CN103716635B - Method and device for improving intelligent analysis performance - Google Patents
Method and device for improving intelligent analysis performance Download PDFInfo
- Publication number
- CN103716635B CN103716635B CN201310684226.3A CN201310684226A CN103716635B CN 103716635 B CN103716635 B CN 103716635B CN 201310684226 A CN201310684226 A CN 201310684226A CN 103716635 B CN103716635 B CN 103716635B
- Authority
- CN
- China
- Prior art keywords
- interest
- data
- area
- internal memory
- video
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Abstract
The invention provides a method for improving the intelligent analysis performance. The method is characterized by comprising the following steps: A, a graphics processing unit (GPU) hard-decodes an input compression coding video code stream, and the hard-decoded video image data is stored in a video memory of the GPU; B, according to a preset region of interest, the video image data, corresponding to the region of interest, in the video memory is copied to an internal memory; and C, the video image data in the internal memory is intelligently analyzed. According to the technical scheme of the invention, the advantages of the decoding performance of the GPU is fully utilized, the region of interest is set so that each intelligent recognition algorithm is enabled to acquire original image data of interest, and as only data in the region of interest is copied from the video memory to the internal memory, the data volume is small, thus enhancing the overall intelligent analysis performance.
Description
Technical field
The present invention relates to technical field of video monitoring, more particularly to a kind of method and apparatus for lifting intellectual analysis performance.
Background technology
With the extensive application of video monitoring service, the intellectual analysis of monitor video are more and more ripe, including virtual
Tripwire detection, quantity of motion detection, human face analysis, packet loss detection（In a region, one kind that object suddenly disappears is intelligent for detection
Analysis）, densely populated place detection etc..But the complete intellectual analysis flow process of monitoring field not only includes intelligent analysis process, together
When it also including video flowing decoding performance.With the lifting of current field of video monitoring video camera resolution, such as 1080P
Resolution is 1920 × 1080, its decoded 1080P yuv datas about 700Mbps, even if i73770 processors can only also be expired
6 tunnel frame per second of load parallel parsing is 30 frames, and resolution is the video code flow of the H264 of 1080P, consumes very big to cpu performance.
The content of the invention
In view of this, the present invention provides a kind of method and apparatus for lifting intellectual analysis performance.
The method of the present invention comprises the steps：A, graphic process unit GPU are carried out to the compressed encoded video code stream being input into
Hard decoder, the vedio data after hard decoder are stored in the video memory of the GPU；
B, according to default area-of-interest, will be the corresponding vedio data of the area-of-interest in the video memory multiple
Make in internal memory；C, intellectual analysis are carried out to the vedio data in internal memory.
Preferably, the corresponding vedio data of the area-of-interest in the video memory is being copied in internal memory it
Before, judge whether the corresponding vedio data of the area-of-interest needs to carry out transposition process, if it is, replicating transposition
Vedio data is to internal memory；If it is not, then replicating the vedio data to internal memory.
Preferably, the corresponding vedio data of the area-of-interest in the video memory is copied in internal memory and is specifically wrapped
Include：According to the classification of intellectual analysis, replicate the Y-component of the corresponding vedio data of the area-of-interest, U components, V component or
Which is combined to internal memory.
Preferably, according to the result that intellectual analysis are carried out to the vedio data in internal memory, judge whether to need into
One step replicates the vedio data of another area-of-interest from video memory, if it is, replicating another interested from video memory
The vedio data in region, does not otherwise replicate；Wherein another area-of-interest is the position obtained according to the intellectual analysis
What information determined.
A kind of device for lifting intellectual analysis performance, the device include：Hard decoder module, for the compressed encoding to being input into
Video code flow carries out hard decoder, and the vedio data after hard decoder is stored in the video memory of GPU；Data duplication module, is used for
According to default area-of-interest, the corresponding vedio data of the area-of-interest in the video memory is copied to into internal memory
In；Intelligent analysis module, for carrying out intellectual analysis to the vedio data in internal memory.
Preferably, the device further includes transposition judge module；By the correspondence of the area-of-interest in the video memory
Vedio data copy in internal memory before, transposition judge module judges the corresponding vedio data of the area-of-interest
Whether need to carry out transposition process, if it is, data duplication module replicates the vedio data of transposition to internal memory；If
No, then data duplication module replicates the vedio data to internal memory.
