CN107122740A - A kind of vehicle type recognition device and method based on twin camera - Google Patents
A kind of vehicle type recognition device and method based on twin camera Download PDFInfo
- Publication number
- CN107122740A CN107122740A CN201710287184.8A CN201710287184A CN107122740A CN 107122740 A CN107122740 A CN 107122740A CN 201710287184 A CN201710287184 A CN 201710287184A CN 107122740 A CN107122740 A CN 107122740A
- Authority
- CN
- China
- Prior art keywords
- msub
- mrow
- vehicle
- image
- alpha
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/80—Analysis of captured images to determine intrinsic or extrinsic camera parameters, i.e. camera calibration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Traffic Control Systems (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of vehicle type recognition device based on twin camera and method, wherein model recognizing method comprises the following steps：Step a pair of dual camera is demarcated, and obtains camera inside and outside parameter；Step 2 carries out vehicle detection and segmentation to the image collected, obtains vehicle image；Step 3 carries out feature extraction and selection to vehicle image, obtains representing the feature of vehicle shape；Step 4 carries out vehicle classification to vehicle image；Step 5 exports recognition result；Vehicle type recognition device includes：Ccd video camera, IMAQ and the processing module of two groups of same models, data transmission module and power management module；Present invention employs binocular camera, computer vision, machine Learning Theory etc. are make use of, the accuracy rate problem recognized under different illumination conditions is improved, greatly reduces vehicle cab recognition cost, large-scale vehicle cab recognition work can be carried out.
Description
Technical field
The invention belongs to technical field of intelligent traffic, particularly a kind of vehicle type recognition device and side based on twin camera
Method.
Background technology
Intelligent transportation system is the developing direction of future transportation system, and it is by advanced information technology, data communication biography
Transferring technology, Electronic transducer technology, control technology and computer technology etc., which are effectively integrated, applies to whole ground traffic control management system
System and set up it is a kind of in a wide range of, it is comprehensive play a role, in real time, accurately and efficiently composite communications transport management system
System.
The vehicle flow of road pavement traveling, vehicle classification are counted, to the vehicle density, lane occupancy ratio, car of road
Road average speed etc. carry out calculating can obtain vehicle supervision department in policies, take measures and professional etiquette entered to means of transportation
Draw, design when science, objective foundation the most.As can be seen here, a complete intelligent transportation system is built, top priority is
The traffic information acquisition system of traffic characteristic parameters can accurately and efficiently be obtained by building one, and the core of this system is related to
The detection of real-time vehicle and identification technology for traffic scene.
Up to the present, vehicle detects two stages of the development experience with identification technology in real time.First stage be based on
" traditional approach " of the detectors such as line of induction ring type, microwave type, ultrasonic type, infrared light curtain formula.Second stage is based on image
The vehicle detection of processing and identification technology.
The existing vehicle cab recognition based on image procossing is typically all that the information for being directed to vehicular sideview is identified, for electricity
The method that the vehicle frontal information that sub- eye is shot is identified is not a lot, and discrimination is not high enough, is mainly due to：
(1) when moving vehicle detection is with segmentation, information of vehicles is imperfect；(2) vehicle characteristics extracted are easily by picture noise, weather condition
Etc. extraneous factor influence.
The content of the invention
Technical problem solved by the invention is offer a kind of vehicle type recognition device and method based on twin camera, with
Discrimination is not high enough during solving vehicle cab recognition, the problem of easily being disturbed by extraneous factor.
The technical solution for realizing the object of the invention is：
A kind of vehicle type recognition device based on twin camera, including two groups of ccd video cameras, IMAQ and processing module,
Data transmission module and power management module；Two groups of ccd video cameras are connected with IMAQ with processing module；Described image is adopted
Collection be connected with processing module with data transmission module, power management module respectively with IMAQ and processing module and data transfer
Module is connected, for for each module for power supply；
Ccd video camera is used to carry out IMAQ to the car face position of the vehicle of traveling；
IMAQ controls two video cameras to be acquired vehicle with processing module, and the module receives two shootings
The vehicle image that machine is passed back, and two images are handled, including dual camera demarcation, vehicle detection and segmentation, vehicle spy
Extraction and vehicle classification processing are levied, vehicle information is finally parsed；
The model data obtained after IMAQ and processing module dissection process is converted to suitable peace by data transmission module
The data transmitted entirely are to application apparatus.
A kind of model recognizing method, comprises the following steps：
Step 1, to dual camera demarcate：The scaling method of single camera is used first, respectively obtains two video cameras
Inside and outside parameter；The position relationship between two video cameras is set up by one group of scaling point in same world coordinates again；
Step 2, image progress vehicle detection and segmentation to collecting：By two groups of camera acquisitions to image schemed
As fusion treatment, strengthen picture quality；Reuse Canny edge detection operators and rim detection is carried out to the image collected, will
The vehicle image normalized detected, vehicle cab recognition is carried out so as to follow-up；
Step 3, to vehicle image carry out feature extraction and selection：First according to the characteristics of vehicle image by car face image from upper
It is divided into all subregion under, extracts the piecemeal SURF characteristic points of car face image；SURF characteristic points distinguishing ability is analyzed again, every
Sub-regions extract characteristic of division vector of the characteristic value selection of SURF characteristic points for the subregion；
Step 4, to vehicle image carry out vehicle classification：According to dual camera external parameter obtained by calibrating and extraction
Vehicle image, calculates length of wagon and height, is large, medium and small three kinds of vehicles by vehicle Preliminary division；By grader identification not
Know vehicle shape；
Step 5, the vehicle information got outflow：The model data obtained after dissection process is converted to and is adapted to safety biography
Defeated data are to application apparatus.
