 503. The recording support apparatus 503 records the data when it is received. In a simultaneous manner, decoder 501 receives the video data stream and progressively decodes SEI messages. The decoded regions of interest are transmitted to the video indexing module 502 responsible for temporarily recording them before sending them to the recording support device 503.

6 illustrates a method implemented by decoder 501 and indexing module 502.

During step T1, the video data stream is received by decoder 501. During step T2, decoder 501 decodes SEI messages present in the video data stream. The decoded SEI messages are the SEI messages as described above in Tables 1 and 2. The decoder may also decode other SEI messages, but that is not the purpose of the present invention. Each SEI message may describe one or more regions of interest per image, as described above in Tables 1 and 2. During step T3, the decoder 501 analyzes each SEI message and decodes each image. During this step, the weights indicated in the SEI message are used to select the region of interest to be recorded for each picture. In a preferred embodiment, the region of interest with the largest protrusion, ie with the highest weight, is preserved.

Once the region of interest is decoded, during step T4 it is sent to the indexing module 502. The recording of the region of interest per picture, and for all the pictures, is of little interest because it presents a large volume of information and also does not enable efficient indexing of the video. The indexing module thus determines the image used to index the video. According to the preferred embodiment described herein, only about 10 images will be selected for an hour and a half video. It can be guessed that in other embodiments the number of images will be larger. These ten images are taken at regular intervals. These selected pictures are included in the indexing module 502 and temporarily recorded in a RAM type memory not shown. In order to display them in the best way, the images are enlarged during step T5, ie they are enlarged so that they are all the same size. According to a preferred embodiment, this size may be the size of the image. To do so, they are read from the temporary memory and recorded again after their enlargement. According to another embodiment, the pictures are enlarged before recording them to temporary memory.

According to another embodiment, the images are provided as a mosaic on the display. Thus, without magnification, the images are reduced to one single size, the same for all of them.

When the entire video is received and thus recorded in the recording support device 503, during step T6, the indexed pictures are also transferred from the temporary memory to the recording support device 503 and recorded in a file.

Then, according to the desired usage, the regions of interest may be used for indexing or for displaying on a PVR type device when the user wants to consult the content of the database.

According to another aspect of the invention, it is also possible to encrypt the position data of the regions of interest during the coding of SEI messages. Thus, only users with the decryption key can access the regions of interest and thus have access to the visualization of the regions of interest or the indexing of the video streams due to the location information of the regions of interest. 2, this encryption step is step E4 '(not shown) but will be inserted after step E4.

The acquisition of the decryption key may be, for example, an object of payment for a service from a program broadcaster.

To do this, SEI messages about areas of interest are encapsulated in Real Time Protocol (RTP) type packets and transmitted on different video ports. Temporary CTS type labels may link SEI messages about regions of interest with corresponding pictures. Advantageously, this transmission mode can only encrypt RTP packets containing SEI messages and no video.

Decryption is performed at the level of the terminal receiver.

In the case of MPEG-2 TS encapsulation, the encryption standard used is DVB-CSA and the SEI messages for the regions of interest are encapsulated in a different PID than that of the video. SEI messages about regions of interest are linked to corresponding pictures via a timestamp (PTS) in the PES packet header. This transmission mode contains SEI messages for areas of interest, not video PIDs, and only allows encryption of PIDs.

According to another embodiment, the video data stream is coded according to coding standard H.264 / AVC using Flexible Macroblock Ordering (FMO), which independently enables coding of different parts of an image and thus enables their decoding independently. . FMO mode uses "slice groups". "Slice groups" are defined in the standard. In this embodiment, the regions of interest are coded in different groups than the rest of the image. PPS type NAL includes a map of "slice groups". SEI messages, such as those described below, indicating which regions of interest are coded are inserted.

The following tables illustrate the format of SEI messages used in accordance with this embodiment:

uuid_iso_iec_11578 : 128-bit single word to indicate our message type to the decoder.

user_data_payload_byte : 8 bits containing part of SEI message.