Preferably, the corresponding vedio data of the area-of-interest in the video memory is copied to by data duplication module
It is concrete in internal memory to perform following operation：According to the classification of intellectual analysis, the corresponding vedio data of the area-of-interest is replicated
Y-component, U components, V component or its be combined to internal memory.
Preferably, data duplication module is sentenced according to the result that intellectual analysis are carried out to the vedio data in internal memory
It is disconnected whether to need the vedio data of another area-of-interest is further replicated from video memory, if it is, multiple from video memory
The vedio data of another area-of-interest is made, is not otherwise replicated；Wherein another area-of-interest is according to the intelligence point
What the positional information that analysis is obtained determined.
Compared to prior art, technical solution of the present invention makes full use of the decoding performance advantage of GPU, interested by arranging
Region, cause decoding after each Intelligent Recognition algorithm can obtain oneself original image data interested, and due to aobvious
Deposit replicate and internal memory between be only area-of-interest data, so data volume is smaller, thus improve intelligence on the whole
The performance of analysis.
Description of the drawings
Fig. 1 is the flow chart of embodiment one.
Fig. 2 is the flow chart of embodiment two.
Fig. 3 is the flow chart of embodiment three.
Fig. 4 is area-of-interest exemplary plot.
Fig. 5 is the area-of-interest exemplary plot after transposition.
Fig. 6 is another area-of-interest exemplary plot.
Fig. 7 is apparatus of the present invention logic chart.
Specific embodiment
Technical solution of the present invention makes full use of the decoding performance advantage of GPU, after arranging area-of-interest, causing decoding
Each Intelligent Recognition algorithm can obtain oneself original image data interested（Loss without primary data information (pdi)）, video memory
It is little with the direct data copy amount of internal memory such that it is able to greatly improve the performance of intellectual analysis；And multiple intelligent algorithms can
Common memory, makes the memory consumption of whole scheme also minimum；And as the original video data information for treating intellectual analysis does not have
Lose, it is possible to adapt to various intellectual analysis.Describe technical solution of the present invention with reference to embodiments in detail.
Embodiment one
Please join Fig. 1, the method for the lifting intellectual analysis performance of the present embodiment comprises the steps：
S11, graphic process unit GPU carry out hard decoder to the compressed encoded video code stream being input into, the video figure after hard decoder
As data storage is in the video memory of the GPU.
S12, according to default area-of-interest, by the corresponding vedio data of the area-of-interest in the video memory
Copy in internal memory.
S13, intellectual analysis are carried out to the vedio data in internal memory.
The video code flow of compressed encoding is input into intellectual analysis equipment in units of frame, and the GPU of the intellectual analysis equipment is to defeated
The each frame compressed encoded video view data for entering is decoded, and the vedio data obtained after decoding is stored in the GPU's
In video memory.Decoded vedio data is typically all in the form of the Y-component of yuv space, U components, V component.By
Belong to hard decoder in the decoding process, so performance is very high.Above-mentioned intellectual analysis equipment can be that a single physics sets
It is standby, or be present on other physical equipments as a logic entity.
After the completion of decoding, need the vedio data in video memory to be copied in internal memory, after such CPU could be to decoding
Vedio data carry out intellectual analysis.CPU may only need to the video figure using local when intellectual analysis are carried out
As data.Such as, when carrying out mixing line detection, it is only necessary to use some vedio datas mixed near line；Entering
When row Face datection, may only need to use the corresponding vedio data of face.So from this point of view, only need
Partial video view data in video memory is copied in internal memory.When implementing, user can predefine
The area-of-interest of intellectual analysis, when then vedio data is replicated in video memory is performed, only copies area-of-interest
Vedio data.User presets the mode of area-of-interest and belongs to prior art, will not be described here；Intellectual analysis
How equipment determines that according to the area-of-interest that user sets the corresponding vedio data in video memory falls within prior art, no
Further describe in detail.
After the vedio data of area-of-interest is copied in internal memory, CPU is held to the vedio data in internal memory
Row intellectual analysis.
Although the vedio data data volume being stored in after decoding in video memory is than larger, due to copying in internal memory
Be only the corresponding data of area-of-interest, so data volume will not be very big, the performance of whole intellectual analysis will be had substantially
Improve；And the data for copying belong to the raw video image data lost without information, and the result of intellectual analysis will be more accurate.
Embodiment two
In order to, on the basis of above-described embodiment one, further improve the performance of intellectual analysis, the embodiment of the present invention two increases
The whether process of vedio data transposition added.Please join the flow process of Fig. 2：
S21, graphic process unit GPU carry out hard decoder to the compressed encoded video code stream being input into, the video figure after hard decoder
As data storage is in the video memory of the GPU.
S22, according to default area-of-interest, judge in video memory the corresponding vedio data of the area-of-interest is whether
Need to carry out transposition process, if it is, execution step S23, otherwise, execution step S24.
S23, the vedio data of the area-of-interest of duplication transposition are to internal memory.
S24, the vedio data of the area-of-interest is replicated to internal memory.
S25, intellectual analysis are carried out to the vedio data in internal memory.
Judge whether above-mentioned vedio data needs the principle for carrying out transposition process to be：If the data volume after transposition is little
Data volume before transposition, then carry out transposition, does not otherwise carry out transposition.
As a example by mixing line and detect this intellectual analysis.Please join Fig. 4, user paint in monitoring image one mix line after i.e. phase
When in area-of-interest is determined, because intellectual analysis equipment can determine the model of area-of-interest according to the coordinate information for mixing line
Enclose.The area-of-interest that solid-line rectangle frame in such as Fig. 4 is determined equivalent to user.Performing in the area-of-interest
When vedio data copies to internal memory, actually replicate for the vedio data in dotted rectangle.But if will be real
Vedio data in line rectangle frame carries out transposition, please join the video image after Fig. 5 transposition, it can be clearly seen that data volume is little
Vedio data before transposition.So in this case it is necessary to replicating the video image number of the area-of-interest of transposition
According to the intellectual analysis for performing tripwire detection to internal memory for CPU.
In the case where needing to perform transposition process, the vedio data for replicating the area-of-interest of transposition is deposited into interior
One step reduces the data volume of copy, while the operand of follow-up Intelligent Recognition algorithm can be also reduced, so this enters to performance
One step serves the effect of lifting.
For embodiment one and two, area-of-interest video figure can also be replicated according to the classification of intellectual analysis further
As the Y-component of data, U components, V component or its be combined to internal memory.Such as, for line detection is mixed, need to be paid close attention to due to which and be stumbled
The monochrome information of line periphery, and chromatic component need not be paid close attention to, so when data duplication is carried out, it is only necessary to replicate sense emerging
The Y-component of interesting area video view data（Brightness data）；Such as the detection of color, can only replicate U components and V point
Amount.That is, for the duplication amount for further reducing data, correspondence can be carried out according to actual Intelligent Recognition algorithm needs
The duplication of vector.Six kinds of situations are there is so：Replicate Y-component, or U components, or V component, or YU components, or YV
Component, or UV components, or YUV components.
Embodiment three
Embodiment three relates generally to further return duplication separately according to the analysis result of area-of-interest vedio data
The problem of the vedio data of one area-of-interest.Please join Fig. 3, the handling process of the embodiment is as follows：
S31, graphic process unit GPU carry out hard decoder to the compressed encoded video code stream being input into, the video figure after hard decoder
As data storage is in the video memory of the GPU.
S32, according to default area-of-interest, by the corresponding vedio data of the area-of-interest in the video memory
Copy in internal memory.
S33, intellectual analysis are carried out to the area-of-interest vedio data in internal memory.
S34, according to intellectual analysis result, judge whether to need regarding for another area-of-interest is further replicated from video memory
Frequency view data, if it is, performing S35, otherwise performs S36.
S35, the vedio data that another area-of-interest is replicated from video memory, wherein another area-of-interest is root
What the positional information obtained according to intellectual analysis in step S33 determined, turn S37.
S36, the vedio data that another area-of-interest is no longer replicated from video memory.
S37, the vedio data to another area-of-interest in internal memory carry out intellectual analysis.
One example application of the embodiment is：For Car license recognition, first carry out and mix line detection, if testing result is have
Vehicle crosses line of stumbling, then further perform the Car license recognition to the vehicle.For this application, it is possible to which consideration uses embodiment
Three method.
In step S34, the result of the intellectual analysis, such as have vehicle to cross line of stumbling, then need further to perform car
Board is recognized；Regarding for another area-of-interest required for result further according to intellectual analysis, i.e. Car license recognition is needed now
Whether there is in frequency view data current memory, if it is present without the need for another area-of-interest is further replicated from video memory
Vedio data；If it does not exist, then needing the video image number of another area-of-interest is further replicated from video memory
According to.Determine for the determination of another area-of-interest position can rely on the positional information that intellectual analysis are obtained in step S33.Please
Ginseng Fig. 6, in the example of fig. 6, mixes line detection and has been detected by the positional information that vehicle triggers line of stumbling（A points such as in figure）, now
The region for carrying out Car license recognition can be determined according to the positional information（Such as dotted rectangle in Fig. 6）Coordinate information, i.e., it is another
The coordinate information of area-of-interest.As Car license recognition generally needs the higher data of resolution ratio, so it is emerging directly to replicate sense
The decoded original video data in interesting region（Relative to the number that compressed encoding obtains small in resolution is further performed after decoding
According to）Will be helpful to perform the Car license recognition；And if the video image number of another area-of-interest required for Car license recognition
Exist according in current memory, then before and after intellectual analysis can be in common memory twice vedio data, so it is whole just
The memory consumption of case will be smaller.
The vedio data and step of another area-of-interest replicated from video memory in step S35 in the present embodiment
The vedio data of the area-of-interest replicated in S32 is for same frame video image data, only again
What is made is the data of diverse location in the frame data.
It should be noted that in embodiment three, it is also possible to increase judgement that whether transposition processed and according to intelligent knowledge
Other classification replicates Y-component, U components, V component or its be combined to the process of internal memory.
Based on same design, the present invention also provides a kind of device for lifting intellectual analysis performance.Please join Fig. 7, the device
Including：Hard decoder module, data duplication module, transposition judge module and intelligent analysis module.
The hard decoder module, carries out hard decoder to the compressed encoded video code stream being input into for GPU, the video after hard decoder
View data is stored in the video memory of GPU；Data duplication module, for according to default area-of-interest, by the video memory
The corresponding vedio data of the area-of-interest copy in internal memory, so that CPU carries out intelligence to the vedio data
Analysis；Intelligent analysis module, for carrying out intellectual analysis to the vedio data in internal memory.
Before the corresponding vedio data of the area-of-interest in the video memory is copied in internal memory, transposition is sentenced
Disconnected module judges whether the corresponding vedio data of the area-of-interest needs to carry out transposition process, if it is, data are multiple
The vedio data of molding block reproducing unit is to internal memory；If it is not, then data duplication module directly replicates the video image number
According to internal memory.
The vedio data for replicating transposition can further lift the performance of intellectual analysis to internal memory, play further excellent
The effect of change.
Data duplication module copies to the area-of-interest corresponding vedio data in the video memory in internal memory
It is concrete to perform following operation：According to the classification of intellectual analysis, replicate the corresponding vedio data of the area-of-interest Y point
Amount, U components, V component or its be combined to internal memory.
Data duplication module judges whether to need according to the result for carrying out the vedio data in internal memory intellectual analysis
The vedio data of another area-of-interest is replicated from video memory further, if it is, another sense is replicated from video memory
The vedio data in interest region, does not otherwise replicate；Wherein another area-of-interest is obtained according to the intellectual analysis
What positional information determined.
Presently preferred embodiments of the present invention is the foregoing is only, not to limit the present invention, all essences in the present invention
Within god and principle, any modification, equivalent substitution and improvements done etc. are should be included within the scope of protection of the invention.
