CN105657215A - Abnormal event detection method based on characteristics of compressed domain - Google Patents
Abnormal event detection method based on characteristics of compressed domain Download PDFInfo
- Publication number
- CN105657215A CN105657215A CN201511024728.9A CN201511024728A CN105657215A CN 105657215 A CN105657215 A CN 105657215A CN 201511024728 A CN201511024728 A CN 201511024728A CN 105657215 A CN105657215 A CN 105657215A
- Authority
- CN
- China
- Prior art keywords
- mic
- frame
- abnormal event
- video
- motion
- 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.)
- Pending
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N5/00—Details of television systems
- H04N5/14—Picture signal circuitry for video frequency region
- H04N5/144—Movement detection
- H04N5/145—Movement estimation
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
Abstract
The invention provides an abnormal event detection method based on characteristics of a compressed domain. A characteristic value for quantitatively describing motion intensity, namely motion intensity count (MIC), is provided; the characteristic value is based on video motion vectors in the compressed domain and a decomposition mode thereof, thereby causing tiny computational overhead. Then, an MIC predicting that later video frames based on a certain frame contain all LCUs is provided. Finally, by analyzing difference between an actual MIC of each LCU and a predicted MIC, a corresponding value of motion intensity change tendency can be detected, and the value is utilized as a detection value of a motion abnormal event. Larger the difference is, more inconsistent the motion intensity change tendency within a certain time period is, so that an abnormal event possibly occurs; conversely, the change of an object motion situation within the time period is little, and no abnormal event occurs. According to the abnormal event detection method, complete decompression of a compression-format video can be prevented, extra computational overhead is also tiny, and abnormal events in a monitoring video can be quickly and accurately detected.
Description
Technical field
The present invention relates to field of video content analysis, more specifically, it relates to a kind of accident detection method based on compressed feature.
Background technology
Accident detection is in one of the most important direction of field of video content analysis. Such as, compression technology, the development of network technology and memory technology makes traffic video monitoring system obtain to apply widely. And these systems are recorded and store the video data of magnanimity. Therefore, real-time accident detection and early warning that how these video sequences carry out automatization have huge actual demand.
Current accident detection great majority concentrate on the method for pixel domain. But, these methods can bring bigger computing cost usually because they usually first by the video compression of packed format to pixel domain, then carried out the light stream of estimated image by some methods such as Block-matching or based on gradient. In fact, video data are normally deposited in the compressed format, therefore inevitable complete for video decompressor can must being carried out of the method for detecting abnormality of pixel domain, its extra expense cannot be avoided. For the video monitoring system (such as traffic supervisory system) having real-time event detection demand, the method for pixel domain often can not be suitable for.
On the other hand, the video data of compression often inherently contain a lot of information that event detection is useful: such as resolution model and motion vector, the edge providing object that respectively can be rough and Optic flow information. For the video coding standard HighEfficiencyVideoCoding (HEVC) of a new generation, the information of compression domain is H.264/AVC abundanter than front generation standard, therefore also more is conducive to carrying out the event detection of compression domain. The coding structure of HEVC have employed a kind of coding unit based on quaternary tree and more predictive mode so that it can provide more accurate about the information of object edge and light stream. In addition, the information of these compression domain can be obtained from compressed video bit stream by less computing cost
Summary of the invention
The technical problem to be solved in the present invention is the abnormal event how detected rapidly and accurately in compression video, it is proposed to a kind of accident detection method based on compressed feature.The method is to solve video monitoring system to contradiction universal rapidly along with HD video and day by day serious between the speed of event detection and accuracy.
For solving the problems of the technologies described above, the technical scheme of the present invention is as follows:
Based on an accident detection method for compressed feature, the method is an accident detection method towards monitor video. It is by utilizing motion vector, coding unit size and predicting unit pattern in HEVC code stream, with minimum additional computational overhead, it is to construct a feature based on compression domain. This feature can detect abnormal event rapidly and accurately, and can save the computing cost carrying out the video decode of packed format to pixel domain processing.
Based on an accident detection method for compressed feature, its main contents comprise following several respects:
(1) according to a large amount of experimental analyses, the mode division that the LCU comprising violent mobile has bigger motion vector sum meticulousr usually. Analyze based on this, the present invention proposes MIC to utilize the exercise intensity of these information being representative LCU.
(2) propose and how to utilize the pattern of present frame to decompose and MV value predicts the MIC predictor at its next frame or a few frame below.
(3) based on MIC, it is proposed that detected the algorithm of abnormal event by the difference of analyses and prediction intensity He actual resolution model.
For the ease of setting forth, briefly provide the definition of some concepts. In HEVC, first each frame be divided into coding unit (LCU) all impartially. What each LCU was further is divided into less coding unit (CU). Each CU can be had some alternative predictive modes (PU). To each interframe PU pattern, all can be corresponding have one to point to it at the motion vector (MV) with reference to frame Optimum Matching.
According to a large amount of experimental analyses, the mode division that the LCU comprising violent mobile has bigger motion vector sum meticulousr usually. Analyze based on this, the present invention proposes MIC to utilize the exercise intensity of these information being representative LCU.
Based on an accident detection method for compressed feature, the general steps of the method:
(1) compressed video file is read.
(2) for current encoded frame, judge whether this frame belongs to crucial frame, it is, skip until next non-key frame.
(3) to present frame, the MIC of s (adjustable parameters is set to 4) frame below is upgraded according to motion vector information.
(4) the actual MIC of present frame is calculated according to resolution model and predictive mode.
(5) for each coding unit of present frame, according to abnormality detection algorithm, weigh the difference between predictor and true value, judge whether there is abnormal event.
(6) judging whether last frame, be, detection terminates, otherwise reads next frame and as present frame and return to step (2).
Compared with prior art, the useful effect of technical solution of the present invention is: the present invention is directed to HEVC standard, it is proposed to one utilizes the quick event detecting method of compressed domain, and the method detects abnormal event by detecting the accident change of exercise intensity. In order to measure exercise intensity, it is proposed that a feature based on compression domain, exercise intensity counting (MIC).
Accompanying drawing explanation
Fig. 1 is the basic flow sheet of the present invention.
Fig. 2 is that MIC calculates exemplary plot.
Fig. 3 is MIC exercise intensity transfer method figure.
Fig. 4 is the accident detection algorithm exemplary plot based on prediction MIC.
Embodiment
Accompanying drawing, only for exemplary illustration, can not be interpreted as the restriction to this patent;
In order to the present embodiment is better described, some parts of accompanying drawing have omission, zoom in or out, and do not represent the size of actual product;
To those skilled in the art, some known features and illustrate and may omit and be appreciated that in accompanying drawing.
Below in conjunction with drawings and Examples, the technical scheme of the present invention is described further.
Embodiment 1
(1) general function framework
Algorithm proposed by the invention mainly comprises two parts: based on the eigenwert of exercise intensity and the predictive model thereof of compression domain, and is detected the algorithm of abnormal event by the difference of analyses and prediction exercise intensity and actual motion intensity.
By utilizing the motion vector definition of the predictive mode size in compression domain and its correspondence eigenwert, can approximate representation motion severe degree. Then, describe the exercise intensity of next frame of information prediction how utilizing present frame, it is proposed that the model of prediction MIC. Finally, give and how to utilize MIC to be truly worth with predictor to carry out the specific algorithm of accident detection.
(2) implementing procedure
Fig. 1 is the basic procedure of algorithm
Step 1, the video reading packed format.
Step 2, each LCU to each two field picture, calculate MIC value according to compressed domain, and calculate the prediction MIC value of next frame.
Step 3, MIC to each LCU are truly worth the difference with MIC value and judge, are greater than threshold value and are abnormal LCU occurs.
Step 4, present frame terminate, and utilize resolution model and the motion vector of present frame, and the MCC table carrying out next frame upgrades.
Step 5, repeating step 2 are until the analysis of compression video all terminates.
(3) detailed construction design
Fig. 2 is that MIC eigenwert calculates exemplary plot. Fig. 3 is the metastasis model that M predicts MIC. Fig. 4 is the accident detection method exemplary plot based on MIC.
1, exercise intensity counting and prediction thereof
(1) in order to utilize the pattern in compression domain to decompose and motion vector information, invention defines a kind of new eigenwert, exercise intensity counting (MIC). The MIC of a LCU refers to the area of all PU that its resolution model comprises and the product sum of MV length. Fig. 1 gives the example how calculating MIC.
(1) MIC predicts metastasis model
Utilize all motion vectors of present frame and relevant PU thereof, it is proposed that according to the direction of MV, the trend of predicted motion, and estimate that its corresponding PU is in the possible position of next frame. For the LCU of certain next frame, PU and MV having certain number is within the scope of it, and the MIC now calculated according to this predicted position is exactly MIC predictor corresponding to this LCU. Fig. 2 describes the prediction transfer how carrying out exercise intensity.
2, based on MIC accident detection
For each LCU of video, there is the MIC predictor that a MIC is truly worth and some are obtained by the prediction of its adjacent frame. Both the actual motion intensity representing this LCU respectively and the motion predict value meeting certain rule estimated according to the adjacent frame in its short period
(1) the true value of MIC and predictor difference
In shorter video cycle, the motion of object should be at the uniform velocity, straight line. If there occurs that the movement tendency of the irregulars such as violent collision, deformation changes, then the regularity of the pattern of these motions will decline to some extent. Utilizing the true value of MIC as standard, the MIC that the frame of video analyzed in time cycle on one point predicts, if there occurs abnormal event, then may there is certain fluctuation in this difference value. By analyzing, these fluctuations detect abnormal event to this algorithm.
(2) event detection
Fig. 2 is a schematic diagram of event detection.For frame currently to be detected, algorithm generates the MIC predictor about present frame by its front s frame. Then, calculate the true MIC of present frame, and combine its difference value d of s predictor calculation, pass through calculation formula:
Wherein MIC_Pred and MIC_Actual is the predictor of current LCU and true value respectively. When d is greater than the threshold value of certain setting, then think that this position there occurs abnormal event
Utilize the MV of the LCU resolution model in compression domain and correspondence, define the eigenwert that can quantize to describe exercise intensity, exercise intensity counting (MIC). In monitor video accident detection, as how lower expense accurately detects the difficult point that abnormal event is problem. The present invention proposes the detection method based on HEVC compression domain, for similar research opens a kind of new thinking. The core of this invention is, it is extracted useful information in the compressed domain, construct exercise intensity counting (MIC) this eigenwert, and observed the true value of MIC and the difference value changing conditions of observed value when standard state and when abnormal event occurs respectively, thus devise abnormal event monitoring method fast and accurately.
The parts that same or similar label is corresponding same or similar;
Accompanying drawing describes position relation for only for exemplary illustration, the restriction to this patent can not be interpreted as;
Obviously, the above embodiment of the present invention is only for example of the present invention is clearly described, and is not the restriction to embodiments of the present invention. For those of ordinary skill in the field, can also make other changes in different forms on the basis of the above description. Here without the need to also cannot all enforcement modes be given exhaustive. All any amendment, equivalent replacement and improvement etc. done within the spirit and principles in the present invention, all should be included within the protection domain of the claims in the present invention.
Claims (1)
1. the accident detection method based on compressed feature, it is characterised in that, the general steps of the method:
(1) compressed video file is read;
(2) for current encoded frame, judge whether this frame belongs to crucial frame, it is, skip until next non-key frame;
(3) to present frame, the MIC of s frame below is upgraded according to motion vector information;
(4) the actual MIC of present frame is calculated according to resolution model and predictive mode;
(5) for each coding unit of present frame, according to abnormality detection algorithm, weigh the difference between predictor and true value, judge whether there is abnormal event;
(6) judging whether last frame, be, detection terminates, otherwise reads next frame and as present frame and return to step (2).
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201511024728.9A CN105657215A (en) | 2015-12-29 | 2015-12-29 | Abnormal event detection method based on characteristics of compressed domain |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201511024728.9A CN105657215A (en) | 2015-12-29 | 2015-12-29 | Abnormal event detection method based on characteristics of compressed domain |
Publications (1)
Publication Number | Publication Date |
---|---|
CN105657215A true CN105657215A (en) | 2016-06-08 |
Family
ID=56490989
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201511024728.9A Pending CN105657215A (en) | 2015-12-29 | 2015-12-29 | Abnormal event detection method based on characteristics of compressed domain |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105657215A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107592547A (en) * | 2017-08-31 | 2018-01-16 | 浙江工业大学 | A kind of motion perception figure extracting method based on HEVC compression domains |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2009044475A (en) * | 2007-08-09 | 2009-02-26 | Victor Co Of Japan Ltd | Monitor camera system |
CN102254329A (en) * | 2011-08-18 | 2011-11-23 | 上海方奥通信技术有限公司 | Abnormal behavior detection method based on motion vector classification analysis |
CN104159089A (en) * | 2014-09-04 | 2014-11-19 | 四川省绵阳西南自动化研究所 | Abnormal event alarm high-resolution video intelligent processor |
CN104820824A (en) * | 2015-04-23 | 2015-08-05 | 南京邮电大学 | Local abnormal behavior detection method based on optical flow and space-time gradient |
-
2015
- 2015-12-29 CN CN201511024728.9A patent/CN105657215A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2009044475A (en) * | 2007-08-09 | 2009-02-26 | Victor Co Of Japan Ltd | Monitor camera system |
CN102254329A (en) * | 2011-08-18 | 2011-11-23 | 上海方奥通信技术有限公司 | Abnormal behavior detection method based on motion vector classification analysis |
CN104159089A (en) * | 2014-09-04 | 2014-11-19 | 四川省绵阳西南自动化研究所 | Abnormal event alarm high-resolution video intelligent processor |
CN104820824A (en) * | 2015-04-23 | 2015-08-05 | 南京邮电大学 | Local abnormal behavior detection method based on optical flow and space-time gradient |
Non-Patent Citations (1)
Title |
---|
HUANG LI 等: "a rapid abnormal event detection method for surveillance video based on a novel feature in compressed domain of HEVC", 《2014 IEEE INTERNATIONAL CONFERENCE ON MULTIMEDIA AND EXPO (ICME)》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107592547A (en) * | 2017-08-31 | 2018-01-16 | 浙江工业大学 | A kind of motion perception figure extracting method based on HEVC compression domains |
CN107592547B (en) * | 2017-08-31 | 2019-05-31 | 浙江工业大学 | A kind of motion perception figure extracting method based on HEVC compression domain |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108256506B (en) | Method and device for detecting object in video and computer storage medium | |
CN102263955B (en) | Method for detecting video occlusion based on motion vectors | |
CN101090498B (en) | Device and method for motion detection of image | |
US9454902B2 (en) | Performing-time-series based predictions with projection thresholds using secondary time-series-based information stream | |
CN101699865B (en) | Method for fast searching mass movement self-adapting sub pixel | |
CN1078591A (en) | Motion forecast method and device thereof | |
JP2003284076A (en) | Device and method for detecting movement in digital image storage utilizing mpeg image compression technique | |
CN108596109B (en) | Target detection method and device based on neural network and motion vector | |
CN110839040A (en) | Flow monitoring method, model training method, device and storage medium | |
KR102310851B1 (en) | Apparatus and method for monitoring process based on image analysis | |
CN100421472C (en) | Adaptive early exit techniques for image correlation minimum distortion calculation | |
CN109002996A (en) | Methods of risk assessment and system based on water rate | |
CN105657215A (en) | Abnormal event detection method based on characteristics of compressed domain | |
CN104104966A (en) | Video block dividing method and device | |
CN101764923B (en) | Method and device for eliminating noise in motion detection | |
CN116863404A (en) | Video abnormal event detection method for sensing local response | |
CN101308576B (en) | Motion direction detecting method and apparatus in road | |
CN101453559B (en) | Noise detection method and apparatus for video signal | |
CN101854547B (en) | Motion frame in video collection and transmission system, method and system for detecting prospects | |
Srilekha et al. | A novel approach for detection and tracking of vehicles using Kalman filter | |
Raiyn et al. | Real-time short-term forecasting based on information management | |
CN113792697A (en) | Target detection method and device, electronic equipment and readable storage medium | |
CN114743136A (en) | Abnormal behavior detection method, device and storage medium | |
CN102065301A (en) | Method and device for coding detecting information in intelligent monitoring system | |
CN106412580B (en) | The method and device of intracoded frame in a kind of setting video sequence |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160608 |