CN106934319A - People's car objective classification method in monitor video based on convolutional neural networks - Google Patents
People's car objective classification method in monitor video based on convolutional neural networks Download PDFInfo
- Publication number
- CN106934319A CN106934319A CN201511008852.6A CN201511008852A CN106934319A CN 106934319 A CN106934319 A CN 106934319A CN 201511008852 A CN201511008852 A CN 201511008852A CN 106934319 A CN106934319 A CN 106934319A
- Authority
- CN
- China
- Prior art keywords
- convolutional neural
- neural networks
- car
- people
- monitor video
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
- G06V20/54—Surveillance or monitoring of activities, e.g. for recognising suspicious objects of traffic, e.g. cars on the road, trains or boats
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/08—Detecting or categorising vehicles
Abstract
The present invention discloses people's car objective classification method in a kind of monitor video based on convolutional neural networks, including：The sample set of multi-angle is obtained, and sample set is divided into training sample set, checking sample set and test sample collection；Set up convolutional neural networks；The samples pictures that training sample is concentrated first are subtracted into the corresponding average of each pixel, then the convolutional neural networks is input into as training data, the study for having supervision is carried out, the parameter of each layer of the convolutional neural networks after being trained；Using the parameter of each layer of the convolutional neural networks after training, mutually isostructural convolutional neural networks are initialized, obtain the image recognition network with people's car target classification function in monitor video.The present invention is targetedly trained to convolutional neural networks using the samples pictures of multi-angle, adjusted, and can reach people's car classification accuracy higher.
Description
Technical field
The present invention relates to people's car objective classification method in a kind of monitor video based on convolutional neural networks, belong to
Mode identification technology.
Background technology
With continuing to develop for society, public safety takes precautions against the important foundation for having become urban modernization, depending on
Frequency monitoring system plays an important role as the important component of safety and protection system, in city road
Road field of traffic, widely uses the traffic behavior of video monitoring system real time record people, car, using convolution god
Through targets such as the people in Network Recognition monitor video, cars, the disposal of all kinds of traffic casees can greatly be improved
Efficiency.People's car sorting technique in existing monitor video, nicety of grading is not high, and real-time is not strong enough.
The content of the invention
In view of the foregoing, it is an object of the invention to provide a kind of monitor video based on convolutional neural networks
Convolutional neural networks are carried out targetedly by middle people's car objective classification method using the samples pictures of multi-angle
Training, adjustment, can reach people's car classification accuracy higher.
To achieve the above object, the present invention uses following technical scheme：
People's car objective classification method in a kind of monitor video based on convolutional neural networks, including：
S1：The sample set of multi-angle is obtained, and sample set is divided into training sample set, checking sample set
And test sample collection；
S2：Set up convolutional neural networks；
S3：The samples pictures that training sample is concentrated first are subtracted into the corresponding average of each pixel, then conduct
Training data is input into the convolutional neural networks, carries out the study for having supervision, the convolutional Neural net after being trained
The parameter of each layer of network；
S4：Using the parameter of each layer of the convolutional neural networks after training, described in initialization and step S2
Convolutional neural networks structure identical convolutional neural networks, obtain with people's car target classification work(in monitor video
The image recognition network of energy.
In the step S1, the acquisition methods of the sample set of the multi-angle are：
People, car, the picture of inhuman non-car in substantial amounts of monitor video are gathered, all pictures are zoomed to together
Deng the picture of pixel size, the label for distinguishing people, car, inhuman non-car picture is added in all pictures,
Mirror image, rotation processing are carried out to all pictures.
Convolutional neural networks in the step S2 include two convolutional layers, two down-sampling layers, and one complete
Articulamentum, and softmax graders, the size of the first convolutional layer wave filter is 5 × 5 pixels, and characteristic pattern is
6, the size of the first down-sampling layer wave filter is 2 × 2 pixels, and characteristic pattern is 6, the filter of the second convolutional layer
The size of ripple device is 5 × 5 pixels, and characteristic pattern is 16, and the size of the second down-sampling layer wave filter is 2 × 2
Pixel, characteristic pattern is 16, and the characteristic pattern of full articulamentum is 120, and softmax graders export three kinds
The target of type：People, car, other.
Horizontal mirror image processing is carried out to all pictures, 10 degree are then rotated in the horizontal direction.
It is an advantage of the invention that：
1st, the picture in substantial amounts of monitor video is gathered, and picture is pre-processed, increased not same
Convolutional neural networks are targetedly trained, adjusted by the otherness between this picture on this basis,
Classification accuracy higher can be reached, it is ensured that the real-time of assorting process；
2nd, using the method for Machine self-learning, human intervention is reduced, convolutional neural networks is learnt
To comprehensive people's car feature, the generalization ability of network is strong.
Brief description of the drawings
Fig. 1 is the structural representation of convolutional neural networks of the invention.
Fig. 2 is the training process figure of convolutional neural networks of the invention.
Fig. 3 is the procedure chart classified using convolutional neural networks of the invention.
Specific embodiment
As shown in Figures 1 to 3, people's car target in the monitor video based on convolutional neural networks disclosed by the invention
Sorting technique, comprises the following steps：
S1：Sample set is obtained, and sample set is divided into training sample set, checking sample set and test sample
Collection；
People, car, the picture of inhuman non-car in substantial amounts of monitor video are gathered, all pictures are zoomed to 32
All pictures are carried out mean value calculation, and add in all pictures by × 32 pixel sizes in each pixel
Tag, for example, adding 0 in the picture of someone, 1 is added in the picture for have car, in inhuman non-car
Picture in add 2；
Afterwards, as samples pictures after being pre-processed to all pictures, pretreatment include picture mirror image,
Rotation processing, mirror-image fashion is horizon glass picture, and the anglec of rotation is 10 degree of rotation in the horizontal direction；Pretreatment
The samples pictures for obtaining afterwards have a multi-angle feature comprehensively, abundant, and increase different samples pictures (people,
Car, other) between otherness.
All samples pictures are divided into training sample set (accounting for the 85% of total sample), checking sample set and (account for total
Sample 10%) and test sample collection (accounting for the 5% of total sample).
S2：Set up convolutional neural networks；
As shown in figure 1, the convolutional neural networks that the present invention sets up include two convolutional layers, two down-sampling layers,
One full articulamentum, and softmax graders.The size of the first convolutional layer wave filter is 5 × 5 pixels, special
It is 6 to levy figure, and the size of the first down-sampling layer wave filter is 2 × 2 pixels, and characteristic pattern is 6, second
The size of convolutional layer wave filter be 5 × 5 pixels, characteristic pattern be 16, the second down-sampling layer wave filter it is big
Small is 2 × 2 pixels, and characteristic pattern is 16, and the characteristic pattern of full articulamentum is 120, softmax classification
Device exports the target of three types：People, car, other.
S3：The samples pictures that training sample is concentrated first are subtracted into the corresponding average of each pixel, then conduct
Training data is input into convolutional neural networks, carries out the study for having supervision of tape label, the convolution after being trained
The parameter of each layer of neutral net；
In training process, using the parameter of each layer in stochastic gradient descent method adjustment convolutional neural networks, observation
Accuracy rate change of the convolutional neural networks on checking sample set, and the learning rate of convolutional neural networks is adjusted,
Ensure that convolutional neural networks are restrained on training sample set and classification standard higher is reached on checking sample set
True rate, after convolutional neural networks are restrained (threshold value of the rate of accuracy reached to setting), preserves convolutional neural networks
In each layer parameter.
S4：Using the parameter of each layer of the convolutional neural networks after training, the convolution god of same structure is initialized
Through network, the image recognition network with people's car target classification function in monitor video is obtained.
It is follow-up test sample collection to be tested using the image recognition network, is classified.
In a specific embodiment, the training sample set size for using is 24000 identity card pictures, checking
The size of sample set is 2823 identity card pictures, and the size of test sample collection is in 1411 monitor videos
People, car, the picture of inhuman non-car, before test, the samples pictures of test sample collection are first subtracted into each picture
The corresponding average of element, then input picture identification network, the final image recognition network is in test sample collection
Classification accuracy reached 86%, because samples pictures have multi-angle, thus classification accuracy is higher.
The above is presently preferred embodiments of the present invention and its know-why used, for the skill of this area
It is without departing from the spirit and scope of the present invention, any based on the technology of the present invention side for art personnel
Equivalent transformation on the basis of case, it is simple replace etc. it is obvious change, belong to the scope of the present invention it
It is interior.
Claims (4)
1. people's car objective classification method in the monitor video of convolutional neural networks is based on, it is characterised in that bag
Include：
S1：The sample set of multi-angle is obtained, and sample set is divided into training sample set, checking sample set
And test sample collection；
S2：Set up convolutional neural networks；
S3：The samples pictures that training sample is concentrated first are subtracted into the corresponding average of each pixel, then conduct
Training data is input into the convolutional neural networks, carries out the study for having supervision, the convolutional Neural net after being trained
The parameter of each layer of network；
S4：Using the parameter of each layer of the convolutional neural networks after training, described in initialization and step S2
Convolutional neural networks structure identical convolutional neural networks, obtain with people's car target classification work(in monitor video
The image recognition network of energy.
2. people's car target classification in the monitor video based on convolutional neural networks according to claim 1
Method, it is characterised in that in the step S1, the acquisition methods of the sample set of the multi-angle are：
People, car, the picture of inhuman non-car in substantial amounts of monitor video are gathered, all pictures are zoomed to together
Deng the picture of pixel size, the label for distinguishing people, car, inhuman non-car picture is added in all pictures,
Mirror image, rotation processing are carried out to all pictures.
3. people's car target classification in the monitor video based on convolutional neural networks according to claim 2
Method, it is characterised in that the convolutional neural networks in the step S2 are including under two convolutional layers, two
Sample level, a full articulamentum, and softmax graders, the size of the first convolutional layer wave filter is 5 × 5
Pixel, characteristic pattern is 6, and the size of the first down-sampling layer wave filter is 2 × 2 pixels, and characteristic pattern is 6
Individual, the size of the second convolutional layer wave filter is 5 × 5 pixels, and characteristic pattern is 16, the second down-sampling metafiltration
The size of ripple device is 2 × 2 pixels, and characteristic pattern is 16, and the characteristic pattern of full articulamentum is 120, softmax
Grader exports the target of three types：People, car, other.
4. people's car target classification in the monitor video based on convolutional neural networks according to claim 2
Method, it is characterised in that horizontal mirror image processing is carried out to all pictures, 10 are then rotated in the horizontal direction
Degree.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201511008852.6A CN106934319A (en) | 2015-12-29 | 2015-12-29 | People's car objective classification method in monitor video based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201511008852.6A CN106934319A (en) | 2015-12-29 | 2015-12-29 | People's car objective classification method in monitor video based on convolutional neural networks |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106934319A true CN106934319A (en) | 2017-07-07 |
Family
ID=59457597
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201511008852.6A Pending CN106934319A (en) | 2015-12-29 | 2015-12-29 | People's car objective classification method in monitor video based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106934319A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107679525A (en) * | 2017-11-01 | 2018-02-09 | 腾讯科技（深圳）有限公司 | Image classification method, device and computer-readable recording medium |
| CN108091141A (en) * | 2017-10-12 | 2018-05-29 | 西安天和防务技术股份有限公司 | Vehicle License Plate Recognition System |
| CN108495287A (en) * | 2018-02-13 | 2018-09-04 | 大唐高鸿信息通信研究院（义乌）有限公司 | Target identification and learning method suitable for vehicle-mounted short haul connection net automatic Pilot |
| CN109389136A (en) * | 2017-08-08 | 2019-02-26 | 上海为森车载传感技术有限公司 | Classifier training method |
| CN109753978A (en) * | 2017-11-01 | 2019-05-14 | 腾讯科技（深圳）有限公司 | Image classification method, device and computer readable storage medium |
| CN110378191A (en) * | 2019-04-25 | 2019-10-25 | 东南大学 | Pedestrian and vehicle classification method based on millimeter wave sensor |
| CN110516615A (en) * | 2019-08-29 | 2019-11-29 | 广西师范大学 | Human and vehicle shunting control method based on convolutional neural networks |
| CN111383721A (en) * | 2018-12-27 | 2020-07-07 | 江苏金斯瑞生物科技有限公司 | Construction method of prediction model, and prediction method and device of polypeptide synthesis difficulty |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8135614B2 (en) * | 2001-10-17 | 2012-03-13 | United Toll Systems, Inc. | Multiple RF read zone system |
| CN104463194A (en) * | 2014-11-04 | 2015-03-25 | 深圳市华尊科技有限公司 | Driver-vehicle classification method and device |
| CN104504395A (en) * | 2014-12-16 | 2015-04-08 | 广州中国科学院先进技术研究所 | Method and system for achieving classification of pedestrians and vehicles based on neural network |
| CN104537387A (en) * | 2014-12-16 | 2015-04-22 | 广州中国科学院先进技术研究所 | Method and system for classifying automobile types based on neural network |
-
2015
- 2015-12-29 CN CN201511008852.6A patent/CN106934319A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8135614B2 (en) * | 2001-10-17 | 2012-03-13 | United Toll Systems, Inc. | Multiple RF read zone system |
| CN104463194A (en) * | 2014-11-04 | 2015-03-25 | 深圳市华尊科技有限公司 | Driver-vehicle classification method and device |
| CN104504395A (en) * | 2014-12-16 | 2015-04-08 | 广州中国科学院先进技术研究所 | Method and system for achieving classification of pedestrians and vehicles based on neural network |
| CN104537387A (en) * | 2014-12-16 | 2015-04-22 | 广州中国科学院先进技术研究所 | Method and system for classifying automobile types based on neural network |
Non-Patent Citations (1)
| Title |
|---|
| 赵伟达 等: "人脸检测中基于自适应ICA的特征提取算法", 《计算机仿真》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109389142A (en) * | 2017-08-08 | 2019-02-26 | 上海为森车载传感技术有限公司 | Classifier training method |
| CN109389136A (en) * | 2017-08-08 | 2019-02-26 | 上海为森车载传感技术有限公司 | Classifier training method |
| CN108091141A (en) * | 2017-10-12 | 2018-05-29 | 西安天和防务技术股份有限公司 | Vehicle License Plate Recognition System |
| CN108091141B (en) * | 2017-10-12 | 2020-10-02 | 西安天和防务技术股份有限公司 | License plate recognition system |
| CN110399929A (en) * | 2017-11-01 | 2019-11-01 | 腾讯科技（深圳）有限公司 | Eye fundus image classification method, device and computer readable storage medium |
| CN107679525B (en) * | 2017-11-01 | 2022-11-29 | 腾讯科技（深圳）有限公司 | Image classification method and device and computer readable storage medium |
| CN110348428A (en) * | 2017-11-01 | 2019-10-18 | 腾讯科技（深圳）有限公司 | Eye fundus image classification method, device and computer readable storage medium |
| CN110348428B (en) * | 2017-11-01 | 2023-03-24 | 腾讯科技（深圳）有限公司 | Fundus image classification method and device and computer-readable storage medium |
| CN107679525A (en) * | 2017-11-01 | 2018-02-09 | 腾讯科技（深圳）有限公司 | Image classification method, device and computer-readable recording medium |
| CN109753978B (en) * | 2017-11-01 | 2023-02-17 | 腾讯科技（深圳）有限公司 | Image classification method, device and computer readable storage medium |
| CN109753978A (en) * | 2017-11-01 | 2019-05-14 | 腾讯科技（深圳）有限公司 | Image classification method, device and computer readable storage medium |
| CN108495287A (en) * | 2018-02-13 | 2018-09-04 | 大唐高鸿信息通信研究院（义乌）有限公司 | Target identification and learning method suitable for vehicle-mounted short haul connection net automatic Pilot |
| CN108495287B (en) * | 2018-02-13 | 2020-12-01 | 大唐高鸿信息通信(义乌)有限公司 | Target identification and learning method suitable for automatic driving of vehicle-mounted short-distance communication network |
| CN111383721A (en) * | 2018-12-27 | 2020-07-07 | 江苏金斯瑞生物科技有限公司 | Construction method of prediction model, and prediction method and device of polypeptide synthesis difficulty |
| CN110378191A (en) * | 2019-04-25 | 2019-10-25 | 东南大学 | Pedestrian and vehicle classification method based on millimeter wave sensor |
| CN110378191B (en) * | 2019-04-25 | 2023-09-22 | 东南大学 | Pedestrian and vehicle classification method based on millimeter wave sensor |
| CN110516615B (en) * | 2019-08-29 | 2022-04-08 | 广西师范大学 | Pedestrian and vehicle distribution control method based on convolutional neural network |
| CN110516615A (en) * | 2019-08-29 | 2019-11-29 | 广西师范大学 | Human and vehicle shunting control method based on convolutional neural networks |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106934319A (en) | People's car objective classification method in monitor video based on convolutional neural networks | |
| CN106845424B (en) | Pavement remnant detection method based on deep convolutional network | |
| Tahmid et al. | Density based smart traffic control system using canny edge detection algorithm for congregating traffic information | |
| CN105303193B (en) | A kind of passenger number statistical system based on single-frame images processing | |
| CN105744232B (en) | A kind of method of the transmission line of electricity video external force damage prevention of Behavior-based control analytical technology | |
| CN106934408A (en) | Identity card picture sorting technique based on convolutional neural networks | |
| CN103984961B (en) | A kind of image detecting method for being used to detect underbody foreign matter | |
| CN105761261B (en) | A method of detection camera suffers artificial malicious sabotage | |
| CN105184271A (en) | Automatic vehicle detection method based on deep learning | |
| Bu et al. | Crack detection using a texture analysis-based technique for visual bridge inspection | |
| CN108446630A (en) | Airfield runway intelligent control method, application server and computer storage media | |
| CN104537387A (en) | Method and system for classifying automobile types based on neural network | |
| CN105975929A (en) | Fast pedestrian detection method based on aggregated channel features | |
| CN106778645A (en) | A kind of image processing method and device | |
| CN104766097B (en) | Surface of aluminum plate defect classification method based on BP neural network and SVMs | |
| CN108491797A (en) | A kind of vehicle image precise search method based on big data | |
| CN103605971B (en) | Method and device for capturing face images | |
| CN104616032A (en) | Multi-camera system target matching method based on deep-convolution neural network | |
| CN104077577A (en) | Trademark detection method based on convolutional neural network | |
| CN111653023A (en) | Intelligent factory supervision method | |
| Kumar et al. | Automatic vehicle number plate recognition system using machine learning | |
| CN109460787A (en) | IDS Framework method for building up, device and data processing equipment | |
| CN105894701A (en) | Large construction vehicle identification and alarm method for preventing external damage to transmission lines | |
| CN107480679A (en) | The road network extracting method of classification and connected component analysis based on convolutional neural networks | |
| CN106250911A (en) | A kind of picture classification method based on convolutional neural networks |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20170707 |