Claims (6)
1. it is a kind of lifted intellectual analysis performance method, it is characterised in that the method comprises the steps：
A, graphic process unit GPU carry out hard decoder to the compressed encoded video code stream being input into, the vedio data after hard decoder
It is stored in the video memory of the GPU；
B, according to default area-of-interest, the corresponding vedio data of the area-of-interest in the video memory is copied to
In internal memory；
C, intellectual analysis are carried out to the vedio data in internal memory；
Wherein, according to the result that intellectual analysis are carried out to the vedio data in internal memory, judge whether need further from
The vedio data of another area-of-interest is replicated in video memory, if it is, another area-of-interest is replicated from video memory
Vedio data, does not otherwise replicate；Wherein another area-of-interest is that the positional information obtained according to the intellectual analysis is true
Fixed.
2. the method for claim 1, it is characterised in that by the corresponding video of the area-of-interest in the video memory
Before view data is copied in internal memory, judge whether the corresponding vedio data of the area-of-interest needs to carry out at transposition
Reason, if it is, replicating the vedio data of transposition to internal memory；If it is not, then replicating the vedio data to internal memory.
3. method as claimed in claim 1 or 2, it is characterised in that regard the area-of-interest in the video memory corresponding
Frequency view data is specifically included in copying to internal memory：According to the classification of intellectual analysis, the corresponding video of the area-of-interest is replicated
The Y-component of view data, U components, V component or its be combined to internal memory.
4. it is a kind of lifted intellectual analysis performance device, it is characterised in that the device includes：
Hard decoder module, carries out hard decoder to the compressed encoded video code stream being input into for GPU, the video image number after hard decoder
According to being stored in the video memory of the GPU；
Data duplication module, for according to default area-of-interest, regarding the area-of-interest in the video memory corresponding
Frequency view data is copied in internal memory, so that CPU carries out intellectual analysis to the vedio data；
Intelligent analysis module, for carrying out intellectual analysis to the vedio data in internal memory；
Wherein, data duplication module judges whether according to the result that intellectual analysis are carried out to the vedio data in internal memory
Needs further replicate the vedio data of another area-of-interest from video memory, if it is, replicating from video memory another
The vedio data of area-of-interest, does not otherwise replicate；Wherein another area-of-interest is obtained according to the intellectual analysis
Positional information determine.
5. device as claimed in claim 4, it is characterised in that the device further includes transposition judge module；
Before the corresponding vedio data of the area-of-interest in the video memory is copied in internal memory, transposition judges mould
Block judges whether the corresponding vedio data of the area-of-interest needs to carry out transposition process, if it is, data duplication mould
Block replicates the vedio data of transposition to internal memory；If it is not, then data duplication module replicates the vedio data to internal memory.
6. the device as described in claim 4 or 5, it is characterised in that data duplication module will be this in the video memory interested
During the corresponding vedio data in region copies to internal memory, concrete execution operates as follows：According to the classification of intellectual analysis, replicating should
The Y-component of the corresponding vedio data of area-of-interest, U components, V component or its be combined to internal memory.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310684226.3A CN103716635B (en) | 2013-12-12 | 2013-12-12 | Method and device for improving intelligent analysis performance |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310684226.3A CN103716635B (en) | 2013-12-12 | 2013-12-12 | Method and device for improving intelligent analysis performance |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103716635A CN103716635A (en) | 2014-04-09 |
| CN103716635B true CN103716635B (en) | 2017-04-19 |
Family
ID=50409140
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310684226.3A Active CN103716635B (en) | 2013-12-12 | 2013-12-12 | Method and device for improving intelligent analysis performance |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103716635B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104063313B (en) * | 2014-04-15 | 2017-05-17 | 深圳英飞拓科技股份有限公司 | Intelligent analytical algorithm test system and method |
| CN108206937B (en) * | 2016-12-20 | 2020-05-19 | 浙江宇视科技有限公司 | Method and device for improving intelligent analysis performance |
| CN110691224A (en) * | 2019-10-31 | 2020-01-14 | 上海电力大学 | Transformer substation perimeter video intelligent detection system |
| CN112040148A (en) * | 2020-09-02 | 2020-12-04 | 北京锐马视讯科技有限公司 | Video value-added service method, device and storage medium |
| CN114143595A (en) * | 2021-12-08 | 2022-03-04 | 珠海豹趣科技有限公司 | Video wallpaper playing method and device, electronic equipment and readable storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101261683A (en) * | 2008-03-26 | 2008-09-10 | 北京航空航天大学 | A vehicle detection method based on color video |
| CN101908035A (en) * | 2010-07-30 | 2010-12-08 | 北京华傲精创科技开发有限公司 | Video coding and decoding method, GPU (Graphics Processing Unit) as well as interacting method and system of same and CPU (Central Processing Unit) |
| CN102034265A (en) * | 2010-11-24 | 2011-04-27 | 清华大学 | Three-dimensional view acquisition method |
| CN102143386A (en) * | 2010-01-28 | 2011-08-03 | 复旦大学 | Streaming media server acceleration method based on graphics processing unit |
| CN102695040A (en) * | 2012-05-03 | 2012-09-26 | 中兴智能交通（无锡）有限公司 | Parallel high definition video vehicle detection method based on GPU |
| WO2013097098A1 (en) * | 2011-12-27 | 2013-07-04 | 华为技术有限公司 | Data processing method, graphics processing unit (gpu) and first node device |
-
2013
- 2013-12-12 CN CN201310684226.3A patent/CN103716635B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101261683A (en) * | 2008-03-26 | 2008-09-10 | 北京航空航天大学 | A vehicle detection method based on color video |
| CN102143386A (en) * | 2010-01-28 | 2011-08-03 | 复旦大学 | Streaming media server acceleration method based on graphics processing unit |
| CN101908035A (en) * | 2010-07-30 | 2010-12-08 | 北京华傲精创科技开发有限公司 | Video coding and decoding method, GPU (Graphics Processing Unit) as well as interacting method and system of same and CPU (Central Processing Unit) |
| CN102034265A (en) * | 2010-11-24 | 2011-04-27 | 清华大学 | Three-dimensional view acquisition method |
| WO2013097098A1 (en) * | 2011-12-27 | 2013-07-04 | 华为技术有限公司 | Data processing method, graphics processing unit (gpu) and first node device |
| CN102695040A (en) * | 2012-05-03 | 2012-09-26 | 中兴智能交通（无锡）有限公司 | Parallel high definition video vehicle detection method based on GPU |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103716635A (en) | 2014-04-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103716635B (en) | Method and device for improving intelligent analysis performance | |
| CN105163127B (en) | video analysis method and device | |
| Herglotz et al. | Modeling the energy consumption of the HEVC decoding process | |
| US20100128798A1 (en) | Video processor using optimized macroblock sorting for slicemap representations | |
| US10477249B2 (en) | Video processing for masking coding artifacts using dynamic noise maps | |
| EP2238764A1 (en) | Coding mode selection for block-based encoding | |
| CN110418177A (en) | Method for video coding, device, equipment and storage medium | |
| CN1981533A (en) | Method and system for performing deblocking filtering | |
| US20180103251A1 (en) | Method for selecting prediction mode of intra prediction, video encoding device and image processing apparatus | |
| CN113473148B (en) | Computing system for video coding and video coding method | |
| CN110139104A (en) | Video encoding/decoding method, device, computer equipment and storage medium | |
| CN107197297A (en) | A kind of video steganalysis method of the detection based on DCT coefficient steganography | |
| CN116208770A (en) | Enhanced real-time visual quality metric generation for video coding | |
| Wang et al. | End-to-end compression towards machine vision: Network architecture design and optimization | |
| JP2009522650A (en) | Hardware motion compensation in video recording | |
| US20100158105A1 (en) | Post-processing encoding system and method | |
| BR102012025882A2 (en) | METHOD AND DEVICE FOR DETERMINING A BALANCE VALUE OF A BLOCK OF A PREDICTIVE BLOCK VIDEO CODED IN A DATA FLOW | |
| Wang et al. | Semantic-aware video compression for automotive cameras | |
| CN103581688A (en) | Video image coding and decoding method and device | |
| CN108765503B (en) | Skin color detection method, device and terminal | |
| CN116980604A (en) | Video encoding method, video decoding method and related equipment | |
| US8934539B2 (en) | Vector processor acceleration for media quantization | |
| Baldev et al. | Scalable wavefront parallel streaming deblocking filter hardware for HEVC decoder | |
| US11403783B2 (en) | Techniques to dynamically gate encoded image components for artificial intelligence tasks | |
| US11170260B2 (en) | Techniques for determining importance of encoded image components for artificial intelligence tasks |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |