CN110378201A - A kind of hinged angle measuring method of multiple row vehicle based on side ring view fisheye camera input - Google Patents
A kind of hinged angle measuring method of multiple row vehicle based on side ring view fisheye camera input Download PDFInfo
- Publication number
- CN110378201A CN110378201A CN201910486361.4A CN201910486361A CN110378201A CN 110378201 A CN110378201 A CN 110378201A CN 201910486361 A CN201910486361 A CN 201910486361A CN 110378201 A CN110378201 A CN 110378201A
- Authority
- CN
- China
- Prior art keywords
- angle
- hinged
- fisheye camera
- splice
- multiple row
- 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
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B11/00—Measuring arrangements characterised by the use of optical techniques
- G01B11/26—Measuring arrangements characterised by the use of optical techniques for measuring angles or tapers; for testing the alignment of axes
-
- 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
- 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
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Landscapes
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to field of vehicle control, more particularly to a kind of hinged angle measuring method of multiple row vehicle based on side ring view fisheye camera input, the following steps are included: A) installation fisheye camera, the ROI section of original image is mapped in aerial view, label creating sample image；B it) establishes and neural network is divided in training image identification；C splice angle Recurrent networks) are established and trained, will be exported after bound amendment and filter correction, as hinged angle measurement.Substantial effect of the invention is: splice angle is obtained by image procossing in real time, viewing system can be integrated in, additional hardware cost will not be introduced, it is not easy to it is disturbed, it is stable and reliable in work；Splice angle estimation is carried out based on tire semantic feature, compared to the method based on body contour feature, robustness is stronger with transplantability, promotes the use of at low cost.
Description
Technical field
The present invention relates to field of vehicle control, and in particular to a kind of multiple row vehicle based on side ring view fisheye camera input is hinged
Angle measuring method.
Background technique
More piece truck combination has the advantages that loading capacity is big, conevying efficiency is high, fuel-economizing is good and transportation cost is low etc., gradually at
For the main force of highway freight.With the continuous construction and the continuous development of economy of network of highways, more and more semi-dragging truck column
Vehicle will participate in shipping.The semi-trailer train being made of tractor, semitrailer is falling back dynamics state when driving
Unstable and uncontrollable, it is easy to happen the reversings accident such as folding, collision, rollover.It is insufficient for those driving experiences
New hand driver completes reversing task and is difficult.Therefore, it is necessary to electronic technology, automatic control technology are applied in trailer,
To assist the control of multiple row vehicle, the safety of more train drivings is improved.Currently, for the control stability automatic control side of bicycle
Method have been relatively mature, but to more piece truck combination, due to needing using splice angle as control variable to guarantee truck combination
Coordination and stablize traveling, thus need to consider the measuring technique of angular displacement.It is directly measured by angle transducer and obtains automobile column
The splice angle of vehicle is the method routinely used in current non-vision measurement method, but this measurement method needs established angle to sense
Device, the restricted application of angle transducer, sensors with auxiliary electrode are only applicable to certain specific articulated form, and need to consider that angle passes
The failure and maintenance of sensor.And vision measuring method, such as based on the rear view camera for being installed on tractor rear portion, based on to trailer vehicle
The vision contour feature of body, which is analyzed, to carry out, and such method is more difficult to non-box-type vehicle application, and is based on traditional characteristic
Method it is poor for the robustness of scene changes.In addition, such vision system can be only used to splice angle estimation, Wu Fayong
In support other application.It thus needs to develop a kind of reliable and stable hinged angular measurement skill of the accurate multiple row vehicle of measurement result
Art.
Chinese patent CN102582686B, publication date on September 18th, 2013, a kind of measurement of truck combination hinge angle and control
System is mainly made of truck combination hinge angle measurement part and splice angle control section, specifically includes GPS tester, input
Equipment, microprocessor, steering wheel angle sensor, active front wheel steering controller, executing agency and power supply unit, GPS test
Instrument, input equipment, steering wheel angle sensor are connected with microprocessor respectively, the steering wheel angle sensor also with vehicle side
It is connected to disk, microprocessor is connected with active front wheel steering controller, and active front wheel steering controller is connected with executing agency, holds
Row mechanism is connected with vehicle steering mechanism, and whole system is powered by the power supply unit.The system passes through the GPS location meter of train
Train splice angle is calculated, although its device is less likely to be damaged.But it needs to rely only on good GPS signal, and to GPS device
The requirement of installation site precision is also very high ON TRAINS, disposes low efficiency.
