CN105357523A - High-order singular value decomposition (HOSVD) algorithm based video compression system and method - Google Patents
High-order singular value decomposition (HOSVD) algorithm based video compression system and method Download PDFInfo
- Publication number
- CN105357523A CN105357523A CN201510677893.8A CN201510677893A CN105357523A CN 105357523 A CN105357523 A CN 105357523A CN 201510677893 A CN201510677893 A CN 201510677893A CN 105357523 A CN105357523 A CN 105357523A
- Authority
- CN
- China
- Prior art keywords
- video
- hosvd
- memory
- frame
- compression system
- 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.)
- Granted
Links
Abstract
The invention relates to a high-order singular value decomposition (HOSVD) algorithm based video compression system and method. The video compression system comprises a camera device and a corresponding sensor. The camera device and the corresponding sensor are respectively connected with a complex programmable logic device (CPLD) processor and a first in first out (FIFO) memory. The CPLD processor and the FIFO memory are connected with a digital signal processor (DSP). The CPLD processor is connected with the FIFO memory and an audio video standard (AVS) coder separately. The AVS coder is connected with a read only memory (ROM) video memory. According to a tensor decomposition theory, hardware and software are combined, so that time domain and spatial domain redundant information is compressed, data storage capacity is compressed furthest, and effective storage of color videos and minimum information loss are realized.
Description
Technical field
The present invention relates to monitored video compression technical field, be specifically related to a kind of based on HOSVD algorithm video compression system and method.
Background technology
In some public places as residential quarter, bank, airport waiting room, shopping center, traffic intersection, office building, campus and residential quarter etc., watch-dog is very universal, the lives and properties that this one side is people and public safety provide guarantee, simultaneously also for relevant unit adds administrative burden and financial burden: the monitor video flow data of 24 hours, annual more than 360 days every day, add the costs such as a large amount of labor managements, plant maintenance, data-storing and data processing, for unit brings huge financial burden to monitoring unit.Some units, for cutting down expenses, deliberately close watch-dog, or allow watch-dog locate idle state for a long time, and watch-dog performs practically no function, and leave many potential safety hazards.At present, the monitoring on market comprises the supervisory control systems such as closed-circuit control, video monitoring, video recording and picture control, remote monitoring, infrared monitoring and network monitoring.Under normal circumstances, every platform watch-dog needs continuous operation 86400 hours in 1 year, for high definition 720P(1280 × 720) form single channel video, within its 24 hours, store 1 day the video data volume and be about 42G, within 1 year, (calculating by 365 days) data volume is about 15TB (15330GB), takies huge memory space, requires high to storage facilities, process and search data and take time and effort, data preserve difficulty.But as carried out later stage excess compression to monitor data, then video pictures quality can be badly damaged, and greatly reduces the efficient of video monitoring.On the other hand, a large amount of lengthy and jumbled information that supervisory control system obtains produces interference to main information, adds the difficulty of data analysis, makes the amount of information that need obtain occur too much redundancy, brings challenge to information preservation and data analysis.
The implementation the most close with the present invention, i.e. AVS compress technique [1].
AVS core technology be decoded by entropy, reorder, the nucleus module such as inverse transformation and inverse quantization, inter prediction, infra-frame prediction and loop filtering represents.
A) entropy decoding: adopt adaptive variable length coding techniques removing data redundancy;
B) reorder: also claim inverse scan, mainly the encoding block residual error coefficient parsed from one-dimensional transform for two dimension;
C) inverse transformation and inverse quantization: replace discrete cosine transform (DCT) with integer transform, smallest blocks prediction is based on 8x8 Integer DCT Transform matrix.Inverse transformation comprises horizontal and vertical and converts two kinds;
D) inter prediction: utilize frame of video correlation and temporal correlation to realize frame number compression, utilizing last decoded picture as the reference diagram of current encoded image, select reference sample in reference diagram, is the basic thought of inter prediction;
E) infra-frame prediction: for removing spatial redundancy in frame, improves inter-coded macroblocks code efficiency.AVS technology utilizes two field picture neighbor correlation (using the left neighbour of current block and upper adjacent pixel as reference pixel) to realize infra-frame prediction;
F) loop filtering: be used for eliminating blocking effect, improve frame quality and code efficiency.Because block-based encoding and decoding easily cause blocking artifact.Deblocking effect is put into coding closed loop by loop filtering, to raise the efficiency.
The present invention relates generally to the HOSVD decomposition algorithm technology of 3 rank or 4 rank tensors, requirement makes decomposition coefficient tool sparsity structure, and HOSVD algorithm is applied in video compression, realize the compressed in layers strategy of video data, better effects can be obtained in Video coding compression.
The form that tensor is number, vector sum matrix is expanded to higher-dimension by low-dimensional in institutional framework, the data point gathered in reality such as colour picture sequence or video flowing etc. are high dimensional data, use traditional 2 rank tensors (matrix) to represent the dimension adding data point, also destroy the internal structure of raw data points (frame) simultaneously.Tensor representation keeps data point inner structural features by the method retaining raw data points rank (dimension), as pixel adjacency, outline line continuity and foreground object is connective and globality etc.Use tensor HOSVD to decompose and can greatly reduce data storage capacity, and effectively denoising, realize effective compression of video data.
AVS technological deficiency and tensor resolution technical advantage:
1.AVS technology is disposable compression, and data to be compressed occupy a large amount of buffer memory; And tensor resolution technology can realize Real Time Compression, avoid data and take buffer memory;
2, AVS technology utilizes the temporal correlation of frame to reach the object of compression, and tensor resolution considers the similarity between sequential frame image under time correlation prerequisite simultaneously, improves compression accuracy;
3, AVS technology lacks the method for effectively searching abnormity point, and tensor resolution method carries out real-time mark to abnormal nodes (frame), easy-to-look-up and early warning;
4, in AVS technology, frame is preserved with matrix (namely frame is for gray-scale map) form, reduces video definition and identification; And frame is preserved with three-dimensional tensor in tensor resolution technology, greatly increase image definition and identification.
The present invention passes through the segmentation chronologically of video every day, and per period video pictures frame is identified automatically, to judge the similitude of frame in Different periods video flowing similitude and section, to determine to retain period video and this period reservation frame number, realize time domain and spatial information (si) compression, reduce data-storing amount, thus reduce monitoring cost, realize information optimized database restore; By the identification to conspicuousness change node frame, timely early warning, for monitor staff provides convenience, avoids the worry that monitor staff uninterruptedly has office hours for 24 hours.
The terminological interpretation that the present invention relates to:
1. video compression: refer to not lose under useful information prerequisite downscaled video data volume with the storage space reduced shared by it and improve its transmission, Storage and Processing speed or (by certain algorithm) recombinate to reduce a kind of technical method of information redundancy to data;
2. Real Time Compression: Information Compression prior to or be synchronized with a data compression technique of information storage, mainly realize based on the real-time segmentation of video and conspicuousness node identification;
3. tensor: a kind of organizational form of high dimensional data and method, can be used for the polyteny function of the linear relationship represented between some vectors, other tensors of scalar sum, also can be used to carry out effective expression to higher-dimension array;
4. frame tensor: the every frame Picture Showing in color video is 3 rank tensor A of m × n × 3, it is by 3 matrix A (::, 1), A (::, 2), A (:,:, 3) form, represent the monochromatic matrix that red, green and blue look corresponding, product successively
mnfor the pixel value of every frame picture;
5.HOSVD decomposes: high order tensor singular value decomposition, this refers to a 3(or 4) rank tensor resolution becomes a 3(or 4) rank core tensor and 3(or 4) product of individual orthogonal matrix.
Citing document:
[1] Li Xiaoyu. AVS video decoding system [D] in real time. University of Electronic Science and Technology, 2009.
[2]H.Lu,K.N.Plataniotis,A.Venetsanopoulos.MultilinearSubspaceLearning:DimensionalityReductionofMultidimensionalData[D].CRCPress,2013.
Summary of the invention
The object of the invention is to overcome prior art Problems existing, provide a kind of based on HOSVD algorithm video compression system and method.
For realizing above-mentioned technical purpose, reach above-mentioned technique effect, the present invention is achieved through the following technical solutions:
A kind of based on HOSVD algorithm video compression system and method, this video compression system comprises: camera device and respective sensor, described camera device and respective sensor are connected to CPLD processor and FIFO memory, described CPLD processor is connected DSP with FIFO memory, described CPLD processor is connected with AVS encoder with FIFO memory respectively, and described AVS encoder is connected with ROM video memory;
This video-frequency compression method comprises the following steps:
Step 1) builds CPLD processor, by MATLAB Integrated Development software platform, is identified and compress technique algorithm, generate corresponding file destination, be sent to by code in CPLD processor chips, realize the digital system of design by schematic diagram and MATLAB design;
Step 2) camera device and respective sensor gather image information and stored in FIFO memory, the pixel data collected is sent in the DSP based on MATLAB by CPLD processor from FIFO memory, prepares the HOSVD algorithm signal transacting carried out based on MATLAB.
Step 3) DSP carries out HOSVD decomposition and data recombination compression to original video stream, removes time redundancy, realizes prospect, background separation simultaneously;
Video streaming is carried out further compression coding to AVS encoder by DSP serial communication module by step 4), is finally stored in ROM memory.
Further, in described step 3), HOSVD decomposes employing high order tensor part singular value decomposition, uses sparse tensor representation, between computation complexity and data compression, reaches balance.
Further, the mode that in described step 3), prospect and background separation adopt low-rank to approach, realizes data and compresses further and denoising.
Further, in described step 3), tensor resolution adopts the mode of decomposing overall video flowing, namely carry out process and interframe process in frame simultaneously, a two field picture is often collected at camera device and respective sensor, based on the DSP of MATLAB this two field picture and existing frame carried out similitude contrast and delete the much higher remaining frame of similarity, remain with difference frame, restore in an interim flash storage, carry out next frame IMAQ, continue above step, until the setting period terminates.
Further, described DSP is circumscribed with RAM memory and flash storage as temporary storage, carries out data buffer storage storage, for calling the frame information closed on when HOSVD decomposes.
The invention has the beneficial effects as follows:
1, use HOSVD resolution theory, realize time domain and spatial redundancy Information Compression, maximum compression data-storing amount, realize effective preservation and the minimum loss of information of color video;
2, high order tensor storage and HOSVD decompose the effective preservation that ensure that color video data;
3, front background separation and frame tensor similarity-rough set are conducive to abnormity point and search and compress with video stream data;
4, Real Time Compression contributes to timely Dynamic Discovery conspicuousness node, and automatic alarm, for monitor staff provides convenience, avoid the worry that monitor staff uninterruptedly has office hours for 24 hours.
Accompanying drawing explanation
Fig. 1 (a) is the facial image preserved with 3 rank tensors;
Fig. 1 (b) is the image sequence generated through Gabor filtering;
Fig. 1 (c) is the brain structure chart preserved with 4 rank tensors;
Fig. 2 is system architecture diagram of the present invention.
Embodiment
Below with reference to the accompanying drawings and in conjunction with the embodiments, describe the present invention in detail.
Fig. 1 (a) is a width 3D facial image; B () is the image sequence generated through Gabor filtering; And (c) is video capture tool 3D effect brain structure chart; They are preserved with 3 rank, 3 rank and 4 rank tensors successively;
Following formula display tensor HOSVD decomposition principle:
Wherein A is N rank tensors, and the right S is core tensor (being generally sparse),
represent the orthogonal matrix (projection) acted in i-th dimension of A.
Fig. 2 is a kind of based on HOSVD algorithm video compression system and method, this video compression system comprises: camera device and respective sensor, described camera device and respective sensor are connected to CPLD processor and FIFO memory, described CPLD processor is connected DSP with FIFO memory, described CPLD processor is connected with AVS encoder with FIFO memory respectively, and described AVS encoder is connected with ROM video memory;
This video-frequency compression method comprises the following steps:
Step 1) builds CPLD processor, by Integrated Development software platform, is identified and isolation technics algorithm, generate corresponding file destination, be sent to by code in CPLD processor chips, realize the digital system of design by schematic diagram or hardware description language design;
Step 2) camera device and respective sensor gather image information and stored in FIFO memory, the pixel data collected is sent in DSP by CPLD processor from FIFO memory, prepares the signal transacting carried out based on tensor algorithm.
Step 3) DSP carries out tensor resolution, data recombination compression to original video stream, removes temporal redundancy field, and by prospect and background separation;
Video streaming is carried out further compression coding to AVS encoder by the serial communication module of DSP by step 4), is finally stored in ROM memory.
The 4 rank tensor part singular value decomposition that in described step 3), HOSVD algorithm adopts the frame sequence upgraded in time to generate extract the first principal component of frame set, generation background, realize prospect and background separation, prospect sequence generates sparse tensor, if equipment is rotatable, then generation background sequence, and synthesize panorama background.
Low-rank is adopted to approach to the prospect sequence tensor after separation in described step 3).
Calculate incoming frame (time interval can be 1s or 5s) in described step 3) and preserve consecutive frame (tensor) similarity, determining whether to preserve this incoming frame (stored in temporary storage), carry out next frame IMAQ, realize Real Time Compression.
Described DSP is circumscribed with RAM memory and flash storage as temporary storage, carries out data buffer storage storage, for calling the frame information closed on when tensor resolution.
The foregoing is only the preferred embodiments of the present invention, be not limited to the present invention, for a person skilled in the art, the present invention can have various modifications and variations.Within the spirit and principles in the present invention all, any amendment done, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.
Claims (5)
1. one kind based on HOSVD algorithm video compression system and method, it is characterized in that, this video compression system comprises: camera device and respective sensor, described camera device and respective sensor are connected to CPLD processor and FIFO memory, described CPLD processor is connected DSP with FIFO memory, described CPLD processor is connected with AVS encoder with FIFO memory respectively, and described AVS encoder is connected with ROM video memory;
This video-frequency compression method comprises the following steps:
Step 1) builds CPLD processor, by MATLAB Integrated Development software platform, is identified and compress technique algorithm, generate corresponding file destination, be sent to by code in CPLD processor chips, realize the digital system of design by schematic diagram and MATLAB design;
Step 2) camera device and respective sensor gather image information and stored in FIFO memory, the pixel data collected is sent in the DSP based on MATLAB by CPLD processor from FIFO memory, prepares the HOSVD algorithm signal transacting carried out based on MATLAB;
Step 3) DSP carries out HOSVD decomposition and data recombination compression to original video stream, removes time redundancy, realizes prospect, background separation simultaneously;
Video streaming is carried out further compression coding to AVS encoder by DSP serial communication module by step 4), is finally stored in ROM memory.
2. according to claim 1 based on HOSVD algorithm video compression system and method, it is characterized in that, in described step 3), HOSVD decomposes employing high order tensor part singular value decomposition, uses sparse tensor representation, between computation complexity and data compression, reaches balance.
3. the video compression system based on HOSVD algorithm according to claim 1 and method, is characterized in that, the mode that in described step 3), prospect and background separation adopt low-rank to approach.
4. the video compression system based on HOSVD algorithm according to claim 1 and method, it is characterized in that, in described step 3), tensor resolution adopts the mode of decomposing overall video flowing, namely carry out process and interframe process in frame simultaneously, a two field picture is often collected at camera device and respective sensor, based on the DSP of MATLAB this two field picture and existing frame carried out similitude contrast and delete the much higher remaining frame of similarity, remain with difference frame, restore in an interim flash storage, carry out next frame IMAQ, continue above step, until the setting period terminates.
5. the video compression system based on HOSVD algorithm according to claim 4 and method, it is characterized in that, described DSP is circumscribed with RAM memory and flash storage as temporary storage, carry out data buffer storage storage, for calling the frame information closed on when HOSVD decomposes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510677893.8A CN105357523B (en) | 2015-10-20 | 2015-10-20 | One kind being based on HOSVD algorithm video compression system and method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510677893.8A CN105357523B (en) | 2015-10-20 | 2015-10-20 | One kind being based on HOSVD algorithm video compression system and method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105357523A true CN105357523A (en) | 2016-02-24 |
| CN105357523B CN105357523B (en) | 2019-02-19 |
Family
ID=55333371
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510677893.8A Expired - Fee Related CN105357523B (en) | 2015-10-20 | 2015-10-20 | One kind being based on HOSVD algorithm video compression system and method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105357523B (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106961575A (en) * | 2017-02-24 | 2017-07-18 | 深圳汇创联合自动化控制有限公司 | A kind of efficient video monitoring system |
| CN107155111A (en) * | 2017-06-05 | 2017-09-12 | 李益永 | A kind of video-frequency compression method and device |
| CN107507253A (en) * | 2017-08-15 | 2017-12-22 | 电子科技大学 | Based on the approximate more attribute volume data compression methods of high order tensor |
| CN107528672A (en) * | 2017-09-05 | 2017-12-29 | 北京航空航天大学 | A kind of efficient wireless data transceiving method and device |
| CN107886560A (en) * | 2017-11-09 | 2018-04-06 | 网易（杭州）网络有限公司 | The processing method and processing device of animation resource |
| CN109997130A (en) * | 2016-11-23 | 2019-07-09 | 韩华泰科株式会社 | Video frequency searching device, date storage method and data storage device |
| CN111866443A (en) * | 2019-04-25 | 2020-10-30 | 黄河 | Video stream data storage method, device, system and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101957909A (en) * | 2009-07-15 | 2011-01-26 | 青岛科技大学 | Digital signal processor (DSP)-based face detection method |
| US20110187916A1 (en) * | 2010-02-02 | 2011-08-04 | Samsung Electronics Co., Ltd. | Apparatus for processing digital image and method of controlling the same |
| CN102223520A (en) * | 2011-04-15 | 2011-10-19 | 北京易子微科技有限公司 | Intelligent face recognition video monitoring system and implementation method thereof |
| CN102970546A (en) * | 2012-12-13 | 2013-03-13 | 中国航空无线电电子研究所 | Video encoding unit and realizing method thereof |
-
2015
- 2015-10-20 CN CN201510677893.8A patent/CN105357523B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101957909A (en) * | 2009-07-15 | 2011-01-26 | 青岛科技大学 | Digital signal processor (DSP)-based face detection method |
| US20110187916A1 (en) * | 2010-02-02 | 2011-08-04 | Samsung Electronics Co., Ltd. | Apparatus for processing digital image and method of controlling the same |
| CN102223520A (en) * | 2011-04-15 | 2011-10-19 | 北京易子微科技有限公司 | Intelligent face recognition video monitoring system and implementation method thereof |
| CN102970546A (en) * | 2012-12-13 | 2013-03-13 | 中国航空无线电电子研究所 | Video encoding unit and realizing method thereof |
Non-Patent Citations (1)
| Title |
|---|
| 钱莉，等: "视频监控现状调查与改进方案", 《电脑知识与技术》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109997130A (en) * | 2016-11-23 | 2019-07-09 | 韩华泰科株式会社 | Video frequency searching device, date storage method and data storage device |
| CN109997130B (en) * | 2016-11-23 | 2023-10-13 | 韩华视觉株式会社 | Video search device, data storage method, and data storage device |
| CN106961575A (en) * | 2017-02-24 | 2017-07-18 | 深圳汇创联合自动化控制有限公司 | A kind of efficient video monitoring system |
| CN107155111A (en) * | 2017-06-05 | 2017-09-12 | 李益永 | A kind of video-frequency compression method and device |
| CN107155111B (en) * | 2017-06-05 | 2020-02-18 | 李益永 | Video compression method and device |
| CN107507253A (en) * | 2017-08-15 | 2017-12-22 | 电子科技大学 | Based on the approximate more attribute volume data compression methods of high order tensor |
| CN107507253B (en) * | 2017-08-15 | 2020-09-01 | 电子科技大学 | Multi-attribute body data compression method based on high-order tensor approximation |
| CN107528672A (en) * | 2017-09-05 | 2017-12-29 | 北京航空航天大学 | A kind of efficient wireless data transceiving method and device |
| CN107886560A (en) * | 2017-11-09 | 2018-04-06 | 网易（杭州）网络有限公司 | The processing method and processing device of animation resource |
| CN107886560B (en) * | 2017-11-09 | 2021-05-25 | 网易（杭州）网络有限公司 | Animation resource processing method and device |
| CN111866443A (en) * | 2019-04-25 | 2020-10-30 | 黄河 | Video stream data storage method, device, system and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105357523B (en) | 2019-02-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105357523B (en) | One kind being based on HOSVD algorithm video compression system and method | |
| US9846820B2 (en) | Method and system for coding or recognizing of surveillance videos | |
| US20160050440A1 (en) | Low-complexity depth map encoder with quad-tree partitioned compressed sensing | |
| CN106664411B (en) | Motion compensated segmentation | |
| CN110830803B (en) | Image compression method combining block matching and string matching | |
| Gao et al. | Digital retina: A way to make the city brain more efficient by visual coding | |
| CN106998470A (en) | Coding/decoding method, coding method, decoding device and encoding device | |
| CN103020138A (en) | Method and device for video retrieval | |
| Canh et al. | Rate-distortion optimized quantization: A deep learning approach | |
| CA3222179A1 (en) | Feature data encoding and decoding method and apparatus | |
| Wu et al. | Memorize, then recall: a generative framework for low bit-rate surveillance video compression | |
| WO2017092072A1 (en) | Distributed video encoding framework | |
| CN105930814A (en) | Method for detecting personnel abnormal gathering behavior on the basis of video monitoring platform | |
| CN113591681A (en) | Face detection and protection method and device, electronic equipment and storage medium | |
| Cossalter et al. | Privacy-enabled object tracking in video sequences using compressive sensing | |
| CN110427904B (en) | Mall monitoring system, method and device based on pedestrian re-identification | |
| CN115209147A (en) | Camera video transmission bandwidth optimization method, device, equipment and storage medium | |
| US11503292B2 (en) | Method and apparatus for encoding/decoding video signal by using graph-based separable transform | |
| KR20200068102A (en) | method of providing object classification for compressed video by use of syntax-based MRPN-CNN | |
| CN109862363A (en) | The second-compressed method and its compressibility of video | |
| CN106162196A (en) | A kind of video coding framework towards intellectual analysis and method | |
| CN116939170B (en) | Video monitoring method, video monitoring server and encoder equipment | |
| Wang et al. | A surveillance video compression algorithm based on regional dictionary | |
| Chatterjee et al. | Adaptive Filtering and Voice Compression Using Neural Networks | |
| Ramasamy et al. | Detecting background setting for dynamic scene |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190219 Termination date: 20191020 |