Typically in this case:

payloadSize = 17 (바이트)이고 따라서 UUID에 대해서는 16이고 소유 데이터에 대해서는 1이다.

payloadSize = 17 (bytes) and thus 16 for UUID and 1 for proprietary data.

user_data_payload_byte:

user_data_payload_byte:

Slice_group (i) _id: If slice_group_id is "1", slice_group represents the region of interest and if it is "0", slice_group represents the rest of the image.

For each slice_group representing the region of interest, semantic information, relative weight, and macroblocks relating thereto may be defined.

Thus, since the regions of interest are independently identified and coded, only macroblocks corresponding to the regions of interest can be decoded during reception.

Claims (14)

A method of indexing a coded video data stream, the video data stream comprising information regarding the location of regions of interest of each image, wherein the method comprises:
Receiving a coded video stream (T1),
Recording the coded video stream on a recording support,
Decoding position information of the regions of interest (T2),
Selecting a region of interest for each image (T3),
Decoding the video data (T3),
Selecting (T4) a predetermined number of regions of interest for the video data stream from the regions of interest selected per image,
Recording the selected regions of interest (T6)
Indexing method comprising a. The method of claim 1 wherein during the recording step:
The selected regions of interest are recorded in temporary memory when they are selected and decoded,
When the selected regions of interest are recorded in the temporary memory, the selected regions of interest are transmitted to a permanent memory support 503.
Indexing method characterized in that. The indexing method according to claim 1 or 2, wherein prior to recording the region of interest, the regions of interest are formatted to obtain a homogeneous size for all of the selected regions of interest. The method of claim 1, comprising encrypting the location of the regions of interest with the aid of an encryption key. 5. The method of claim 4 including obtaining a decryption key in accordance with a payment by a user. The indexing method according to any one of claims 1 to 5, wherein the video data stream is coded according to coding standard H.264 / AVC and the position information is included in a Supplemental Enhancement Information (SEI) type message. Way. 7. The method of claim 5 or 6, wherein the SEI messages are encapsulated in real-time protocol (RTP) packets, and the RTP packets are encrypted. 7. The indexing method according to claim 5 or 6, wherein said Supplemental Enhancement Information (SEI) type messages relating to region of interest location information are inserted into said coded data before or after each image they refer to. The method according to any one of claims 1 to 8, wherein the location information,
The number of regions of interest of each image,
The coordinates of each region of interest for each image dimension,
The surface of each region of interest,
A weight relating to the importance of the region of interest with respect to other regions of interest of the image,
Information about the content of each region of interest;
And any combination of this information
Indexing method comprising the information selected from. 10. The indexing method according to any one of claims 1 to 9, wherein in selecting the region of interest for each image (T3), the region of interest is selected according to a weight related to the importance of the region of interest. The video coding standard of claim 6, wherein the video coding standard uses flexible macro-bloc ordering (FMO), and the regions of interest are independently slice groups, independent of other image data. And wherein the location information of the regions of interest includes slice group numbers coded with the region of interest. 12. The method of claim 11, wherein a Supplemental Enhancement Information (SEI) message includes an identifier indicating for each slice group whether it is related to one region of interest. 13. The method of claim 12, comprising the further step of reading the SEI messages and the step of decoding the video data (T3) decodes only the slice groups containing the regions of interest. A device for indexing a coded video data stream, the video data stream comprising information regarding the location of regions of interest of each image, wherein the device comprises:
Means for receiving the coded video stream,
Means for recording the coded video stream on a recording support device 503,
Means (501) for decoding the location information of the regions of interest,
Means for decoding the video data 501,
Means 502 for selecting a region of interest per image,
Means (502) for selecting a predetermined number of regions of interest for the video data stream from among the regions of interest selected per image,
Means (503) for recording the selected regions of interest
And a device for indexing the coded video data stream.

Images