Summary of the invention
The technical problem to be solved by the present invention is it is hinged to lack the simple and fast and high measurement accuracy multiple row vehicle of deployment at present
The technical issues of angular measurement scheme.A kind of hinged angle measuring method of multiple row vehicle based on side ring view fisheye camera input is proposed,
This method has the advantages that deployment is simple and convenient, measurement accuracy is high.
In order to solve the above technical problems, the technical solution used in the present invention are as follows: one kind is defeated based on side ring view fisheye camera
The hinged angle measuring method of multiple row vehicle entered, comprising the following steps: A) in train two sides installation fisheye camera, calibration fisheye camera is former
The ROI section of original image is mapped to using hinge joint between train as the aerial view of coordinate origin by the ROI section of beginning image
In, zoom to pre-set dimension, as real-time sampling image, stores several sampled images and make after manually mark tire region
For sample image；B image recognition segmentation neural network) is established, the real-time sampling image obtained for step A is inputted, is exported to divide
Not Biao Shi pixel be tire region probability and pixel be background area two matrixes of probability, by indicate pixel be tire region
Probability matrix binaryzation after be used as real-time recognition matrix, the sample image training image identification segmentation stored using step A
Neural network, after output handmarking's splice angle of image recognition segmentation neural network, as sample matrix；C it) establishes hinged
Angle Recurrent networks are inputted the real-time recognition matrix obtained for step B, exported as hinge angle, the sample matrix obtained using step B
After training, by output of the splice angle Recurrent networks under real-time recognition matrix, as hinged angle measurement.
Preferably, when neural network is divided in training image identification, using loss function L in step BtyreIndicate nerve
Network output valve at a distance from true value,
Wherein, x, y are the pixel coordinate of input picture, and w, h are respectively the pixel wide and height of input picture, gX, yIt is corresponding
The artificial calibration semantic label of coordinate, value { 0,1 }, 0 indicates background area, and 1 indicates tire region, pX, yFor image recognition point
Neural network is cut in the output valve of respective coordinates, Value Types are percentage, indicate that the position is the percentage of tire region.This is excellent
Selecting scheme is to be directed in the prior art, the mode that the evaluation function of neural metwork training directly compares predicted value and mark value,
It is suitable only for single output nerve network.And problem to be solved by this invention, the output for the image recognition segmentation neural network being related to
It is matrix form.Evaluation function is to need to reflect whole superiority and inferiority, is avoided again coarse.Direction such as is modified in a certain parameter, is made
More partial pixel is exported toward correctly predicted direction, but the direction for also resulting in another fewer parts pixel prediction to mistake is defeated
Out, according to directly comparison and the simply evaluation function that is added in the prior art, will lead to the parameter modify direction obtain it is higher
Evaluation, cause neural network further to the direction modification parameter can not although evaluation function result is more excellent on the whole
Raising neural network integrally identifies accurate.This preferred embodiment is used and is taken using natural logrithm e as the logarithm at bottom, and with loss
Function is as evaluation.Modifying direction when a certain parameter keeps partial pixel result more accurate, but partial pixel result more mistake simultaneously
When, which modifies score of the direction on evaluation function can't be very high, and even better than logarithmic function is non-linear, instead
The score decline that will lead on evaluation function avoids falling into shifting office so that neural network adjusting parameter modifies direction
Portion's trap.
Preferably, will indicate that pixel is the method for the matrix binaryzation of the probability of tire region in step B are as follows:
Wherein, (x, y) is pixel coordinate, and M (x, y) indicates the element value of real-time recognition matrix, ch1(x, y) be pixel (x,
It y) is the probability of tire region, ch0(x, y) is the probability that pixel (x, y) is background area, TminFor given threshold.
Preferably, in sampled images after manually mark tire region, carrying out sample extended operation, specifically in step A
To use random cropping, rotation, distortion or color conversion operation, new sampled images are produced.
