CN109919979A - A kind of method of video real-time modeling method - Google Patents
A kind of method of video real-time modeling method Download PDFInfo
- Publication number
- CN109919979A CN109919979A CN201910174796.5A CN201910174796A CN109919979A CN 109919979 A CN109919979 A CN 109919979A CN 201910174796 A CN201910174796 A CN 201910174796A CN 109919979 A CN109919979 A CN 109919979A
- Authority
- CN
- China
- Prior art keywords
- frame
- video
- target
- detection
- real
- 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.)
- Withdrawn
Links
Landscapes
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of methods of video real-time modeling method, the technology combined using target detection with target following, greatly reduce the calculation amount of video real-time target detection, due to without carrying out traversal detection to each frame video image, therefore the computational efficiency of video real-time target detection is greatly improved, can achieve the frame per second of real-time video.Position of the target frame that the method for video real-time modeling method provided by the invention detected object detector using neural network in next frame image carries out tracking recurrence, significantly reduce the calculation amount of video real-time target detection, without all using detector to detect target each frame image, it is applied in the detection of video real-time target using the technology that detection is combined with tracking, without carrying out the processing such as complicated noise reduction to input picture, to object detector also without specific demand, the rate of detection can be greatly promoted, applicability of the present invention is wide, it can guarantee enough computational efficiencies in the embedded device of low side.
Description
Technical field
The present invention relates to a kind of methods of video real-time modeling method.
Background technique
Target detection has obtained unprecedented development in recent years, and constantly iteration updates algorithm of target detection, from
The deep learning neural network detection algorithm of initial Conventional visual detection algorithm till now, the accuracy rate of target detection and
Stability is greatly improved, and the accuracy of especially neural network detection algorithm has reached commercialized rank.
Target detection technique is the important foundation technology of video real-time target detection field, and video real-time target detects always
Since be academia and industry important subject, real-time target detection can apply in many video scenes, such as
Object locating system, object identification system, Vehicle License Plate Recognition System etc., target detection technique often occupies in these video scenes
Very important status, thereby, it is ensured that the computational efficiency of video real-time detection is of great significance.
The way of video real-time target detection is that traversal detection, this way are carried out to each frame video image mostly at present
Many times are often consumed, each frame video image will carry out traversal and detect whether to reduce real-time video comprising target
Frame per second, lead to not to increase some additional operations, such as analysis, the objective attribute target attribute detection operation of license plate number.
Summary of the invention
The primary purpose of the present invention is that a kind of method of video real-time modeling method is provided, using target detection and target
The technology combined is tracked, the calculation amount of video real-time target detection is greatly reduced, due to without to each frame video image
Traversal detection is carried out, therefore greatly improves the computational efficiency of video real-time target detection, can achieve the frame of real-time video
Rate.
A kind of method of video real-time modeling method, comprising the following steps:
1) real-time video, is acquired as input by hardware device camera, or directly inputs the view comprising multiframe
Frequency file;
2) video, is decomposed, video is decomposed as unit of single frames;
It 3), is the digital image matrix format of object detector support by different digital image matrix format conversions;
4), 1 frame of digital image array of input is into object detector, detector by the testing result that returns after calculating with
The mode of array is saved, and the length of array is the destination number size detected;
5), target basis frame of the target detection frame obtained according to present incoming frame as next frame image, using nerve
Network carries out recurrence calculating in the position of next frame image to present frame target frame, obtains the target detection frame letter of next frame image
Breath, if next frame detection block information is not empty, the circulation execution current procedures in next frame image;If next frame mesh
Mark frame information be sky, then jump to step 4 to next frame image re-call object detector carry out target detection until
Processing terminate for video frame.
What the method for video real-time modeling method provided by the invention detected object detector using neural network
Position of the target frame in next frame image carries out tracking recurrence, significantly reduces the calculation amount of video real-time target detection,
Without all using detector to detect target each frame image, it is real-time that video is applied to using the technology that detection is combined with tracking
In target detection, without carrying out the processing such as complicated noise reduction to input picture, to object detector also without specific demand, Ke Yi great
The big rate for promoting detection, applicability of the present invention is wide, can guarantee enough computational efficiencies in the embedded device of low side.
Detailed description of the invention
The attached drawing constituted part of this application is used to provide further understanding of the present invention, schematic reality of the invention
It applies example and its explanation is used to explain the present invention, do not constitute improper limitations of the present invention.In the accompanying drawings:
Fig. 1 is flow chart of the embodiment of the present invention.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
Embodiment
The present invention is mainly utilized the previous frame object detection results that detector detects in successive frame video and is used as basic frame,
The object detection results of present incoming frame are calculated using neural net regression, then the testing result of present frame is passed to next frame
Recurrence calculating is carried out, target following is carried out.
In the present invention, writing or training for object detector it is not related to, our method is to be promoted in video
Target detection efficiency, for detector without special demand or improvement require, it is only necessary to the digital picture square of input
The testing result that battle array is calculated is the object detector for meeting this method.
As shown in Figure 1, a kind of method of video real-time modeling method is detailed to realize that steps are as follows:
1, real-time video is acquired as input by hardware camera device, or directly inputs the video comprising multiframe
File.Herein we require input video must be effective and each frame video image matrix data type and
Size dimension should be to maintain consistent (in the mutually same input video).
2, video is decomposed at first, in order to be decomposed as unit of single frames to video.A such as view
It include 100 frame of digital images in frequency, then will be divided into 100 inputs, input each time is the matrix function of 1 frame digital image
According to.The data of input can be the digital image matrix of various formats.
3, since the format of the digital image matrix inputted in step 2 may be not quite similar, we need before treatment by
Different digital image matrix format conversions is the digital image matrix format that object detector is supported.The step can also be
Unified conversion is carried out when input video.
4,1 frame of digital image array of input is into object detector, detector by the testing result that returns after calculating with
What the mode of array was saved, the length of array is the destination number size detected.
5, target basis frame of the target detection frame obtained according to present incoming frame as next frame image, using nerve net
Network carries out recurrence calculating in the position of next frame image to present frame target frame, obtains the target detection frame letter of next frame image
Breath.If next frame detection block information is not empty, the circulation execution current procedures in next frame image;If next frame mesh
Mark frame information be sky, then jump to step 4 to next frame image re-call object detector carry out target detection until
Processing terminate for video frame.
As described above, embodiments of the present invention are described in detail, as long as but essence is of the invention without being detached from
Inventive point and effect can have many deformations, this will be readily apparent to persons skilled in the art.Therefore, in this way
Variation be also integrally incorporated within protection scope of the present invention.
Claims (2)
1. a kind of method of video real-time modeling method, it is characterised in that the following steps are included:
1) real-time video, is acquired as input by hardware device camera, or directly inputs the video text comprising multiframe
Part;
2) video, is decomposed, video is decomposed as unit of single frames;
It 3), is the digital image matrix format of object detector support by different digital image matrix format conversions;
4), for 1 frame of digital image array of input into object detector, detector passes through the testing result returned after calculating with array
Mode saved, the length of array is the destination number size detected;
5), target basis frame of the target detection frame obtained according to present incoming frame as next frame image, using neural network
Recurrence calculating is carried out in the position of next frame image to present frame target frame, obtains the target detection frame information of next frame image,
If next frame detection block information is not empty, the circulation execution current procedures in next frame image;If next frame target
Frame information is sky, then jumps to step 4 and re-call object detector progress target detection until view to next frame image
Processing terminate for frequency frame.
2. a kind of method of video real-time modeling method according to claim 1, it is characterised in that:
The step 3) carries out unified conversion when step 1) input video.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910174796.5A CN109919979A (en) | 2019-03-08 | 2019-03-08 | A kind of method of video real-time modeling method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910174796.5A CN109919979A (en) | 2019-03-08 | 2019-03-08 | A kind of method of video real-time modeling method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109919979A true CN109919979A (en) | 2019-06-21 |
Family
ID=66963831
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910174796.5A Withdrawn CN109919979A (en) | 2019-03-08 | 2019-03-08 | A kind of method of video real-time modeling method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109919979A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113160474A (en) * | 2021-03-22 | 2021-07-23 | 浙江大华技术股份有限公司 | Authentication method, authentication terminal, authentication system and storage medium |
-
2019
- 2019-03-08 CN CN201910174796.5A patent/CN109919979A/en not_active Withdrawn
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113160474A (en) * | 2021-03-22 | 2021-07-23 | 浙江大华技术股份有限公司 | Authentication method, authentication terminal, authentication system and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11200424B2 (en) | Space-time memory network for locating target object in video content | |
US10453204B2 (en) | Image alignment for burst mode images | |
CN109214238B (en) | Multi-target tracking method, device, equipment and storage medium | |
CN112861575A (en) | Pedestrian structuring method, device, equipment and storage medium | |
US20210342593A1 (en) | Method and apparatus for detecting target in video, computing device, and storage medium | |
CN108491763B (en) | Unsupervised training method and device for three-dimensional scene recognition network and storage medium | |
WO2011161579A1 (en) | Method, apparatus and computer program product for providing object tracking using template switching and feature adaptation | |
US20210312195A1 (en) | Lane marking detecting method, apparatus, electronic device, storage medium, and vehicle | |
CN112418216A (en) | Method for detecting characters in complex natural scene image | |
CN102457724B (en) | Image motion detecting system and method | |
CN110827320B (en) | Target tracking method and device based on time sequence prediction | |
CN112818955A (en) | Image segmentation method and device, computer equipment and storage medium | |
CN111382647B (en) | Picture processing method, device, equipment and storage medium | |
KR101993085B1 (en) | Semantic image segmentation method based on deep learing | |
KR20220153667A (en) | Feature extraction methods, devices, electronic devices, storage media and computer programs | |
Fortun et al. | Fast piecewise-affine motion estimation without segmentation | |
CN114022531A (en) | Image processing method, electronic device, and storage medium | |
CN108229281B (en) | Neural network generation method, face detection device and electronic equipment | |
CN113256683B (en) | Target tracking method and related equipment | |
CN109919979A (en) | A kind of method of video real-time modeling method | |
CN110298229B (en) | Video image processing method and device | |
CN110555865B (en) | Dynamic visual sensor sample set modeling method based on frame image | |
CN113869163B (en) | Target tracking method and device, electronic equipment and storage medium | |
CN113657218B (en) | Video object detection method and device capable of reducing redundant data | |
CN113486788A (en) | Video similarity determination method and device, electronic equipment and storage medium |
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 | ||
WW01 | Invention patent application withdrawn after publication | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190621 |