The present invention compared with prior art, its remarkable advantage：
(1) present invention uses twin camera, using image processing techniques, compared to monocular-camera, be difficult by illumination and
Shooting angle influences, and can obtain high-quality vehicle image and vehicle related parameters；
(2) in terms of vehicle cab recognition, vehicle classification device employs machine Learning Theory and is trained, compared to tradition side
Method, improves recognition accuracy.
The present invention is described in further detail below in conjunction with the accompanying drawings.
Brief description of the drawings
Fig. 1 is the structural representation of the vehicle type recognition device of the invention based on twin camera.
Fig. 2 is the flow chart of the model recognizing method of the invention based on twin camera.
Embodiment
Below in conjunction with the accompanying drawings, the embodiment to invention is further described.Following examples are only used for more clear
Illustrate to Chu technical scheme, and can not be limited the scope of the invention with this.
Fig. 1 be the vehicle type recognition device based on twin camera structural representation, a kind of vehicle type recognition device of the invention,
Including two groups of ccd video cameras (A video cameras and B video cameras), IMAQ and processing module, data transmission module and power management
Module；Two groups of ccd video cameras are connected with IMAQ with processing module；Described image is gathered and processing module and data transfer
Module is connected, and power management module is connected with IMAQ with processing module and data transmission module respectively, for for each module
Power supply.
Ccd video camera is used to carry out IMAQ to the car face position of the vehicle of traveling；IMAQ and processing module control
Two video cameras of system are acquired to vehicle, and the module receives the vehicle image that two video cameras are passed back, and to two width figures
As being handled, including dual camera demarcation, vehicle detection and segmentation, vehicle characteristics are extracted and vehicle classification is handled, finally
Parse vehicle information；
It is preferred that, two groups of ccd video camera models are identical, between two groups of ccd video camera afterbodys angle between 50 °~80 °,
It it is 3~4 meters with respect to ground level, field of view length is 20~150 meters.Two groups of ccd video cameras are installed on A-frame.
Described image is gathered includes submodule with processing module：Detection and segmentation module, feature extraction and selection module, car
Type sort module；
Detection is with segmentation module to carry out vehicle detection and segmentation to the image collected：Two groups of camera acquisitions are arrived
Image carry out image co-registration processing, strengthen picture quality；Canny edge detection operators are reused to carry out the image collected
Rim detection, by the vehicle image normalized detected；
It is preferred that, when image co-registration is handled, the point higher than threshold values is removed using median filter, threshold value here is every
The intermediate value of individual neighborhood of pixels (square area centered on current pixel), eliminates the mutation of pixel value, keeps light intensity continuous
Property；The vehicle image detected is normalized and stored, is easy to vehicle image feature extraction；For example, by the vehicle image being partitioned into
200 × 200 are uniformly normalized to by linear interpolation method.
Feature extraction and selection module to vehicle image to carry out feature extraction and selection：According to the characteristics of vehicle image
Car face image is divided into all subregion from top to bottom, the piecemeal SURF characteristic points of car face image are extracted；SURF characteristic points are differentiated
Every sub-regions, are extracted characteristic of division vector of the characteristic value selection of SURF characteristic points for the subregion by capability analysis；
Feature extraction and selection module includes feature extraction unit and feature selection unit；
Feature extraction unit：Car face image is divided into 4 sub-regions from top to bottom according to the characteristics of vehicle image, this 4
Subregion is respectively roof position, vehicle window position, front cover position, and the piecemeal SURF features of car face image are extracted at car plate position
Point；If SURF characteristic points are not present in some image-region, the 64 dimension SURF characteristic vectors for putting the image-region are null vector；
, whereas if there is SURF characteristic points, then the 64 all dimension SURF characteristic vectors to the image-region calculate average value；Calculate
Formula is：
In above formula,The corresponding 64 dimension SURF characteristic vectors in sub-image area arranged for the i-th row j, ni,jRepresent current figure
As the SURF features points detected altogether in region,Represent in the image-region the corresponding feature of each SURF characteristic points to
Amount；
Feature selection unit：SURF characteristic points distinguishing ability is analyzed, every sub-regions Extraction and discrimination ability is maximum
The characteristic value selection of preceding 10 SURF characteristic points is the characteristic of division vector of the subregion, and this characteristic of division vector will be as refreshing
Input vector through network is used for vehicle cab recognition；
Vehicle classification module is used to the vehicle image according to dual camera external parameter obtained by calibrating and extraction, calculates
Length of wagon and height, are large, medium and small three kinds of vehicles by vehicle Preliminary division；Unknown vehicle shape is recognized by grader.Car
Type sort module includes first module, second unit, third unit；
First module：According to dual camera external parameter obtained by calibrating (rotating vector and translation vector) and extract
Vehicle image, calculates length of wagon and height, is large, medium and small three kinds of vehicles by vehicle Preliminary division：
Arbitrfary point P coordinate (x on vehiclep,yp,zp) calculated by equation below：
Every video camera measures two angles, respectively video camera A horizontal azimuth αA, video camera A's vertically inclines
Angle betaA, video camera B horizontal azimuth αB, video camera B vertical dip angle βB.B is baseline length, i.e. between two video cameras away from
From；
Coordinate with final endpoint is put foremost by calculating vehicle, obtains length of wagon.By calculating vehicle roof most
The coordinate of high point, obtains bodywork height；
The criterion of vehicle size is：It is compact car that vehicle commander, which is less than or equal to 4.3 meters, and vehicle commander is 4.3 meters -4.9 meters and is
Type car, it is large car that vehicle commander, which is more than 4.9 meters,.