Preferably, when neural network is divided in training image identification, using batch stochastic gradient descent method, being asked in step B
Solve the optimum combination of neural network parameter.
Preferably, demarcating the ROI section of fisheye camera original image specifically includes the following steps: A11 in step A)
When train steady, the artificial ROI region for demarcating fisheye camera original image, and the phase on train body is selected for ROI region
To static reference；A12) when train driving, train low-speed motion model is established, input quantity is speed, length over ends of body, car body hinge
Position and steering wheel angle are connect, geometric modelling is carried out, determines opposing stationary reference position, and then dynamically determines ROI region.
Preferably, when training splice angle Recurrent networks, using loss function L in step CrIndicate that splice angle returns net
Network output valve at a distance from true value,
Wherein, Γ is splice angle Recurrent networks to splice angle predicted value, and G is handmarking's splice angle.
Preferably, further including to hinged angular measurement after splice angle Recurrent networks export hinged angle measurement in step C
Value carries out bound amendment, it may be assumed that
Γk=max (Γk-1-ΔΓmax, min (Γk, Γk-1+ΔΓmax))
Γk=max (Γmin, min (Γk, Γmax))
Wherein, ΓkFor the hinged angle measurement of current time splice angle Recurrent networks output, Γk-1For last moment hinged angular convolution
The hinged angle measurement for returning network to export, Δ ΓmaxIt is the maximum value of splice angle variable quantity absolute value in time at intervals, ΓminWith
And ΓmaxThe respectively lower limit value and upper limit value of splice angle.
Preferably, further including to hinged angular measurement after splice angle Recurrent networks export hinged angle measurement in step C
Value carries out Kalman Filtering for Discrete amendment, and institute's adoption status equation and observational equation are as follows:
xk+1=Axk+Buk+wk
zk+1=cxk+vk
Wherein, xkIt is four dimensional vectors, including tractor side velocity, yaw velocity, splice angle and splice angle for quantity of state
Speed, zkFor observed quantity, i.e. splice angle, ukFor system input, i.e. steering wheel angle, wkFor process noise, vkFor observation noise；
State space matrices A, B, C details are as follows:
E=[C1 b 0 0]T；
A=M-1N；
B=M-1E；
C=[0 00 1]；
Wherein, m1, m2 are tractor and trailer quality；J1, J2 are tractor and trailer yaw rotation inertia；A, b, c are respectively
Tractor front axle and tractor mass center, the distance of the 5th wheel and rear axle；D, e are respectively that trailer mass center and rear axle are taken turns to the 5th
Distance；C1, C2, C3 are tire cornering stiffness；V is longitudinal speed.
Preferably, further including using discrete low pass after splice angle Recurrent networks export hinged angle measurement in step C
It is filtered, the hinged angle measurement after discrete low-pass filtering, as final hinged angle measurement.
Substantial effect of the invention is: splice angle is obtained by image procossing in real time, viewing system can be integrated in, it will not
Introduce additional hardware cost, it is not easy to it is disturbed, it is stable and reliable in work；Splice angle estimation, phase are carried out based on tire semantic feature
Compared with the method based on body contour feature, robustness is stronger with transplantability, promotes the use of at low cost；Auxiliary compensation tractor with
Relative motion between trailer is realized and hingedly looks around panorama auxiliary display；As the input parameter of vehicle low-speed motion controller,
The accuracy that vehicle automatically controls is improved, the realization for active steering and automatic parking provides support.
Detailed description of the invention
Fig. 1 is the hinged angle measuring method flow diagram of one multiple row vehicle of embodiment.
Fig. 2 is one ROI region schematic diagram of embodiment.
Wherein: 1, tractor, 2, preceding fisheye camera, 3, ROI region, 4, rear fisheye camera, 5, trailer.
Specific embodiment
Below by specific embodiment, and in conjunction with attached drawing, a specific embodiment of the invention is further described in detail.
