CN110113616B - Multi-level monitoring video efficient compression coding and decoding device and method - Google Patents
Multi-level monitoring video efficient compression coding and decoding device and method Download PDFInfo
- Publication number
- CN110113616B CN110113616B CN201910488842.9A CN201910488842A CN110113616B CN 110113616 B CN110113616 B CN 110113616B CN 201910488842 A CN201910488842 A CN 201910488842A CN 110113616 B CN110113616 B CN 110113616B
- Authority
- CN
- China
- Prior art keywords
- decoding
- background frame
- frame
- decoder
- coding
- 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
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/134—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or criterion affecting or controlling the adaptive coding
- H04N19/136—Incoming video signal characteristics or properties
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
- H04N19/134—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or criterion affecting or controlling the adaptive coding
- H04N19/142—Detection of scene cut or scene change
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/20—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using video object coding
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/42—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals characterised by implementation details or hardware specially adapted for video compression or decompression, e.g. dedicated software implementation
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
- H04N19/70—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals characterised by syntax aspects related to video coding, e.g. related to compression standards
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention discloses a device and a method for efficiently compressing, encoding and decoding multi-level monitoring videos, and belongs to the technical field of mass camera metropolitan area level video monitoring application. The method comprises the following steps: (1) specific semantic object coding: detecting a particular semantic object dnReconstructing the objectDetecting a key point sequence by a tracking technology and a key point detection technology, and transmitting the key point sequence to a decoder; (2) modeling a long-term background frame: setting a plurality of scene categories, and distinguishing the scene categories by using background frame scene index sequence numbers; detecting scene types in the encoder, and transmitting background frame index sequence numbers to a decoder; (3) short-term background frame modeling: obtaining the short-term background frame prediction value of the current frame by adopting a multi-mode prediction methodTransmitting the encoding control parameters to a decoder by optimally selecting the multi-mode reference prediction; (4) and (3) foreground coding: prediction residualForeground code stream is generated through HEVC coding, and reconstructed foreground is obtained through decodingThe prediction residual is transmitted to the decoder through a channel.
Description
Technical Field
The invention relates to the technical field of metropolitan area level video monitoring application of massive cameras, in particular to a device and a method for efficiently compressing, encoding and decoding multi-level monitoring videos.
Background
Most cameras are applied to security protection, and a video signal has the characteristics of: (1) background does not change or changes little in a period of time, and background frame modeling can provide possibility for more efficient coding compared with applications such as broadcast television, video websites and the like; (2) the public security city snow project deploys a large number of cameras, most of data of the cameras are invalid and cannot be seen by people, and most of information is seen by machines. (3) The intelligent security application usually focuses on target objects with specific semantics, such as pedestrians, vehicles, human faces, license plates and the like in a scene, and the specific semantic objects are focused on practical applications such as city-level retrieval, big data analysis and the like.
The prior art has the defects that: early MPEG-4 object-oriented coding techniques also focused on search-oriented applications. However, the development of machine vision and target detection technologies is not mature enough before 2000 years ago, so that the standard cannot be really applied in practice. In recent years, with the development of deep learning technology and the continuous strong computing power of computing platforms, the high-performance detection of target objects with specific semantics becomes possible. In recent years, an end-to-end image coding framework based on deep learning is broken through, and a high-dimensional feature vector expressed by the deep learning can be used as a compact retrieval descriptor, so that the possibility of extracting, representing and coding the depth feature driven by compressing and retrieving double targets can be provided.
But the above work was explored from a different perspective. Aiming at mass camera cluster perception and machine understanding of scenes such as video coding, the video data coding and compressing appeal is greatly different from that of the traditional video coding. How to effectively utilize the multi-dimensional data redundancy of the video data space-time-Camer is not solved on the premise of ensuring the perception understanding efficiency of a machine, and the effective realization of data coding compression is still achieved.
Disclosure of Invention
Aiming at the problems in the prior art, the invention focuses on the characteristics, makes full use of the video data space-time-Camer multi-dimensional data redundancy, and provides a multi-layer efficient video coding algorithm framework in a targeted manner.
A high-efficiency compression coding method for multi-level surveillance videos comprises the following levels:
(1) specific semantic object coding: detecting a particular semantic object dnReconstructing the objectFinding time domain track by tracking technology, and checking by key pointDetecting key point sequences for all versions of a target object on a time domain track by using a detection technology, and transmitting key sequence structural information to a decoder;
(2) modeling a long-term background frame: setting a plurality of scene categories, and distinguishing the scene categories by using background frame scene index sequence numbers; constructing a background frame a by an offline training method during camera installationn(ii) a Detecting scene type in encoder to obtain long-term background frameTransmitting the background frame index sequence number to a decoder;
(3) short-term background frame modeling: suppose that the current frame fnPreceding adjacent frame fn-1fn-2A decoded reconstructed version has been obtainedObtaining the short-term background frame prediction value of the current frame by adopting a multi-mode prediction methodSelecting a proper reference frame and a weighted prediction coefficient by optimizing and selecting the multi-mode reference prediction, and transmitting an encoding control parameter to a decoder;
(4) and (3) foreground coding: prediction residualForeground code stream is generated through HEVC coding, and reconstructed foreground is obtained through decodingThe prediction residual is transmitted to the decoder through a channel.
Further, in the step (1), a deep learning detector is used to detect the specific semantic object dnThe semantic objects comprise pedestrians, vehicles and human faces, and are subjected to compression, retrieval, dual-target-driven feature extraction, representation and encoding, and deconvolution to obtain decoding reconstruction objects
Further, in the step (4), the prediction residual c is directly predicted by specifying the semantic objectnIs set to 0.
The decoding method of the specific semantic object is to utilize a decoder to decode and obtain key point sequence structured information and an object obtained by deconvolution decoding, and study the decoding version of an adjacent frame target object reconstructed by interpolation through a geometric method
The decoding method of the long-term background frame is that the decoder reconstructs the background frame by utilizing the scene category index number of the long-term background frame
The decoding method of the short-term background frame is that a decoder obtains by using a reference frame and a weighted prediction coefficient
The decoding method of the prediction residual error is that a decoder decodes a foreground code stream
Further, the decoding of the specific semantic object is performed by deconvolution reconstruction of the target object, and finally, the video decoding is as follows:
a multi-level monitoring video high-efficiency compression coding device comprises a specific semantic object coding module, a long-term background frame modeling module, a short-term background frame modeling module and a foreground coding module,
the specific languageSemantic object coding module detects specific semantic object dnReconstructing the objectFinding a time domain track through a tracking technology, detecting key point sequences for all versions of a target object on the time domain track through a key point detection technology, and transmitting key point sequence structured information to a decoder;
the long-term background frame modeling module sets a plurality of scene categories, and distinguishes the scene categories by using background frame scene index sequence numbers; constructing a background frame a by an offline training method during camera installationn(ii) a Detecting scene type in encoder to obtain long-term background frameTransmitting the background frame index sequence number to a decoder;
the short-term background frame modeling module assumes the current frame fnPreceding adjacent frame fn-1fn-2A decoded reconstructed version has been obtainedObtaining the short-term background frame prediction value of the current frame by adopting a multi-mode prediction methodSelecting a proper reference frame and a weighted prediction coefficient by optimizing and selecting the multi-mode reference prediction, and transmitting an encoding control parameter to a decoder;
the foreground coding module predicts the residual errorForeground code stream is generated through HEVC coding, and reconstructed foreground is obtained through decodingThe prediction residual is transmitted to the decoder through a channel.
Further, the specific semantic object coding moduleDetection of a particular semantic object d using a deep learning detectornThe semantic objects comprise pedestrians, vehicles and human faces, and are subjected to compression, retrieval, dual-target-driven feature extraction, representation and encoding, and deconvolution to obtain decoding reconstruction objectsFurther, in the step (4), the prediction residual c is directly predicted by specifying the semantic objectnIs set to 0.
A decoding device for processing a multi-level monitoring video high-efficiency compression coding method comprises an object decoding module, a long-term background frame decoding module, a short-term background frame decoding module and a foreground decoding module, wherein the object decoding module obtains key point sequence structural information by decoding through a decoder and an object obtained by deconvolution decoding, and a decoding version of an adjacent frame target object is reconstructed by interpolation through a geometric method
The long-term background frame decoding module reconstructs the long-term background frame by using the scene category index number of the long-term background frame
The short-term background frame decoding module obtains a short-term background frame by using a reference frame and a weighted prediction coefficient
Further, the final video decoding is as follows:
the invention has the beneficial effects that:
(1) the compression performance is greatly improved;
(2) visual object depth features may support decoding reconstruction and retrieval.
Drawings
FIG. 1 is a block diagram of multi-level efficient predictive coding;
fig. 2 is a block diagram of multi-level efficient predictive decoding.
Detailed Description
The technical scheme of the invention is further explained by combining the drawings in the specification.
The technical problems to be solved by the invention are as follows:
(1) modeling a long and short-term background frame: the video data space-time-Camera multi-dimensional data redundancy in a massive camera cluster is fully utilized,
(2) machine-understood video semantic object depth feature coding: (compression encoding and retrieval dual targets);
(3) semantic object space-time multi-level feature coding;
as shown in fig. 1 and 2, the technical solution of the present invention is as follows:
(1) specific semantic object coding
Detection of a particular semantic object d based on a deep learning detector (SSD or YOLO)n(such as pedestrians, vehicles, human faces and the like identified by a machine), compressing and retrieving double-target-driven feature extraction, representation and encoding, wherein the feature extraction adopts the unstructured depth features represented by CNN high-dimensional feature vectors, and decoding reconstruction objects are obtained by deconvolution(black rectangular frame in figure), considering the video signal time domain correlation, a target object is realized in a camera which usually lasts for a period of time, after once coding description, a time domain track can be found through a tracking technology, a key point sequence is detected for all versions of the target object on the time domain track through a key point detection technology, the tracking technology and the key point detection technology both adopt the prior art to detect, and only key points are detectedTransmitting the point sequence structural information to a decoder, decoding by the decoder to obtain key point sequence structural information and objects obtained by deconvolution decoding, and researching a decoding version for reconstructing adjacent frame target objects by interpolation through a geometric method
(2) Long term background frame modeling
In security application, the background of the camera is fixed in most of time, if no moving target exists in the camera in a late night time period, under the condition, if a background frame can be constructed, the compression efficiency can be greatly improved by the differential encoding technology based on the background frame. The distribution characteristics of the pixel brightness and the chromaticity of the background frame are different under different seasons, illumination and weather conditions. Based on the background modeling method, a plurality of scene categories are set according to different combinations of seasons, illumination and weather, and the scene categories are distinguished by using background frame scene index sequence numbers. Constructing a background frame a by an offline training method during camera installationn. Detecting scene type in encoder to obtain long-term background frameThe background frame index sequence number is transmitted to a decoder, and the decoder can reconstruct the sequence number
(3) Short term background frame modeling
The long-term background frame only describes a common background for a relatively long time, but the actual scene is complex in the daytime, more target motion, occlusion and other areas appear, the coding scheme needs to decode and reconstruct all frames, and in order to fully utilize the short-term time domain correlation, the invention constructs the short-term background frame to utilize the short-term time domain redundancy to the maximum extent. Suppose that the current frame fnPreceding adjacent frame fn-1fn-2A decoded reconstructed version has been obtainedThen adoptMulti-mode prediction methods, e.g. linear weighted prediction, obtaining a prediction of a short-term background frame of a current frameBy optimally selecting multi-mode reference prediction, selecting proper reference frame and weighted prediction coefficient, these coding control parameters are transmitted to a decoder, and the decoding end can obtain the same
(4) Foreground (prediction residual) coding
Current frame f relative to short-term background framenThere is also some foreground irregularity, namely the prediction residualForeground code stream is generated through HEVC coding, and reconstructed foreground is obtained through decodingThe prediction residual is transmitted to the decoder via a channel, and the decoder can decode the prediction residualNote that here also the target object regions of specific semantics need to be considered, for which rectangular regions the prediction residual c is directly predictednSet to 0, these rectangular regions are decoded by deconvolution reconstruction of the target object, i.e.(black rectangular box in the figure). Finally, the video is decoded as follows
The invention fuses a multi-level video coding frame of long and short term background frames, foreground coding and semantic object feature coding; modeling by utilizing a long and short-term background frame of video data space-time-Camera multi-dimensional data redundancy in a massive camera cluster; compressing, coding and retrieving the video semantic object depth feature coding driven by the two targets; semantic object space-time multi-level feature coding.
Claims (10)
1. A high-efficiency compression coding method for multi-level surveillance videos is characterized by comprising the following levels:
(1) specific semantic object coding: detecting a particular semantic object dnReconstructing the objectFinding a time domain track through a tracking technology, detecting key point sequences for all versions of a target object on the time domain track through a key point detection technology, and transmitting key point sequence structured information to a decoder;
(2) modeling a long-term background frame: setting a plurality of scene categories, and distinguishing the scene categories by using background frame scene index sequence numbers; constructing a background frame a by an offline training method during camera installationn(ii) a Detecting scene type in encoder to obtain long-term background frameTransmitting the background frame index sequence number to a decoder;
(3) short-term background frame modeling: suppose that the current frame fnPreceding adjacent frame fn-1fn-2A decoded reconstructed version has been obtainedObtaining the short-term background frame prediction value of the current frame by adopting a multi-mode prediction methodSelecting a proper reference frame and a weighted prediction coefficient by optimizing and selecting the multi-mode reference prediction, and transmitting an encoding control parameter to a decoder;
2. The method according to claim 1, wherein the specific semantic object d detected in step (1) is a deep learning detectornThe semantic objects comprise pedestrians, vehicles and human faces, and are subjected to compression, retrieval, dual-target-driven feature extraction, representation and encoding, and deconvolution to obtain decoding reconstruction objects
3. The method as claimed in claim 1, wherein the prediction residual c is directly encoded by the semantic object specified in step (4)nIs set to 0.
4. A decoding method for processing the multi-level monitoring video high-efficiency compression coding method of claim 1, characterized in that the decoding method of the specific semantic object is to use a decoder to decode and obtain the key point sequence structured information and deconvolute the object obtained by decoding, and study the decoding version of the target object of the adjacent frame reconstructed by interpolation through a geometric methodThe decoding method of the long-term background frame is that the decoder reconstructs the background frame by utilizing the scene category index number of the long-term background frame
The decoding method of the short-term background frame is that a decoder obtains by using a reference frame and a weighted prediction coefficient
6. a multi-level monitoring video high-efficiency compression coding device is characterized by comprising a specific semantic object coding module, a long-term background frame modeling module, a short-term background frame modeling module and a foreground coding module,
the specific semantic object coding module detects a specific semantic object dnReconstructing the objectFinding a time domain track through a tracking technology, detecting key point sequences for all versions of a target object on the time domain track through a key point detection technology, and transmitting key point sequence structured information to a decoder;
the long-term background frame modeling module sets a plurality of scene categories, and distinguishes the scene categories by using background frame scene index sequence numbers; constructing a background frame a by an offline training method during camera installationn(ii) a Detecting scene classes in an encoderObtaining a long-term background frameTransmitting the background frame index sequence number to a decoder;
the short-term background frame modeling module assumes the current frame fnPreceding adjacent frame fn-1fn-2A decoded reconstructed version has been obtainedObtaining the short-term background frame prediction value of the current frame by adopting a multi-mode prediction methodSelecting a proper reference frame and a weighted prediction coefficient by optimizing and selecting the multi-mode reference prediction, and transmitting an encoding control parameter to a decoder;
7. The apparatus according to claim 6, wherein the semantic object-specific coding module detects the semantic object d by a deep learning detectornThe semantic objects comprise pedestrians, vehicles and human faces, and are subjected to compression, retrieval, dual-target-driven feature extraction, representation and encoding, and deconvolution to obtain decoding reconstruction objects
8. According toThe apparatus of claim 6, wherein the semantic object is a direct prediction residual cnIs set to 0.
9. A decoding device for processing the multi-level monitoring video high-efficiency compression coding method of claim 1 is characterized by comprising an object decoding module, a long-term background frame decoding module, a short-term background frame decoding module and a foreground decoding module, wherein the object decoding module obtains key point sequence structured information by decoding through a decoder and an object obtained by deconvolution decoding, and a decoding version of an adjacent frame target object is reconstructed by interpolation through a geometric method
The long-term background frame decoding module reconstructs the long-term background frame by using the scene category index number of the long-term background frame
The short-term background frame decoding module obtains a short-term background frame by using a reference frame and a weighted prediction coefficient
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910488842.9A CN110113616B (en) | 2019-06-05 | 2019-06-05 | Multi-level monitoring video efficient compression coding and decoding device and method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910488842.9A CN110113616B (en) | 2019-06-05 | 2019-06-05 | Multi-level monitoring video efficient compression coding and decoding device and method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110113616A CN110113616A (en) | 2019-08-09 |
CN110113616B true CN110113616B (en) | 2021-06-01 |
Family
ID=67494175
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910488842.9A Active CN110113616B (en) | 2019-06-05 | 2019-06-05 | Multi-level monitoring video efficient compression coding and decoding device and method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110113616B (en) |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110866041B (en) * | 2019-09-30 | 2023-05-30 | 视联动力信息技术股份有限公司 | Query method and device for monitoring camera of visual network |
CN110677624B (en) * | 2019-10-21 | 2020-09-18 | 浙江大学 | Monitoring video-oriented foreground and background parallel compression method based on deep learning |
CN111654724B (en) * | 2020-06-08 | 2021-04-06 | 上海纽菲斯信息科技有限公司 | Low-bit-rate coding transmission method of video conference system |
CN111918071A (en) * | 2020-06-29 | 2020-11-10 | 北京大学 | Data compression method, device, equipment and storage medium |
CN112115860A (en) * | 2020-09-18 | 2020-12-22 | 深圳市威富视界有限公司 | Face key point positioning method and device, computer equipment and storage medium |
CN112634303B (en) * | 2020-12-29 | 2022-02-25 | 北京深睿博联科技有限责任公司 | Method, system, device and storage medium for assisting blind person in visual reconstruction |
CN113315972B (en) * | 2021-05-19 | 2022-04-19 | 西安电子科技大学 | Video semantic communication method and system based on hierarchical knowledge expression |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101742319A (en) * | 2010-01-15 | 2010-06-16 | 北京大学 | Background modeling-based static camera video compression method and background modeling-based static camera video compression system |
CN103283226A (en) * | 2010-12-30 | 2013-09-04 | 派尔高公司 | Searching recorded video |
CN104301735A (en) * | 2014-10-31 | 2015-01-21 | 武汉大学 | Method and system for global encoding of urban traffic surveillance video |
CN107341445A (en) * | 2017-06-07 | 2017-11-10 | 武汉大千信息技术有限公司 | The panorama of pedestrian target describes method and system under monitoring scene |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8848802B2 (en) * | 2009-09-04 | 2014-09-30 | Stmicroelectronics International N.V. | System and method for object based parametric video coding |
-
2019
- 2019-06-05 CN CN201910488842.9A patent/CN110113616B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101742319A (en) * | 2010-01-15 | 2010-06-16 | 北京大学 | Background modeling-based static camera video compression method and background modeling-based static camera video compression system |
CN103283226A (en) * | 2010-12-30 | 2013-09-04 | 派尔高公司 | Searching recorded video |
CN104301735A (en) * | 2014-10-31 | 2015-01-21 | 武汉大学 | Method and system for global encoding of urban traffic surveillance video |
CN107341445A (en) * | 2017-06-07 | 2017-11-10 | 武汉大千信息技术有限公司 | The panorama of pedestrian target describes method and system under monitoring scene |
Non-Patent Citations (2)
Title |
---|
基于视频的车辆检测及车牌识别系统的研究;张晓晶;《中国优秀硕士学位论文全文数据库》;20110831;全文 * |
视频序列中运动物体分割的研究;何毓知;《中国优秀硕士学位论文全文数据库》;20100131;全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN110113616A (en) | 2019-08-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110113616B (en) | Multi-level monitoring video efficient compression coding and decoding device and method | |
Que et al. | Voxelcontext-net: An octree based framework for point cloud compression | |
Duan et al. | Compact descriptors for video analysis: The emerging MPEG standard | |
US10390040B2 (en) | Method, apparatus, and system for deep feature coding and decoding | |
US9846820B2 (en) | Method and system for coding or recognizing of surveillance videos | |
CN101389029B (en) | Method and apparatus for video image encoding and retrieval | |
Zhang et al. | An efficient coding scheme for surveillance videos captured by stationary cameras | |
CN103067702B (en) | Video concentration method used for video with still picture | |
EP4373086A1 (en) | Image processing method and apparatus, medium, and electronic device | |
CA2692250A1 (en) | Video encoding and decoding methods using residual prediction, and corresponding apparatuses | |
CN104837031B (en) | A kind of method of high-speed adaptive extraction key frame of video | |
EP3818502A1 (en) | A method, an apparatus and a computer program product for image compression | |
CN103020138A (en) | Method and device for video retrieval | |
CN1926879A (en) | A video signal encoder, a video signal processor, a video signal distribution system and methods of operation therefor | |
CN110677624A (en) | Monitoring video-oriented foreground and background parallel compression method based on deep learning | |
KR102090775B1 (en) | method of providing extraction of moving object area out of compressed video based on syntax of the compressed video | |
CN103533353B (en) | A kind of near video coding system | |
Qian et al. | Video text detection and localization in intra-frames of H. 264/AVC compressed video | |
Ouyang et al. | The comparison and analysis of extracting video key frame | |
CN111723735A (en) | Pseudo high bit rate HEVC video detection method based on convolutional neural network | |
Shinohara et al. | Video compression estimating recognition accuracy for remote site object detection | |
Zhang et al. | Macro-block-level selective background difference coding for surveillance video | |
CN116886922A (en) | Video processing method, video processing device, electronic equipment and computer readable storage medium | |
CN113205010B (en) | Intelligent disaster-exploration on-site video frame efficient compression system and method based on target clustering | |
Huang et al. | An efficient coding framework for compact descriptors extracted from video sequence |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |