CN108717703B - HEVC-based moving target detection and tracking method - Google Patents

HEVC-based moving target detection and tracking method Download PDF

Info

Publication number
CN108717703B
CN108717703B CN201810252619.XA CN201810252619A CN108717703B CN 108717703 B CN108717703 B CN 108717703B CN 201810252619 A CN201810252619 A CN 201810252619A CN 108717703 B CN108717703 B CN 108717703B
Authority
CN
China
Prior art keywords
motion
block
motion vector
hevc
target
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
Application number
CN201810252619.XA
Other languages
Chinese (zh)
Other versions
CN108717703A (en
Inventor
姚英彪
杨旭
姜显扬
刘兆霆
刘晴
许晓荣
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hangzhou Dianzi University
Original Assignee
Hangzhou Dianzi University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hangzhou Dianzi University filed Critical Hangzhou Dianzi University
Priority to CN201810252619.XA priority Critical patent/CN108717703B/en
Publication of CN108717703A publication Critical patent/CN108717703A/en
Application granted granted Critical
Publication of CN108717703B publication Critical patent/CN108717703B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/223—Analysis of motion using block-matching
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/207—Analysis of motion for motion estimation over a hierarchy of resolutions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30241—Trajectory

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Compression Or Coding Systems Of Tv Signals (AREA)

Abstract

The invention discloses a moving target detection and tracking method based on HEVC. The method comprises the following steps: step 1: extracting a coding unit structure, a motion vector and a prediction mode of a current frame coded video from an HEVC code stream; step 2: normalizing the size of the coding block; and step 3: determining a motion macro block; and 4, step 4: calculating the weight of the motion vector of the motion macro block; and 5: detecting a moving target; step 6: classifying the moving target; and 7: and tracking the moving target. The invention utilizes the information of motion vectors, coding unit division modes, prediction modes and the like generated in the HEVC coding and decoding process to detect and track the moving target, avoids unnecessary digital image processing process, thereby reducing the complexity of calculation and rapidly detecting and tracking the moving target on the basis of not damaging the original compressed video data.

Description

HEVC-based moving target detection and tracking method
Technical Field
The invention belongs to the technical field of video processing, and particularly relates to a moving target detection and tracking method based on HEVC.
Background
In the face of the increasing frequency of road traffic accidents, scientists around the world have made extensive and intensive studies on the existing traffic systems, and then intelligent traffic systems have been proposed as a solution and rapidly developed. At present, moving object detection and tracking in an intelligent traffic system mainly refers to detecting and tracking moving tracks of automobiles and pedestrians running on roads.
The moving target detection mainly means that a moving target is segmented from a complex dynamic background, and can be understood as a classification problem, namely information in a video image is correctly classified according to prior knowledge so as to achieve the purpose of target identification. The moving target tracking mainly refers to the step of carrying out target recognition on each frame of a video, so that the continuous acquisition of the moving track and the posture of a moving target is realized.
In the field of intelligent transportation systems, various target detection and tracking technologies are also being continuously generated and developed, such as acoustic detection, induction coil detection, infrared video detection, microwave detection, video detection and the like. Compared with other modes, the video-based target detection and tracking technology has the advantages of multiple parameters, low investment cost, no need of traffic interruption during installation and maintenance, easiness in traffic scene reproduction and the like.
At present, the more classical video-based moving object detection and tracking is mainly performed in a pixel domain, which generally analyzes texture information, motion information and temporal-spatial neighborhood information of each pixel, and uses these abundant feature information to find out an interested moving object in a video image by means of digital image processing. The current detection technology of moving objects in a pixel domain mainly comprises methods based on a background difference method, an interframe difference method, an optical flow method and the like, and the tracking technology of the moving objects mainly comprises methods based on motion estimation, mean shift, template matching and the like. Due to some self factors of an image processing algorithm, the pixel domain method faces some bottlenecks, the detection effect is not ideal when moving objects face special conditions, and meanwhile, due to too much pixel information, a large amount of calculation time and storage space are needed.
In recent years, with the continuous development of digital video compression technology, the video codec format standard gradually transits from the former MPEG to the mainstream HEVC format. The HEVC format is one of the internationally recognized video coding standards today, with a doubling of the coding efficiency compared to the h.264 format. The infrastructure of HEVC remains consistent with h.264/AVC, but some new techniques are also introduced, such as improved motion vector prediction, new intra prediction modes: MERGE mode, inter-prediction asymmetric motion partitioning, quad-tree transformation, DCT-based sub-pixel filters, new multi-reference frame management approach, simplified deblocking filtering, new loop filter SAO, and high throughput entropy and content-adaptive binary arithmetic coding CABAC, among others. Based on the above features, more and more applications are turning to the adoption of HEVC instead of h.264, and HEVC has become the mainstream standard of video coding and decoding nowadays. In many practical applications today, the video source format is typically the HEVC encoded source format. A section of video often generates more reliable information such as a Motion Vector (MV), a prediction mode, a DCT residual coefficient, etc. in the encoding and decoding process, many scholars begin to research on extracting a moving object by directly using the information in the video encoding and decoding process, which is completely different from the previous pixel domain method and is called a moving object detection and tracking method based on a compressed domain.
The HEVC encoder adopts a Coding Tree Unit (CTU), a Coding Unit (CU), a Prediction Unit (PU), and a Transform Unit (TU) structure, so that HEVC can encode videos of different resolutions and application environments. In HEVC coding, a picture may be divided into a plurality of mutually non-overlapping CTUs, each CTU is divided into one or more CUs, and the CU division mode is as shown in fig. 1; when intra-frame or inter-frame prediction coding is performed, a CU can be selectively divided into one or more PUs, and the PU division mode is shown in fig. 2; in a transform, quantization operation, one CU may be divided into one or more TUs.
Disclosure of Invention
The invention aims to provide a moving target detection and tracking method based on HEVC. The invention utilizes the information of motion vectors, coding unit division modes, prediction modes and the like generated in the HEVC coding and decoding process to detect and track the moving target, avoids unnecessary digital image processing process, thereby reducing the complexity of calculation and rapidly detecting and tracking the moving target on the basis of not damaging the original compressed video data.
In order to achieve the purpose, the technical scheme adopted by the invention is as follows:
a moving target detection and tracking method based on HEVC comprises the following steps:
step 1: extracting a coding unit structure, a motion vector and a prediction mode of a current frame coded video from an HEVC code stream;
step 2: normalizing the size of the coding block;
and step 3: determining a motion macro block;
and 4, step 4: calculating the weight of the motion vector of the motion macro block;
and 5: detecting a moving target;
step 6: classifying the moving target;
and 7: and tracking the moving target.
Further, the encoding block in step 2 is a basic unit that handles a coding unit of size 32 × 32 as a motion vector.
Further, the step of determining the motion macroblock in step 3 is:
3.1 after determining that the size of the coding block is 32 multiplied by 32, defining the sub coding unit with the motion vector not being 0 as a motion sub macro block;
3.2 judging whether the moving sub-macro block is the moving macro block according to a threshold limiting method.
Further, the formula for calculating the Motion Vector Weight (MVW) of the motion macroblock in step 4 is as follows:
Figure BDA0001608189100000031
Figure BDA0001608189100000032
wherein, { MViI ∈ σ } represents a set of motion vectors in a coding block, area (PU)i) Denotes the area of the ith PU, and x and y denote the length and width of the motion vector.
Further, the step of detecting the moving object in the step 5 is as follows:
according to the magnitude of the weight (MVW) value of the motion vector, threshold selection is carried out through an Otsu threshold method, and the motion macro block is determined to be a motion target or a non-motion target, and the specific algorithm is as follows:
assuming that the motion vector weight range is divided into [0, L-1], Totsu is a threshold value that divides the image into a motion target block a and a non-motion target block B, the probability distribution of a and B is:
Figure BDA0001608189100000041
Figure BDA0001608189100000042
pia probability representing the weight of the motion vector; the average motion vector weights are:
Figure BDA0001608189100000043
Figure BDA0001608189100000044
where μ is the average motion vector weight of the whole graph, the formula is as follows:
Figure BDA0001608189100000045
the variance of A and B is:
σ2＝ωA(μA-μ)2+(1-ωA)(μB-μ)2
the motion macroblock that makes the variance the largest, i.e., MVW > Totsu, is determined as the motion target block.
Further, the moving object classification method in step 6 is as follows: classifying the automobiles or pedestrians by adopting a Support Vector Machine (SVM); training the SVM by using a training set to obtain a classification model, and predicting a class label of a test set by using the obtained model; each sample contains 5 feature components: CU, PU, MV, MVW and prediction mode; finally, the purpose of classifying the vehicles and the pedestrians is achieved.
Further, the method for tracking the moving object in step 7 includes:
7.1, performing motion estimation on each motion target block of the current frame by adopting a motion target block matching method, predicting the Motion Vector Intensity (MVI) of each block, and adopting the following calculation formula:
MVI＝||MV||2×32×32
7.2 for each motion target block, extracting a motion vector from the block, shifting the motion target block to the current frame according to the motion vector, and predicting the motion target block of the current frame; the motion estimation formula is as follows:
Upred＝Ucur-(m-n)*MV
wherein, UpredRepresenting a predicted moving object block, UcurRepresenting the current moving target block, m and n representing the video frame number;
7.3 verifying whether the MVI of the predicted motion target block is approximate to the real MVI, and judging the formula as follows:
Figure BDA0001608189100000051
epsilon represents the allowable error and successful verification represents successful tracking.
By this point, the entire HEVC-based moving object detection and tracking method ends.
The invention has the beneficial effects that:
the method is based on an HEVC video coding and decoding format, video compression code stream data (HEVC code stream data) to be analyzed is decoded and analyzed, and information such as motion vectors, coding unit division modes, prediction modes and the like contained in the code stream data is extracted; then, integrating and analyzing the data, and extracting the characteristic information of the moving target; and finally, the detection (extraction), classification and tracking of the moving target are realized by means of the characteristic information.
The invention avoids unnecessary digital image processing process, simplifies the complexity of calculation, and has simple and rapid method.
Drawings
Fig. 1 CU partition diagram.
FIG. 2 illustrates PU partition.
Fig. 3 is a flow chart of a HEVC-based moving object detection and tracking method.
Fig. 4 is a diagram of extraction results in an HEVC code stream.
Fig. 5 shows a coding block size normalization process.
Fig. 6 is a schematic diagram of a motion macroblock.
Fig. 7 is a schematic diagram of motion vector weight calculation.
Fig. 8 is a schematic diagram of motion vector intensity calculation.
Detailed Description
The present invention is further illustrated by the following examples, which are not intended to limit the invention to these embodiments. It will be appreciated by those skilled in the art that the present invention encompasses all alternatives, modifications and equivalents as may be included within the scope of the claims.
The invention provides a moving target detection and tracking method based on HEVC. Firstly, HEVC code stream data is decoded and analyzed, and information such as motion vectors, coding unit division modes, prediction modes and the like contained in the code stream data is extracted; then, integrating and analyzing the data, and extracting the characteristic information of the moving target; and finally, the detection (extraction), classification and tracking of the moving target are realized by means of the characteristic information. Meanwhile, unnecessary digital image processing process is avoided, so that the complexity of calculation is reduced, and the rapid moving object detection and tracking are carried out on the basis of not damaging the original compressed video data.
As shown in fig. 3, the following steps are required for specific implementation:
step 1: and extracting a coding unit structure, a motion vector and a prediction mode of a current frame coded video from the HEVC code stream.
The extraction results are shown in fig. 4. Wherein the white square box represents the coding unit structure, the black square box represents the largest coding unit LCU, the black dot diagonal line represents the motion vector, and the curved surrounding area represents the intra prediction mode.
Step 2: and normalizing the size of the coding block.
Since HEVC autonomously determines the sample size of a coding unit according to the characteristics of the input video, such as resolution, content, and picture size, different HEVC videos may have different coding unit sample sizes, and therefore, we must perform normalization processing on the size. The coding unit having a size of 32 × 32 is collectively referred to as a coding block as a basic unit of motion vector processing, and the processing result is as shown in fig. 5.
And step 3: a motion macroblock is determined.
After determining the size of the coding block to be 32 × 32, it is calculated that one coding block contains 64 4 × 4 sub-coding units at most, and we define the sub-coding units with motion vectors different from 0 as motion sub-macroblocks. Whether each coding block is a motion macroblock is weighed according to the number of motion sub-macroblocks contained in the coding block, and the specific method is threshold limitation: the number of sub-coding blocks with motion vectors not equal to 0 is counted through 64 sub-coding blocks, and if the number is not less than 3/4 (i.e. 48) of the total number, the coding block is defined as a motion macroblock, and the determination result is shown in fig. 6.
And 4, step 4: the motion vector weight of the moving macroblock is calculated.
According to observation and analysis, a moving object in a video generally has the characteristics of long motion vector and small coding unit. Therefore, a new feature Motion Vector Weight (MVW) is proposed, which is calculated as follows:
Figure BDA0001608189100000071
Figure BDA0001608189100000072
{MVii ∈ σ } represents a set of motion vectors in a coding block, area (PU)i) Denotes the area of the ith PU, and x and y denote the length and width of the motion vector. The moving object can be known to have a large MVW value through the formula, and the specific calculation is shown in fig. 7.
And 5: and detecting a moving object.
Depending on the magnitude of the Motion Vector Weight (MVW) value, a moving macroblock may be determined as either a moving object or a non-moving object. The Otsu threshold method is adopted for threshold value selection. The specific algorithm is as follows:
let the weight range of motion vector be divided into [0, L-1]]，TotsuWhich is a threshold value, divides the image into a moving target block a and a non-moving target block B. The probability distribution of a and B is:
Figure BDA0001608189100000081
Figure BDA0001608189100000082
pi represents the probability of the motion vector weight. The average motion vector weights are:
Figure BDA0001608189100000083
Figure BDA0001608189100000084
where μ is the average motion vector weight of the whole graph, the formula is as follows:
Figure BDA0001608189100000085
the variance of A and B is:
σ2＝ωA(μA-μ)2+(1-ωA)(μB-μ)2
t maximizing the varianceotsuI.e. the threshold value, i.e. MVW>The moving macroblock of T is determined as a moving target block.
Step 6: and (4) classifying the moving object.
The moving object in the video may be a car or a pedestrian. Here classification is done using a support vector machine SVM. SVMs are developed based on an optimal classification hyperplane of linearly separable samples. The optimal classification hyperplane can correctly distinguish classes in a space containing two types of samples and maximize the separation of the two types of sample points to the plane. For a given linearly separable dataset { xi, yi }, i ═ 1,2,3, …, N, yi ∈ { -1, +1}, xi ∈ Rd, then the classification can be done using hyperplane:
wTx+b＝0
where w is the weight vector and b is the classification threshold.
According to Lagrange duality principle, the following formula can be obtained:
Figure BDA0001608189100000091
Figure BDA0001608189100000092
Figure BDA0001608189100000093
Figure BDA0001608189100000094
ai is a Lagrange factor solved by the quadratic programming optimization problem, n is a support vector number, and C is a penalty factor.
Because the problem is linear inseparable in the definition space, the original input space is input into a new feature space through a kernel function, and the original linear inseparable task is converted into a linear separable task. The resulting formula is as follows:
Figure BDA0001608189100000095
Figure BDA0001608189100000096
Figure BDA0001608189100000097
Figure BDA0001608189100000098
k (xi, yi) is the radial basis function:
K(xi,xj)＝exp(-g||xi-xj||2)
the parameters C and g in the formula are obtained by a grid search method.
And training the SVM by using the training set to obtain a classification model, and predicting the class label of the test set by using the obtained model. Each sample contains 5 feature components: CU, PU, MV, MVW, and prediction mode. Finally, the purpose of classifying the vehicles and the pedestrians is achieved.
And 7: and tracking the moving target.
The moving target tracking is an important link after target detection and is a basis for understanding and analyzing target behaviors. Here, a motion target block matching method is adopted to perform motion estimation on each motion target block of the current frame, and the motion vector intensity MVI of each block is predicted, and the calculation formula is as follows:
MVI＝||MV||2×32×32
for each motion target block, a motion vector is extracted from the block, the motion target block is shifted to the current frame according to the motion vector, and the motion target block of the current frame is predicted. The motion estimation formula is as follows:
Upred＝Ucur-(m-n)*MV
Upredrepresenting a predicted moving object block, UcurRepresenting the current motion target block, and m and n represent the video frame number. And then verifying whether the MVI of the predicted motion target block is approximate to the real MVI. The judgment formula is as follows:
Figure BDA0001608189100000101
epsilon represents the allowable error and successful verification represents successful tracking. The schematic diagram is shown in fig. 8.
The above description is only a preferred embodiment of the present invention, and the scope of the present invention should not be limited thereby, and all the equivalent changes and modifications made by the claims and the content of the specification should be covered by the scope of the present invention.

Claims (4)

1. A moving target detection and tracking method based on HEVC is characterized by comprising the following steps:
step 1: extracting a coding unit structure, a motion vector and a prediction mode of a current frame coded video from an HEVC code stream;
step 2: normalizing the size of the coding block;
and step 3: determining a motion macro block;
and 4, step 4: calculating the weight of the motion vector of the motion macro block;
and 5: detecting a moving target;
step 6: classifying the moving target;
and 7: tracking a moving target;
the encoding block in step 2 is a basic unit for processing a coding unit with a size of 32 × 32 as a motion vector;
the step of determining the motion macro block in step 3 is as follows:
3.1 after determining that the size of the coding block is 32 multiplied by 32, defining the sub coding unit with the motion vector not being 0 as a motion sub macro block;
3.2 judging whether the moving sub-macro block is a moving macro block according to a threshold limiting method;
the step of detecting the moving target in the step 5 is as follows:
according to the value of the motion vector weight MVW, threshold selection is performed through an Otsu threshold method, and a motion macro block is determined as a motion target or a non-motion target, wherein the specific algorithm is as follows:
assuming that the motion vector weight range is divided into [0, L-1], Totsu is a threshold value that divides the image into a motion target block a and a non-motion target block B, the probability distribution of a and B is:
Figure FDA0003193539140000011
Figure FDA0003193539140000021
pi represents the probability of the motion vector weight; the average motion vector weights are:
Figure FDA0003193539140000022
Figure FDA0003193539140000023
where μ is the average motion vector weight of the whole graph, the formula is as follows:
Figure FDA0003193539140000024
the variance of A and B is:
σ2＝ωA(μA-μ)2+(1-ωA)(μB-μ)2
the motion macroblock that makes the variance the largest, i.e., MVW > Totsu, is determined as the motion target block.
2. An HEVC-based moving object detection and tracking method according to claim 1, wherein the formula for calculating motion vector weights MVW of motion macroblocks in step 4 is as follows:
Figure FDA0003193539140000025
Figure FDA0003193539140000026
wherein, { MViI ∈ σ } represents a set of motion vectors in a coding block, area (PU)i) Denotes the area of the ith PU, and x and y denote the length and width of the motion vector.
3. An HEVC-based moving object detection and tracking method as claimed in claim 1, wherein in step 6, the moving object classification method is as follows: classifying the automobiles or pedestrians by adopting a Support Vector Machine (SVM); training the SVM by using a training set to obtain a classification model, and predicting a class label of a test set by using the obtained model; each sample contains 5 feature components: CU, PU, MV, MVW and prediction mode; finally, the purpose of classifying the vehicles and the pedestrians is achieved.
4. An HEVC-based moving object detecting and tracking method according to claim 1, wherein the moving object tracking method in step 7 is:
7.1, performing motion estimation on each motion target block of the current frame by adopting a motion target block matching method, predicting the motion vector intensity MVI of each block, wherein the calculation formula is as follows:
MVI＝||MV||2×32×32
7.2 for each motion target block, extracting a motion vector from the block, shifting the motion target block to the current frame according to the motion vector, and predicting the motion target block of the current frame; the motion estimation formula is as follows:
Upred＝Ucur-(m-n)*MV
wherein, UpredRepresenting a predicted moving object block, UcurRepresenting the current moving target block, m and n representing the video frame number;
7.3 verifying whether the MVI of the predicted motion target block is approximate to the real MVI, and judging the formula as follows:
Figure FDA0003193539140000031
epsilon represents the allowable error and successful verification represents successful tracking.
CN201810252619.XA 2018-03-26 2018-03-26 HEVC-based moving target detection and tracking method Active CN108717703B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810252619.XA CN108717703B (en) 2018-03-26 2018-03-26 HEVC-based moving target detection and tracking method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810252619.XA CN108717703B (en) 2018-03-26 2018-03-26 HEVC-based moving target detection and tracking method

Publications (2)

Publication Number Publication Date
CN108717703A CN108717703A (en) 2018-10-30
CN108717703B true CN108717703B (en) 2021-11-02

Family

ID=63898805

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810252619.XA Active CN108717703B (en) 2018-03-26 2018-03-26 HEVC-based moving target detection and tracking method

Country Status (1)

Country Link
CN (1) CN108717703B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111079567B (en) * 2019-11-28 2020-11-13 中科驭数（北京）科技有限公司 Sampling method, model generation method, video behavior identification method and device
CN111083488B (en) * 2019-12-10 2021-08-31 四川新视创伟超高清科技有限公司 High-speed decoding method for 8K ultra-high-definition video based on CPU

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103607590A (en) * 2013-11-28 2014-02-26 北京邮电大学 High efficiency video coding sensing rate-distortion optimization method based on structural similarity
CN104113754A (en) * 2014-07-17 2014-10-22 四川大学 Method for high-performance video interframe coding based on time domain relevance and transcoder thereof

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103607590A (en) * 2013-11-28 2014-02-26 北京邮电大学 High efficiency video coding sensing rate-distortion optimization method based on structural similarity
CN104113754A (en) * 2014-07-17 2014-10-22 四川大学 Method for high-performance video interframe coding based on time domain relevance and transcoder thereof

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
A RAPID ABNORMAL EVENT DETECTION METHOD FOR SURVEILLANCE VIDEO BASED ON A NOVEL FEATURE IN COMPRESSED DOMAIN OF HEVC;Huang LI et al.;《2014 IEEE International Conference on Multimedia and Expo》;20140908;第3-8页 *
Moving Object Detection in the HEVC Compressed Domain for Ultra-High-Resolution Interactive Video;Johan De Praeter et al.;《2017 IEEE International Conference on Consumer Electronics》;20170330;第2-3页 *
Real-Time Moving Object Segmentation and Classification From HEVC Compressed Surveillance Video;Liang Zhao et al.;《IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY》;20161228;第1346-1357页 *
基于HEVC压缩域的车辆行为事件检测研究;尹烁;《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》;20180215;第17-41页 *

Also Published As

Publication number Publication date
CN108717703A (en) 2018-10-30

Similar Documents

Publication Publication Date Title
Zhao et al. Real-time moving object segmentation and classification from HEVC compressed surveillance video
US9253503B2 (en) Computationally efficient motion estimation with learning capabilities for video compression in transportation and regularized environments
CN102917225B (en) HEVC intraframe coding unit fast selecting method
WO2010057170A1 (en) Analytics-modulated coding of surveillance video
Gao et al. Recent standard development activities on video coding for machines
CN106231315B (en) A kind of video coding system and the method for video coding using the system
CN108717703B (en) HEVC-based moving target detection and tracking method
CN107404653B (en) A kind of Parking rapid detection method of HEVC code stream
KR101149522B1 (en) Apparatus and method for detecting scene change
US9736493B2 (en) System and method for achieving computationally efficient motion estimation in video compression based on motion direction and magnitude prediction
CN106331730B (en) H.264 video same quantization factor double compression detection method
KR102261669B1 (en) Artificial Neural Network Based Object Region Detection Method, Device and Computer Program Thereof
Du et al. Accmpeg: Optimizing video encoding for accurate video analytics
Li et al. A rapid abnormal event detection method for surveillance video based on a novel feature in compressed domain of HEVC
Du et al. Accmpeg: Optimizing video encoding for video analytics
CN103051891A (en) Method and device for determining a saliency value of a block of a video frame block-wise predictive encoded in a data stream
Fischer et al. Boosting neural image compression for machines using latent space masking
US20080253670A1 (en) Image Signal Re-Encoding Apparatus And Image Signal Re-Encoding Method
Laumer et al. Compressed domain moving object detection by spatio-temporal analysis of H. 264/AVC syntax elements
Chen et al. Fast object detection in hevc intra compressed domain
Laumer et al. Moving object detection in the H. 264/AVC compressed domain
CN113422959A (en) Video encoding and decoding method and device, electronic equipment and storage medium
Solana-Cipres et al. Real-time moving object segmentation in H. 264 compressed domain based on approximate reasoning
CN106878754B (en) A kind of 3D video depth image method for choosing frame inner forecast mode
CN110933415B (en) Object region detection method, device and non-transitory computer readable medium for object region detection

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