Embodiment one: a kind of hinged angle measuring method of multiple row vehicle based on side ring view fisheye camera input, as shown in Figure 1,
The present embodiment is the following steps are included: A) fisheye camera is installed in train two sides, the ROI section of fisheye camera original image is demarcated,
The ROI section of original image is mapped to using hinge joint between train to zoom to pre-set dimension in the aerial view of coordinate origin,
As real-time sampling image, several sampled images are stored and after manually mark tire region as sample image.Tire is
It General Motors component and is contacted with road surface, applicability and robustness are relative to wider based on other vehicle parts such as vehicle bodies
It is general.Installation side ring in position as shown in Figure 2 regards camera.I.e. height is located at 5 near top of trailer, and angle defaults oblique 45 ° towards ground.
Preceding fisheye camera 2 and rear fisheye camera 4 include two fisheye cameras.Preceding fisheye camera 2 is mounted on the two sides of tractor 1,
Fisheye camera 4 is mounted on the two sides at 5 middle part of trailer afterwards.When using the present embodiment method, 2 He of fisheye camera before can enabling simultaneously
Fisheye camera 4 afterwards, fisheye camera 2 or rear fisheye camera 4 before can also only enabling.As long as at least one fisheye camera can cover
Side rear-wheel in 1 two sides rear-wheel of lid tractor.Thus the ROI region 3 of the present embodiment is 1 two sides rear-wheel front of tractor
To 5 front of trailer and to two sides extend two regions.When tractor 1 keeps keeping straight on, the rear-wheel of 1 two sides of tractor is whole
Or part, in the region, when tractor 1 turns to, the rear-wheel of at least one tractor 1 can be still fallen in the two regions.It is logical
Cross neural metwork training, so that it may directly by 1 rear wheel position of tractor, export splice angle.
The pixel at the original visual angle of bird is generated to the mapping table of vertical view 3 pixel of visual angle ROI region using following steps.A11) exist
When train steady, the artificial ROI region 3 for demarcating fisheye camera original image, and be the selection of ROI region 3 on train body
Opposing stationary reference；A12) when train driving, train low-speed motion model is established, input quantity is speed, length over ends of body, car body
Articulated position and steering wheel angle carry out geometric modelling, determine opposing stationary reference position, and then the dynamic determining area ROI
Domain 3.Artificial calibration tire region and joint angle.In sampled images after manually mark tire region, sample extended operation is carried out,
Random cropping, rotation, distortion or color conversion operation are specially used, new sampled images are produced.
B image recognition segmentation neural network) is established, the real-time sampling image obtained for step A is inputted, is exported as table respectively
Show the probability that pixel is tire region and two matrixes of probability that pixel is background area, uses loss function LtyreIndicate nerve
Network output valve, using batch stochastic gradient descent method, solves the optimum combination of neural network parameter at a distance from true value,
Wherein, x, y are the pixel coordinate of input picture, and w, h are respectively the pixel wide and height of input picture, gX, yIt is corresponding
The artificial calibration semantic label of coordinate, pX, yDivide neural network in the output valve of respective coordinates for image recognition.
Pixel will be indicated to be used as real-time recognition matrix after the matrix binaryzation of the probability of tire region, will indicate that pixel is
The method of the matrix binaryzation of the probability of tire region are as follows:
Wherein, (x, y) is pixel coordinate, and M (x, y) indicates the element value of real-time recognition matrix, ch1(x, y) be pixel (x,
It y) is the probability of tire region, ch0(x, y) is the probability that pixel (x, y) is background area, TminFor given threshold.
The sample image training image identification segmentation neural network stored using step A, divides nerve net for image recognition
After output handmarking's splice angle of network, as sample matrix.
C splice angle Recurrent networks) are established, the real-time recognition matrix obtained for step B is inputted, are exported as hinge angle, use
Loss function LrIndicate splice angle Recurrent networks output valve at a distance from true value,
Wherein, Γ is splice angle Recurrent networks to splice angle predicted value, and G is handmarking's splice angle.To hinged angle measurement into
The amendment of row bound, it may be assumed that
Γk=max (Γk-1-ΔΓmax, min (Γk, Γk-1+ΔΓmax))
Γk=max (Γmin, min (Γk, Γmax))
Wherein, ΓkFor the hinged angle measurement of current time splice angle Recurrent networks output, Γk-1For last moment hinged angular convolution
The hinged angle measurement for returning network to export, Δ ΓmaxIt is the maximum value of splice angle variable quantity absolute value in time at intervals, ΓminWith
And ΓmaxThe respectively lower limit value and upper limit value of splice angle.Kalman Filtering for Discrete amendment is carried out to hinged angle measurement, is adopted
It is as follows with observational equation with state equation:
xk+1=Axk+Buk+wk
zk+1=Cxk+vk
Wherein, xkIt is four dimensional vectors for quantity of state, including 1 side velocity of tractor, yaw velocity, splice angle and hinged
Angular speed, zkFor observed quantity, i.e. splice angle, ukFor system input, i.e. steering wheel angle, wkFor process noise, vkIt makes an uproar for observation
Sound；State space matrices A, B, C details are as follows:
E=[C1 b 0 0]T；
A=M-1N；
B=M-1E；
C=[0 00 1]；
Wherein, m1, m2 are tractor 1 and 5 mass of trailer；J1, J2 are tractor 1 and 5 yaw rotation inertia of trailer；A, b, c points
Not Wei 1 front axle of tractor and 1 mass center of tractor, the distance of the 5th wheel and rear axle；D, e are respectively 5 mass center of trailer and rear axle
To the distance of the 5th wheel；C1, C2, C3 are tire cornering stiffness；V is longitudinal speed.When complete above-mentioned parameter can not be acquired, adopt
With discrete low-pass filtering treatment, hinged angle measurement after discrete low-pass filtering, as final hinged angle measurement.Pass through figure
Splice angle is obtained in real time as handling, viewing system can be integrated in, will not introduce additional hardware cost, it is not easy to disturbed, work
It is reliable and stable；Carry out splice angle estimation based on tire semantic feature, compared to the method based on body contour feature, robustness with
Transplantability is stronger, promotes the use of at low cost.
Above-mentioned embodiment is only a preferred solution of the present invention, not the present invention is made in any form
Limitation, there are also other variations and modifications on the premise of not exceeding the technical scheme recorded in the claims.
Claims (10)
1. a kind of hinged angle measuring method of multiple row vehicle based on side ring view fisheye camera input, which is characterized in that
The following steps are included:
A fisheye camera) is installed in train two sides, the ROI section of fisheye camera original image is demarcated, by the portion ROI of original image
It point is mapped to using hinge joint between train to zoom to pre-set dimension in the aerial view of coordinate origin, as real-time sampling image,
Store several sampled images and after manually mark tire region and splice angle as sample image；
B image recognition segmentation neural network) is established, the real-time sampling image obtained for step A is inputted, exports to respectively indicate picture
Element be tire region probability and pixel be background area probability two matrixes, will indicate that pixel is the probability of tire region
Real-time recognition matrix is used as after matrix binaryzation, the sample image training image identification segmentation neural network stored using step A,
After the output of image recognition segmentation neural network is associated with corresponding handmarking's splice angle, as sample matrix；
C splice angle Recurrent networks) are established, the real-time recognition matrix obtained for step B is inputted, exports as hinge angle, uses step B
After the sample matrix training of acquisition, by output of the splice angle Recurrent networks under real-time recognition matrix, as hinged angle measurement.
2. a kind of hinged angle measuring method of multiple row vehicle based on side ring view fisheye camera input according to claim 1,
It is characterized in that,
In step B, when neural network is divided in training image identification, loss function L is usedtyreIndicate neural network output valve and true
The distance of real value,
Wherein, x, y are the pixel coordinate of input picture, and w, h are respectively the pixel wide and height of input picture, gX, yIt is corresponding
The artificial calibration semantic label of coordinate, pX, yDivide neural network in the output valve of respective coordinates for image recognition.
3. a kind of hinged angle measuring method of multiple row vehicle based on side ring view fisheye camera input according to claim 1 or 2,
It is characterized in that,
In step B, it will indicate that pixel is the method for the matrix binaryzation of the probability of tire region are as follows:
Wherein, (x, y) is pixel coordinate, and M (x, y) indicates the element value of real-time recognition matrix, ch1(x, y) be pixel (x,
It y) is the probability of tire region, ch0(x, y) is the probability that pixel (x, y) is background area, TminFor given threshold.
4. a kind of hinged angle measuring method of multiple row vehicle based on side ring view fisheye camera input according to claim 1 or 2,
It is characterized in that,
In step A, in sampled images after manually mark tire region, sample extended operation is carried out, is specially cut out using random
It cuts, rotate, distorting or color conversion operation, producing new sampled images.
5. a kind of hinged angle measuring method of multiple row vehicle based on side ring view fisheye camera input according to claim 2,
It is characterized in that,
In step B, when neural network is divided in training image identification, batch stochastic gradient descent method is used, solves neural network ginseng
Several optimum combinations.
6. a kind of hinged angle measuring method of multiple row vehicle based on side ring view fisheye camera input according to claim 1 or 2,
It is characterized in that,
In step A, demarcate fisheye camera original image ROI section specifically includes the following steps:
A11) in train steady, the artificial ROI region for demarcating fisheye camera original image, and select for ROI region in train
Opposing stationary reference on car body；
A12) when train driving, train low-speed motion model is established, input quantity is speed, length over ends of body, car body articulated position
And steering wheel angle, geometric modelling is carried out, determines opposing stationary reference position, and then dynamically determines ROI region.
7. a kind of hinged angle measuring method of multiple row vehicle based on side ring view fisheye camera input according to claim 1 or 2,
It is characterized in that,
In step C, when training splice angle Recurrent networks, loss function L is usedrIndicate splice angle Recurrent networks output valve and true
The distance of value,
Wherein, Γ is splice angle Recurrent networks to splice angle predicted value, and G is handmarking's splice angle.
8. a kind of hinged angle measuring method of multiple row vehicle based on side ring view fisheye camera input according to claim 1 or 2,
It is characterized in that,
It further include carrying out bound to hinged angle measurement to repair after splice angle Recurrent networks export hinged angle measurement in step C
Just, it may be assumed that
Γk=max (Γk-1-ΔΓmax, min (Γk, Γk-1+ΔΓmax))
Γk=max (Γmin, min (Γk, Γmax))
Wherein, ΓkFor the hinged angle measurement of current time splice angle Recurrent networks output, Γk-1For last moment hinged angular convolution
The hinged angle measurement for returning network to export, Δ ΓmaxIt is the maximum value of splice angle variable quantity absolute value in time at intervals, ΓminWith
And ΓmaxThe respectively lower limit value and upper limit value of splice angle.
9. a kind of hinged angle measuring method of multiple row vehicle based on side ring view fisheye camera input according to claim 1 or 2,
It is characterized in that,
It further include that discrete karr is carried out to hinged angle measurement after splice angle Recurrent networks export hinged angle measurement in step C
Graceful filter correction, institute's adoption status equation and observational equation are as follows:
xk+1=Axk+Buk+wk
zk+1=Cxk+vk
Wherein, xkIt is four dimensional vectors, including tractor side velocity, yaw velocity, splice angle and splice angle for quantity of state
Speed, zkFor observed quantity, i.e. splice angle, ukFor system input, i.e. steering wheel angle, wkFor process noise, vkFor observation noise；
State space matrices A, B, C details are as follows:
E=[C1 b 0 0]T；
A=M-1N；
B=M-1E；
C=[0 00 1]；
Wherein, m1, m2 are tractor and trailer quality；J1, J2 are tractor and trailer yaw rotation inertia；A, b, c are respectively
Tractor front axle and tractor mass center, the distance of the 5th wheel and rear axle；D, e are respectively that trailer mass center and rear axle are taken turns to the 5th
Distance；C1, C2, C3 are tire cornering stiffness；V is longitudinal speed.
10. a kind of hinged angular measurement side of multiple row vehicle based on side ring view fisheye camera input according to claim 1 or 2
Method, which is characterized in that
It further include using discrete low-pass filtering treatment after splice angle Recurrent networks export hinged angle measurement in step C, it is discrete
Hinged angle measurement after low-pass filtering, as final hinged angle measurement.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910486361.4A CN110378201A (en) | 2019-06-05 | 2019-06-05 | A kind of hinged angle measuring method of multiple row vehicle based on side ring view fisheye camera input |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910486361.4A CN110378201A (en) | 2019-06-05 | 2019-06-05 | A kind of hinged angle measuring method of multiple row vehicle based on side ring view fisheye camera input |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110378201A true CN110378201A (en) | 2019-10-25 |
Family
ID=68249755
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910486361.4A Pending CN110378201A (en) | 2019-06-05 | 2019-06-05 | A kind of hinged angle measuring method of multiple row vehicle based on side ring view fisheye camera input |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110378201A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111637852A (en) * | 2020-05-27 | 2020-09-08 | 中国汽车技术研究中心有限公司 | System and method for measuring articulation angle of full-trailer automobile train |
| CN112613603A (en) * | 2020-12-25 | 2021-04-06 | 西安交通大学 | Neural network training method based on amplitude limiter and application thereof |
| CN115205311A (en) * | 2022-07-15 | 2022-10-18 | 小米汽车科技有限公司 | Image processing method, image processing apparatus, vehicle, medium, and chip |
| CN116499420A (en) * | 2023-05-23 | 2023-07-28 | 清华大学 | Method and system for measuring pinch angle between semitrailer and tractor |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106274907A (en) * | 2016-08-12 | 2017-01-04 | 浙江零跑科技有限公司 | A kind of many trains splice angle vision measurement optimization method based on Kalman filtering |
| CN108263283A (en) * | 2018-01-25 | 2018-07-10 | 长沙立中汽车设计开发股份有限公司 | More marshalling varied angle vehicle panoramic viewing system calibration and joining method |
| CN108961269A (en) * | 2018-06-22 | 2018-12-07 | 深源恒际科技有限公司 | Pig weight measuring method and system based on image |
| CN109190171A (en) * | 2018-08-02 | 2019-01-11 | 武汉中海庭数据技术有限公司 | A method of the vehicle movement model optimization based on deep learning |
| CN109191515A (en) * | 2018-07-25 | 2019-01-11 | 北京市商汤科技开发有限公司 | A kind of image parallactic estimation method and device, storage medium |
| CN109712118A (en) * | 2018-12-11 | 2019-05-03 | 武汉三江中电科技有限责任公司 | A kind of substation isolating-switch detection recognition method based on Mask RCNN |
| CN109711295A (en) * | 2018-12-14 | 2019-05-03 | 北京航空航天大学 | A kind of remote sensing image offshore Ship Detection |
| CN109741395A (en) * | 2018-12-14 | 2019-05-10 | 北京市商汤科技开发有限公司 | Biventricular quantization method, device, electronic equipment and storage medium |
| US20190147245A1 (en) * | 2017-11-14 | 2019-05-16 | Nuro, Inc. | Three-dimensional object detection for autonomous robotic systems using image proposals |
-
2019
- 2019-06-05 CN CN201910486361.4A patent/CN110378201A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106274907A (en) * | 2016-08-12 | 2017-01-04 | 浙江零跑科技有限公司 | A kind of many trains splice angle vision measurement optimization method based on Kalman filtering |
| US20190147245A1 (en) * | 2017-11-14 | 2019-05-16 | Nuro, Inc. | Three-dimensional object detection for autonomous robotic systems using image proposals |
| CN108263283A (en) * | 2018-01-25 | 2018-07-10 | 长沙立中汽车设计开发股份有限公司 | More marshalling varied angle vehicle panoramic viewing system calibration and joining method |
| CN108961269A (en) * | 2018-06-22 | 2018-12-07 | 深源恒际科技有限公司 | Pig weight measuring method and system based on image |
| CN109191515A (en) * | 2018-07-25 | 2019-01-11 | 北京市商汤科技开发有限公司 | A kind of image parallactic estimation method and device, storage medium |
| CN109190171A (en) * | 2018-08-02 | 2019-01-11 | 武汉中海庭数据技术有限公司 | A method of the vehicle movement model optimization based on deep learning |
| CN109712118A (en) * | 2018-12-11 | 2019-05-03 | 武汉三江中电科技有限责任公司 | A kind of substation isolating-switch detection recognition method based on Mask RCNN |
| CN109711295A (en) * | 2018-12-14 | 2019-05-03 | 北京航空航天大学 | A kind of remote sensing image offshore Ship Detection |
| CN109741395A (en) * | 2018-12-14 | 2019-05-10 | 北京市商汤科技开发有限公司 | Biventricular quantization method, device, electronic equipment and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| YUANZHOUHAN CAO 等: "Exploiting Depth from Single Monocular Images for Object Detection and Semantic Segmentation", 《APPEARING IN IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
| 谢元澄 等: "小麦麦穗几何表型测量的精准分割方法研究", 《南京农业大学学报》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111637852A (en) * | 2020-05-27 | 2020-09-08 | 中国汽车技术研究中心有限公司 | System and method for measuring articulation angle of full-trailer automobile train |
| CN111637852B (en) * | 2020-05-27 | 2023-03-21 | 中国汽车技术研究中心有限公司 | System and method for measuring articulation angle of full-trailer automobile train |
| CN112613603A (en) * | 2020-12-25 | 2021-04-06 | 西安交通大学 | Neural network training method based on amplitude limiter and application thereof |
| CN112613603B (en) * | 2020-12-25 | 2023-04-07 | 西安交通大学 | Neural network training method based on amplitude limiter and application thereof |
| CN115205311A (en) * | 2022-07-15 | 2022-10-18 | 小米汽车科技有限公司 | Image processing method, image processing apparatus, vehicle, medium, and chip |
| CN115205311B (en) * | 2022-07-15 | 2024-04-05 | 小米汽车科技有限公司 | Image processing method, device, vehicle, medium and chip |
| CN116499420A (en) * | 2023-05-23 | 2023-07-28 | 清华大学 | Method and system for measuring pinch angle between semitrailer and tractor |
| CN116499420B (en) * | 2023-05-23 | 2023-10-17 | 清华大学 | Method and system for measuring pinch angle between semitrailer and tractor |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110378201A (en) | A kind of hinged angle measuring method of multiple row vehicle based on side ring view fisheye camera input | |
| CN109074069B (en) | Autonomous vehicle with improved visual detection capability | |
| US7969326B2 (en) | Parking assist method and parking assist apparatus | |
| CN107554430A (en) | Vehicle blind zone view method, apparatus, terminal, system and vehicle | |
| US11693423B2 (en) | Model for excluding vehicle from sensor field of view | |
| CN111142091B (en) | Automatic driving system laser radar online calibration method fusing vehicle-mounted information | |
| CN107229908A (en) | A kind of method for detecting lane lines | |
| DE112015004011B4 (en) | VEHICLE IMAGE PROCESSING DEVICE | |
| CN114572183A (en) | Automobile pavement self-adaptive vehicle control method and equipment | |
| CN107284455B (en) | A kind of ADAS system based on image procossing | |
| DE112017007019T5 (en) | Vehicle display control device | |
| CN108107897B (en) | Real-time sensor control method and device | |
| CN114556253A (en) | Sensor field of view in self-driving vehicles | |
| CN106157386A (en) | Vehicular video filming control method and device | |
| CN114715158A (en) | Device and method for measuring road adhesion coefficient based on road texture features | |
| CN115503747A (en) | Road condition identification and reminding system based on intelligent automobile steer-by-wire system | |
| EP3815044B1 (en) | Method for sensor and memory-based depiction of an environment, display apparatus and vehicle having the display apparatus | |
| CN106601076B (en) | A kind of automobile self training device and method based on inertial navigation and area array cameras | |
| DE102016208369A1 (en) | Method for determining data representing part of the environment below the vehicle | |
| EP1283406A2 (en) | Image processing apparatus for a vehicle and method for the same | |
| CN104504677B (en) | A kind of image generating method based on variable visual angle | |
| CN114290995B (en) | Implementation method and device of transparent A column, automobile and medium | |
| DE102016208370A1 (en) | Method for determining data representing an environment below a vehicle | |
| CN115294211A (en) | Vehicle-mounted camera installation external parameter calibration method, system, device and storage medium | |
| CN107082049A (en) | Vehicle glazing control device |
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 | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 310051 1st and 6th floors, no.451 Internet of things street, Binjiang District, Hangzhou City, Zhejiang Province Applicant after: Zhejiang Zero run Technology Co.,Ltd. Address before: 310051 1st and 6th floors, no.451 Internet of things street, Binjiang District, Hangzhou City, Zhejiang Province Applicant before: ZHEJIANG LEAPMOTOR TECHNOLOGY Co.,Ltd. |
|
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191025 |