Second unit：Substantial amounts of vehicle image is collected, and the characteristic vector extracted is entered to the grader of vehicle shape
Row training, the method for the classification of vehicle shape is inputted in grader, allows vehicle classification device to produce classifying identification rule；
Third unit：Use the decision region of each vehicle shape class of the vehicle shape classifier calculated trained, identification
Unknown vehicle shape.
The model data obtained after IMAQ and processing module dissection process is converted to suitable peace by data transmission module
The data transmitted entirely are to application apparatus.
Based on the above-mentioned vehicle type recognition device based on twin camera, a kind of model recognizing method of proposition of the invention, including with
Lower step：
Step 1, to dual camera demarcate：The scaling method of single camera is used first, respectively obtains two video cameras
Inside and outside parameter；The position relationship between two video cameras is set up by one group of scaling point in same world coordinates again.
Specifically, the intrinsic parameter of video camera is：(three radial directions, two are cut for camera intrinsic parameter matrix and distortion factor
To)；The outer parameter of video camera is：Rotating vector (size is 1 × 3 vector or spin matrix 3 × 3) and translation vector.
Preferably, the method that the demarcation of single camera is converted using perspective matrix first solves the video camera ginseng of linear system
Number；Again in the hope of parameter be initial value, it is considered to distorted factor, and nonlinear solution, i.e., two shooting are tried to achieve using optimal method
The inside and outside parameter of machine.
Step 2, image progress vehicle detection and segmentation to collecting：By two groups of camera acquisitions to image schemed
As fusion treatment, strengthen picture quality；Reuse Canny edge detection operators and rim detection is carried out to the image collected, will
The vehicle image normalized detected, vehicle cab recognition is carried out so as to follow-up.
Preferably, when image co-registration is handled, the point higher than threshold values is removed using median filter, threshold value here is every
The intermediate value of individual neighborhood of pixels (square area centered on current pixel), eliminates the mutation of pixel value, keeps light intensity continuous
Property；The vehicle image detected is normalized and stored, is easy to vehicle image feature extraction；For example, by the vehicle image being partitioned into
200 × 200 are uniformly normalized to by linear interpolation method.
Step 3, to vehicle image carry out feature extraction and selection：First according to the characteristics of vehicle image by car face image from upper
It is divided into all subregion under, extracts the piecemeal SURF characteristic points of car face image；SURF characteristic points distinguishing ability is analyzed again, every
Sub-regions extract characteristic of division vector of the characteristic value selection of SURF characteristic points for the subregion, and this characteristic of division vector will
It is used for vehicle cab recognition as the input vector of neutral net.Specifically：
Car face image is first divided into 4 sub-regions by 3.1 from top to bottom according to the characteristics of vehicle image, this 4 sub-regions point
Wei not roof position, vehicle window position, front cover position, car plate position, the piecemeal SURF characteristic points of extraction car face image.If certain
SURF characteristic points are not present in individual image-region, then the 64 dimension SURF characteristic vectors for putting the image-region are null vector；, whereas if
There is SURF characteristic points, then the 64 all dimension SURF characteristic vectors to the image-region calculate average value.Calculation formula is：
In above formula,The corresponding 64 dimension SURF characteristic vectors in sub-image area arranged for the i-th row j, ni,jRepresent current figure
As the SURF features points detected altogether in region,Represent in the image-region the corresponding feature of each SURF characteristic points to
Amount.
3.2 analyze SURF characteristic points distinguishing ability again, first 10 of every sub-regions Extraction and discrimination ability maximum
The characteristic value selection of SURF characteristic points is vectorial for the characteristic of division of the subregion, and this characteristic of division vector will be used as neutral net
Input vector be used for vehicle cab recognition.
It will be appreciated that the dimension for the vehicle image characteristic point extracted in the present embodiment must be consistent, so to different
Vehicle image, the SURF characteristic points extracted all can be different in picture position and number, and the automatic of vehicle is carried out with this difference
Identification.
Step 4, to vehicle image carry out vehicle classification：According to dual camera external parameter obtained by calibrating and extraction
Vehicle image, calculates length of wagon and height, is large, medium and small three kinds of vehicles by vehicle Preliminary division；By grader identification not
Know vehicle shape.
4.1 first according to dual camera external parameter obtained by calibrating (rotating vector and translation vector) and the car extracted
Image, calculates length of wagon and height, is large, medium and small three kinds of vehicles by vehicle Preliminary division：
Arbitrfary point P coordinate (x on vehiclep,yp,zp) calculated by equation below：
Every video camera measures two angles, respectively video camera A horizontal azimuth αA, video camera A's vertically inclines
Angle betaA, video camera B horizontal azimuth αB, video camera B vertical dip angle βB.B is baseline length, i.e. between two video cameras away from
From.
Coordinate with final endpoint is put foremost by calculating vehicle, obtains length of wagon.By calculating vehicle roof most
The coordinate of high point, obtains bodywork height.
The criterion of vehicle size is：It is compact car that vehicle commander, which is less than or equal to 4.3 meters, and vehicle commander is 4.3 meters -4.9 meters and is
Type car, it is large car that vehicle commander, which is more than 4.9 meters,.
Then 4.2 collect substantial amounts of vehicle image again, and the characteristic vector extracted is entered to the grader of vehicle shape
Row training, will allow vehicle to divide in method (characteristic of division vector) i.e. in step 3.2 input grader of the classification of vehicle shape
Class device produces classifying identification rule；
The 4.3 last decision regions for reusing each vehicle shape class of the vehicle shape classifier calculated trained, identification
Unknown vehicle shape.
Preferably, sorting technique uses RBF neural, and an implicit sheaf space is constituted in hidden layer with RBF functions, this
Individual implicit sheaf space carries out nonlinear mapping to the input vector of neutral net.The core of grader is sorting algorithm herein,
Compared to conventional method, recognition accuracy is improved.
Step 5, the vehicle information got outflow：The model data obtained after dissection process is converted to and is adapted to safety biography
Defeated data are to application apparatus.
The embodiment of the present invention, using twin camera, using image processing techniques, compared to monocular-camera, is difficult light
Influenceed according to shooting angle, high-quality vehicle image and vehicle related parameters can be obtained；In terms of vehicle cab recognition, vehicle
Grader employs machine Learning Theory and is trained, compared to conventional method, improves recognition accuracy.
Described above is only the preferred embodiment for the present invention, it is noted that for those skilled in the art
For, without departing from the technical principles of the invention, it is regarded as protection scope of the present invention.
Claims (10)
1. a kind of vehicle type recognition device based on twin camera, it is characterised in that including two groups of ccd video cameras, IMAQ with
Processing module, data transmission module and power management module；Two groups of ccd video cameras are connected with IMAQ with processing module；
Described image collection is connected with processing module with data transmission module, power management module respectively with IMAQ and processing module
It is connected with data transmission module, for for each module for power supply；
Ccd video camera is used to carry out IMAQ to the car face position of the vehicle of traveling；
IMAQ controls two video cameras to be acquired vehicle with processing module, and the module receives two video cameras and passed
The vehicle image returned, and two images are handled, including dual camera demarcation, vehicle detection is with splitting, and vehicle characteristics are carried
Take and vehicle classification processing, finally parse vehicle information；
Data transmission module, which is converted to the model data obtained after IMAQ and processing module dissection process, is adapted to safety biography
Defeated data are to application apparatus.
2. the vehicle type recognition device as claimed in claim 1 based on twin camera, it is characterised in that described image is gathered and place
Manage module submodule：Detection and segmentation module, feature extraction and selection module, vehicle classification module；
Detection is with segmentation module to carry out vehicle detection and segmentation to the image collected：The figure that two groups of camera acquisitions are arrived
As carrying out image co-registration processing, strengthen picture quality；Reuse Canny edge detection operators and edge is carried out to the image collected
Detection, by the vehicle image normalized detected；
Feature extraction and selection module to vehicle image to carry out feature extraction and selection：First will according to the characteristics of vehicle image
Car face image is divided into all subregion from top to bottom, extracts the piecemeal SURF characteristic points of car face image；SURF characteristic points are differentiated again
Every sub-regions, are extracted characteristic of division vector of the characteristic value selection of SURF characteristic points for the subregion by capability analysis；
Vehicle classification module is used to the vehicle image according to dual camera external parameter obtained by calibrating and extraction, calculates vehicle body
Length and height, are large, medium and small three kinds of vehicles by vehicle Preliminary division；Unknown vehicle shape is recognized by grader.
3. the vehicle type recognition device as claimed in claim 2 based on twin camera, it is characterised in that the detection and segmentation mould
Block removes the point higher than threshold values when image co-registration is handled using median filter, eliminates the mutation of pixel value, keeps light intensity to connect
Continuous property.
4. the vehicle type recognition device as claimed in claim 2 based on twin camera, it is characterised in that the feature extraction and choosing
Selecting module includes feature extraction unit and feature selection unit；
Feature extraction unit：Car face image is divided into 4 sub-regions from top to bottom according to the characteristics of vehicle image, this 4 sub-districts
Domain is respectively roof position, vehicle window position, front cover position, and the piecemeal SURF characteristic points of car face image are extracted at car plate position；Such as
Really SURF characteristic points are not present in some image-region, then the 64 dimension SURF characteristic vectors for putting the image-region are null vector；Conversely,
If there is SURF characteristic points, then to the 64 all dimension SURF characteristic vectors calculating average values of the image-region；Calculation formula
For：
<mrow>
<mover>
<msub>
<mi>A</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>=</mo>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<mfrac>
<mn>1</mn>
<msub>
<mi>n</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
</mfrac>
<mi>&Sigma;</mi>
<mover>
<msub>
<mi>B</mi>
<mi>k</mi>
</msub>
<mo>&OverBar;</mo>
</mover>
</mrow>
</mtd>
<mtd>
<mrow>
<msub>
<mi>n</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>></mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mo>(</mo>
<mn>0</mn>
<mo>,</mo>
<mn>0</mn>
<mo>,</mo>
<mn>...</mn>
<mo>,</mo>
<mn>0</mn>
<mo>)</mo>
</mrow>
</mtd>
<mtd>
<mrow>
<msub>
<mi>n</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>=</mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
</mrow>
In above formula,The corresponding 64 dimension SURF characteristic vectors in sub-image area arranged for the i-th row j, ni,jRepresent present image area
In detect altogether SURF features points,Represent the corresponding characteristic vector of each SURF characteristic points in the image-region；
Feature selection unit：SURF characteristic points distinguishing ability is analyzed, before every sub-regions Extraction and discrimination ability maximum 10
The characteristic value selection of individual SURF characteristic points is vectorial for the characteristic of division of the subregion, and this characteristic of division vector will be used as nerve net
The input vector of network is used for vehicle cab recognition.
5. the vehicle type recognition device as claimed in claim 2 based on twin camera, it is characterised in that the vehicle classification module
Including first module, second unit, third unit；
First module：According to dual camera external parameter obtained by calibrating and the vehicle image of extraction, calculate length of wagon and
Highly, it is large, medium and small three kinds of vehicles by vehicle Preliminary division：Arbitrfary point P coordinate (x on vehiclep,yp,zp) by following public
Formula is calculated：
<mrow>
<msub>
<mi>x</mi>
<mi>p</mi>
</msub>
<mo>=</mo>
<mi>b</mi>
<mfrac>
<mrow>
<msub>
<mi>cos&alpha;</mi>
<mi>A</mi>
</msub>
<msub>
<mi>sin&alpha;</mi>
<mi>B</mi>
</msub>
</mrow>
<mrow>
<mi>s</mi>
<mi>i</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>&alpha;</mi>
<mi>A</mi>
</msub>
<mo>+</mo>
<msub>
<mi>&alpha;</mi>
<mi>B</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
</mrow>
<mrow>
<msub>
<mi>y</mi>
<mi>p</mi>
</msub>
<mo>=</mo>
<mi>b</mi>
<mfrac>
<mrow>
<msub>
<mi>sin&alpha;</mi>
<mi>A</mi>
</msub>
<msub>
<mi>sin&alpha;</mi>
<mi>B</mi>
</msub>
</mrow>
<mrow>
<mi>s</mi>
<mi>i</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>&alpha;</mi>
<mi>A</mi>
</msub>
<mo>+</mo>
<msub>
<mi>&alpha;</mi>
<mi>B</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
</mrow>
<mrow>
<msub>
<mi>z</mi>
<mi>p</mi>
</msub>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mn>2</mn>
</mfrac>
<mo>&lsqb;</mo>
<mi>b</mi>
<mfrac>
<mrow>
<msub>
<mi>sin&alpha;</mi>
<mi>B</mi>
</msub>
<msub>
<mi>tan&beta;</mi>
<mi>A</mi>
</msub>
<mo>+</mo>
<msub>
<mi>sin&alpha;</mi>
<mi>A</mi>
</msub>
<msub>
<mi>tan&beta;</mi>
<mi>B</mi>
</msub>
</mrow>
<mrow>
<mi>s</mi>
<mi>i</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>&alpha;</mi>
<mi>A</mi>
</msub>
<mo>+</mo>
<msub>
<mi>&alpha;</mi>
<mi>B</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>&rsqb;</mo>
</mrow>
Every video camera measures two angles, respectively video camera A horizontal azimuth αA, video camera A vertical dip angle βA,
Video camera B horizontal azimuth αB, video camera B vertical dip angle βB；B is baseline length, i.e. the distance between two video cameras；
Coordinate with final endpoint is put foremost by calculating vehicle, obtains length of wagon；By calculating vehicle roof peak
Coordinate, obtain bodywork height；
Second unit：Substantial amounts of vehicle image is collected again, and the characteristic vector extracted is carried out to the grader of vehicle shape
Training, the method for the classification of vehicle shape is inputted in grader, allows vehicle classification device to produce classifying identification rule；
Third unit：Using the decision region of each vehicle shape class of the vehicle shape classifier calculated trained, recognize unknown
Vehicle shape.
6. a kind of model recognizing method, it is characterised in that comprise the following steps：
Step 1, to dual camera demarcate：The scaling method of single camera is used first, respectively obtains the inside and outside of two video cameras
Parameter；The position relationship between two video cameras is set up by one group of scaling point in same world coordinates again；
Step 2, image progress vehicle detection and segmentation to collecting：By two groups of camera acquisitions to image carry out image melt
Conjunction is handled, and strengthens picture quality；Reuse Canny edge detection operators and rim detection is carried out to the image collected, will detect
The vehicle image normalized arrived, vehicle cab recognition is carried out so as to follow-up；
Step 3, to vehicle image carry out feature extraction and selection：First according to the characteristics of vehicle image by car face image from top to bottom
It is divided into all subregion, extracts the piecemeal SURF characteristic points of car face image；SURF characteristic points distinguishing ability is analyzed again, per height
The characteristic value selection of extracted region SURF characteristic points is vectorial for the characteristic of division of the subregion；
Step 4, to vehicle image carry out vehicle classification：According to dual camera external parameter obtained by calibrating and the vehicle of extraction
Image, calculates length of wagon and height, is large, medium and small three kinds of vehicles by vehicle Preliminary division；Unknown car is recognized by grader
Shape；
Step 5, the vehicle information got outflow：The model data obtained after dissection process is converted into suitable safe transmission
Data are to application apparatus.
7. a kind of model recognizing method as claimed in claim 6, it is characterised in that in step 2 when image co-registration is handled, profit
The point higher than threshold values is removed with median filter, the mutation of pixel value is eliminated, light intensity continuity is kept.
8. a kind of model recognizing method as claimed in claim 6, it is characterised in that feature is carried out to vehicle image in step 3
Extract with selecting, specifically include following steps：
Car face image is first divided into 4 sub-regions by 3.1 from top to bottom according to the characteristics of vehicle image, and this 4 sub-regions is respectively
Roof position, vehicle window position, front cover position, the piecemeal SURF characteristic points of car face image are extracted at car plate position；If some figure
As SURF characteristic points are not present in region, then put the 64 of the image-region and tie up SURF characteristic vectors for null vector；, whereas if in the presence of
SURF characteristic points, then to the 64 all dimension SURF characteristic vectors calculating average values of the image-region；Calculation formula is：
<mrow>
<mover>
<msub>
<mi>A</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>&OverBar;</mo>
</mover>
<mo>=</mo>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<mfrac>
<mn>1</mn>
<msub>
<mi>n</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
</mfrac>
<mi>&Sigma;</mi>
<mover>
<msub>
<mi>B</mi>
<mi>k</mi>
</msub>
<mo>&OverBar;</mo>
</mover>
</mrow>
</mtd>
<mtd>
<mrow>
<msub>
<mi>n</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>></mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mo>(</mo>
<mn>0</mn>
<mo>,</mo>
<mn>0</mn>
<mo>,</mo>
<mn>...</mn>
<mo>,</mo>
<mn>0</mn>
<mo>)</mo>
</mrow>
</mtd>
<mtd>
<mrow>
<msub>
<mi>n</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>=</mo>
<mn>0</mn>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
</mrow>
In above formula,The corresponding 64 dimension SURF characteristic vectors in sub-image area arranged for the i-th row j, ni,jRepresent present image area
In detect altogether SURF features points,Represent the corresponding characteristic vector of each SURF characteristic points in the image-region；
3.2 analyze SURF characteristic points distinguishing ability again, preceding 10 SURF spies that every sub-regions Extraction and discrimination ability is maximum
Characteristic of division vector of the characteristic value selection a little for the subregion is levied, this characteristic of division vector will be used as the input of neutral net
Vector is used for vehicle cab recognition.
9. a kind of model recognizing method as claimed in claim 8, it is characterised in that step 3.1 divides the standard of vehicle size
For：It is compact car that vehicle commander, which is less than or equal to 4.3 meters, and it is in-between car that vehicle commander, which is 4.3 meters -4.9 meters, and it is large car that vehicle commander, which is more than 4.9 meters,.
10. a kind of model recognizing method as claimed in claim 6, it is characterised in that vehicle is carried out to vehicle image in step 4
Classification, specifically includes following steps：
4.1, first according to dual camera external parameter obtained by calibrating and the vehicle image of extraction, calculate length of wagon and height
Degree, is large, medium and small three kinds of vehicles by vehicle Preliminary division：Arbitrfary point P coordinate (x on vehiclep,yp,zp) pass through equation below
Calculate：
<mrow>
<msub>
<mi>x</mi>
<mi>p</mi>
</msub>
<mo>=</mo>
<mi>b</mi>
<mfrac>
<mrow>
<msub>
<mi>cos&alpha;</mi>
<mi>A</mi>
</msub>
<msub>
<mi>sin&alpha;</mi>
<mi>B</mi>
</msub>
</mrow>
<mrow>
<mi>s</mi>
<mi>i</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>&alpha;</mi>
<mi>A</mi>
</msub>
<mo>+</mo>
<msub>
<mi>&alpha;</mi>
<mi>B</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
</mrow>
<mrow>
<msub>
<mi>y</mi>
<mi>p</mi>
</msub>
<mo>=</mo>
<mi>b</mi>
<mfrac>
<mrow>
<msub>
<mi>sin&alpha;</mi>
<mi>A</mi>
</msub>
<msub>
<mi>sin&alpha;</mi>
<mi>B</mi>
</msub>
</mrow>
<mrow>
<mi>s</mi>
<mi>i</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>&alpha;</mi>
<mi>A</mi>
</msub>
<mo>+</mo>
<msub>
<mi>&alpha;</mi>
<mi>B</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
</mrow>
<mrow>
<msub>
<mi>z</mi>
<mi>p</mi>
</msub>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mn>2</mn>
</mfrac>
<mo>&lsqb;</mo>
<mi>b</mi>
<mfrac>
<mrow>
<msub>
<mi>sin&alpha;</mi>
<mi>B</mi>
</msub>
<msub>
<mi>tan&beta;</mi>
<mi>A</mi>
</msub>
<mo>+</mo>
<msub>
<mi>sin&alpha;</mi>
<mi>A</mi>
</msub>
<msub>
<mi>tan&beta;</mi>
<mi>B</mi>
</msub>
</mrow>
<mrow>
<mi>s</mi>
<mi>i</mi>
<mi>n</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>&alpha;</mi>
<mi>A</mi>
</msub>
<mo>+</mo>
<msub>
<mi>&alpha;</mi>
<mi>B</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>&rsqb;</mo>
</mrow>
Every video camera measures two angles, respectively video camera A horizontal azimuth αA, video camera A vertical dip angle βA,
Video camera B horizontal azimuth αB, video camera B vertical dip angle βB；B is baseline length, i.e. the distance between two video cameras；
Coordinate with final endpoint is put foremost by calculating vehicle, obtains length of wagon；By calculating vehicle roof peak
Coordinate, obtain bodywork height；
Then 4.2 collect substantial amounts of vehicle image again, and the characteristic vector extracted is instructed to the grader of vehicle shape
Practice, the method for the classification of vehicle shape is inputted in grader, allow vehicle classification device to produce classifying identification rule；
The 4.3 last decision regions for reusing each vehicle shape class of the vehicle shape classifier calculated trained, are recognized unknown
Vehicle shape.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710287184.8A CN107122740A (en) | 2017-04-26 | 2017-04-26 | A kind of vehicle type recognition device and method based on twin camera |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710287184.8A CN107122740A (en) | 2017-04-26 | 2017-04-26 | A kind of vehicle type recognition device and method based on twin camera |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107122740A true CN107122740A (en) | 2017-09-01 |
Family
ID=59726364
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710287184.8A Pending CN107122740A (en) | 2017-04-26 | 2017-04-26 | A kind of vehicle type recognition device and method based on twin camera |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107122740A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108198426A (en) * | 2018-01-30 | 2018-06-22 | 河南郑大嘉源环保技术有限公司 | A kind of motor vehicle in tunnel experiment vehicle automatic identifying method and its system |
| CN109141595A (en) * | 2018-06-15 | 2019-01-04 | 东南大学 | A method of it is ceased by video identification vehicle dimension acquisition of information traffic axis information carrying |
| CN109271907A (en) * | 2018-09-03 | 2019-01-25 | 北京万集科技股份有限公司 | The control method and system of vehicle ETC equipment |
| CN109840474A (en) * | 2018-12-28 | 2019-06-04 | 广州粤建三和软件股份有限公司 | A kind of vehicle enters and leaves specification recognition methods, system and storage medium |
| CN109948610A (en) * | 2019-03-14 | 2019-06-28 | 西南交通大学 | A kind of vehicle fine grit classification method in the video based on deep learning |
| CN110414357A (en) * | 2019-06-28 | 2019-11-05 | 上海工程技术大学 | A kind of front vehicles localization method based on vehicle type recognition |
| CN111899389A (en) * | 2020-07-27 | 2020-11-06 | 上海福赛特智能科技有限公司 | Method for identifying vehicle type and license plate of fleet management system |
| CN111951563A (en) * | 2020-07-20 | 2020-11-17 | 苏州第四度信息科技有限公司 | Vehicle identification number detection and identification method and device |
| CN112015941A (en) * | 2020-09-03 | 2020-12-01 | 湖南同冈科技发展有限责任公司 | Intelligent vehicle body identification method and system in automobile production line |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104299008A (en) * | 2014-09-23 | 2015-01-21 | 同济大学 | Vehicle type classification method based on multi-feature fusion |
| CN105469046A (en) * | 2015-11-23 | 2016-04-06 | 电子科技大学 | Vehicle model identification method based on PCA and SURF characteristic cascade |
-
2017
- 2017-04-26 CN CN201710287184.8A patent/CN107122740A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104299008A (en) * | 2014-09-23 | 2015-01-21 | 同济大学 | Vehicle type classification method based on multi-feature fusion |
| CN105469046A (en) * | 2015-11-23 | 2016-04-06 | 电子科技大学 | Vehicle model identification method based on PCA and SURF characteristic cascade |
Non-Patent Citations (3)
| Title |
|---|
| 武宏伟 等: "一种基于支持向量机的车型自动分类器设计方案", 《计算机应用》 * |
| 武宏伟: "基于双目视觉的车辆自动识别与分类技术研究", 《WWW.IRGRID.AC.CN/HANDLE/1471X/483698》 * |
| 袁爱龙 等: "基于视频的汽车车型识别研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108198426A (en) * | 2018-01-30 | 2018-06-22 | 河南郑大嘉源环保技术有限公司 | A kind of motor vehicle in tunnel experiment vehicle automatic identifying method and its system |
| CN109141595B (en) * | 2018-06-15 | 2020-06-30 | 东南大学 | Method for acquiring traffic axle load information by identifying vehicle size information through video |
| CN109141595A (en) * | 2018-06-15 | 2019-01-04 | 东南大学 | A method of it is ceased by video identification vehicle dimension acquisition of information traffic axis information carrying |
| CN109271907A (en) * | 2018-09-03 | 2019-01-25 | 北京万集科技股份有限公司 | The control method and system of vehicle ETC equipment |
| CN109271907B (en) * | 2018-09-03 | 2021-03-23 | 北京万集科技股份有限公司 | Control method and system of vehicle ETC device |
| CN109840474A (en) * | 2018-12-28 | 2019-06-04 | 广州粤建三和软件股份有限公司 | A kind of vehicle enters and leaves specification recognition methods, system and storage medium |
| CN109948610A (en) * | 2019-03-14 | 2019-06-28 | 西南交通大学 | A kind of vehicle fine grit classification method in the video based on deep learning |
| CN110414357A (en) * | 2019-06-28 | 2019-11-05 | 上海工程技术大学 | A kind of front vehicles localization method based on vehicle type recognition |
| CN110414357B (en) * | 2019-06-28 | 2023-04-07 | 上海工程技术大学 | Front vehicle positioning method based on vehicle type identification |
| CN111951563A (en) * | 2020-07-20 | 2020-11-17 | 苏州第四度信息科技有限公司 | Vehicle identification number detection and identification method and device |
| CN111951563B (en) * | 2020-07-20 | 2022-01-14 | 苏州第四度信息科技有限公司 | Vehicle identification number detection and identification method and device |
| CN111899389A (en) * | 2020-07-27 | 2020-11-06 | 上海福赛特智能科技有限公司 | Method for identifying vehicle type and license plate of fleet management system |
| CN112015941A (en) * | 2020-09-03 | 2020-12-01 | 湖南同冈科技发展有限责任公司 | Intelligent vehicle body identification method and system in automobile production line |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107122740A (en) | A kind of vehicle type recognition device and method based on twin camera | |
| CN109460709B (en) | RTG visual barrier detection method based on RGB and D information fusion | |
| CN102722704B (en) | Method and system for recognizing vehicle license plate by integrating video dynamic tracking | |
| CN101739829B (en) | Video-based vehicle overspeed monitoring method and system | |
| CN103235938B (en) | The method and system of car plate detection and indentification | |
| CN103824066B (en) | A kind of licence plate recognition method based on video flowing | |
| CN104951784B (en) | A kind of vehicle is unlicensed and license plate shading real-time detection method | |
| CN103832433B (en) | Deviation and front truck anti-collision alarm system and its implementation | |
| CN103366154B (en) | Reconfigurable clear path detection system | |
| CN103258432A (en) | Traffic accident automatic identification processing method and system based on videos | |
| CN106127107A (en) | The model recognizing method that multi-channel video information based on license board information and vehicle's contour merges | |
| CN105404857A (en) | Infrared-based night intelligent vehicle front pedestrian detection method | |
| CN102880863B (en) | Method for positioning license number and face of driver on basis of deformable part model | |
| CN103065138A (en) | Recognition method of license plate number of motor vehicle | |
| CN103400113B (en) | Freeway tunnel pedestrian detection method based on image procossing | |
| CN109670449B (en) | Vehicle illegal judgment method based on vertical snapshot mode | |
| Sharma et al. | A hybrid technique for license plate recognition based on feature selection of wavelet transform and artificial neural network | |
| CN104063882A (en) | Vehicle video speed measuring method based on binocular camera | |
| CN108269407A (en) | A kind of automatic management stream of people, the security robot of logistics | |
| CN107358236A (en) | A kind of number-plate number identifying system and method based on camera device | |
| CN106778646A (en) | Model recognizing method and device based on convolutional neural networks | |
| CN105118305A (en) | Vehicle management platform of outdoor parking lot exit | |
| Ma et al. | An all-weather lane detection system based on simulation interaction platform | |
| CN113673326A (en) | Unmanned aerial vehicle platform crowd counting method and system based on image deep learning | |
| Su et al. | A new local-main-gradient-orientation HOG and contour differences based algorithm for object classification |
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 |
Application publication date: 20170901 |
|
| RJ01 | Rejection of invention patent application after publication |