CN110398226A - A kind of monocular vision distance measuring method for advanced DAS (Driver Assistant System) - Google Patents
A kind of monocular vision distance measuring method for advanced DAS (Driver Assistant System) Download PDFInfo
- Publication number
- CN110398226A CN110398226A CN201910425229.2A CN201910425229A CN110398226A CN 110398226 A CN110398226 A CN 110398226A CN 201910425229 A CN201910425229 A CN 201910425229A CN 110398226 A CN110398226 A CN 110398226A
- Authority
- CN
- China
- Prior art keywords
- model
- transverse
- target
- longitudinal
- road surface
- 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
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C3/00—Measuring distances in line of sight; Optical rangefinders
- G01C3/10—Measuring distances in line of sight; Optical rangefinders using a parallactic triangle with variable angles and a base of fixed length in the observation station, e.g. in the instrument
- G01C3/12—Measuring distances in line of sight; Optical rangefinders using a parallactic triangle with variable angles and a base of fixed length in the observation station, e.g. in the instrument with monocular observation at a single point, e.g. coincidence type
Landscapes
- Physics & Mathematics (AREA)
- Electromagnetism (AREA)
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Image Analysis (AREA)
Abstract
The present invention is a kind of monocular vision distance measuring method for assisting driving.Objects ahead is detected by YOLOv3, and obtains the location information of target in the picture；Establish the transverse and longitudinal ranging model based on road surface end point；Neural network compensation model is established according to the dimension information of the error function of YOLOv3 and target location information in the picture and target, it is therefore an objective to reduce the Systematic Errors of transverse and longitudinal ranging model and the position error of YOLOv3 algorithm.Finally, obtaining the transverse and longitudinal range information of target in transverse and longitudinal ranging model of the result input based on road surface end point after error compensation.
Description
Technical field
The present invention relates to advanced auxiliary driving field more particularly to a kind of monocular vision distance measuring methods.
Background technique
Inevitably there is the case where absent-minded or error in judgement in driving procedure in driver.Advanced DAS (Driver Assistant System) goes out
Now driver can be reminded to make correct operation before danger occurs, so that driver's reduction effectively be helped to drive
Danger coefficient.One of the core function that warning function is advanced DAS (Driver Assistant System) is hit before anti-.The premise of early warning realization is hit before anti-
Be detect the target on road surface, and using road surface target positional information calculation target in the picture and the transverse and longitudinal of itself away from
From.
Conventionally used monocular vision distance measuring method is that road surface target is obtained using a target detection model in image
In location information, then by target in the picture location information input monocular ranging model in obtain target range itself
Transverse and longitudinal distance.Showing the location information obtained by target detection model according to actual experiment, there are certain errors.Together
When, there is also Systematic Errors for monocular vision ranging model itself.Influence of the two to distance measurement result can be with target object
Distance increases and increases, and is unfavorable for driving safety.
In order to solve the drawbacks described above of monocular vision location algorithm, need one kind that can reduce above two error simultaneously
Monocular distance measuring method.
This patent is in such a way that target detection error function and object detection results establish Neural Network Optimization model, very
Good solves the above problem.
Summary of the invention
Set forth herein a kind of monocular distance measuring method for assisting driving, specific embodiments are as follows:
(1) target detection is carried out using YOLOv3 target detection model, obtains the location information of road surface target in the picture.
(2) the transverse and longitudinal ranging model based on road surface end point is established, according to the positional information calculation of target in the picture
The transverse and longitudinal distance of target and itself.
(3) position and classification information of the error function of YOLOv3 target detection model and target detection acquisition are utilized
Neural network compensation model is established, the Systematic Errors of ranging model and the position error of target detection model are optimized.
(4) location information resulting after optimization is inputted into the transverse and longitudinal ranging model based on road surface end point, obtains target
Transverse and longitudinal distance.
Detailed description of the invention
Fig. 1 is overall plan flow chart
Fig. 2 is longitudinal ranging model schematic；
Fig. 3 is lateral ranging model schematic；
Fig. 4 is error compensation neural network model schematic diagram；
Specific embodiment
Further illustrate the present invention with reference to the accompanying drawings and detailed description: the present invention is monocular vision distance measuring method,
Implementation step is as follows:
Step1: realtime graphic is obtained by imaging sensor, the real-time image information that will acquire is calculated by bilinear interpolation
Method is passed in YOLOv3 target detection model after narrowing down to 608 × 608 sizes.Image is passed to after model, and model will export mesh
The coordinate information of mark in the picture.
Step2: longitudinal ranging model based on road surface end point is established.Fig. 2 is longitudinal ranging based on road surface end point
Model schematic.Wherein O is camera position, and E' is the road surface end point of the plane of delineation, and E is real world road surface end point,
It ascends the throne infinite point in world coordinate system.S is testee bottom in the position of real world.S' is that testee bottom exists
The position of the plane of delineation.G' is plane of delineation midpoint.G is plane of delineation midpoint in the position of real world.C' is camera bat
Bottom recently a bit of the ground distance plane of delineation taken the photograph.C is C' in the position of real world.d1For C to video camera O
Fore-and-aft distance.
OG'=f in Fig. 2, wherein f focal length of camera, if G', C', S', E' ordinate are n0, n1, n2, n3.If video camera
Pixel focal length αy=f/dy, You Tuke are derived by
It is dead end street since E' is corresponding, so OE is parallel to ground, and it is nothing in world coordinate system that E' is corresponding
Poor far point.There is figure that can be derived by
α=∠ EOG- ∠ SO (5)
β=∠ EOG+ ∠ COG (6)
(2)-(6), which are brought into (1) abbreviation, to be obtained
From the above equation, we can see that as long as measuring d1It can be obtained by video camera at a distance from the testee of front with n2.n2For
The boundary frame bottom ordinate that YOLOv3 target detection network generates in real time.
Step3: Fig. 3 is lateral ranging schematic diagram, and wherein the abscissa of S' and Q' in the picture is respectively u2, u4.
It can be obtained by geometrical relationship in Fig. 3:
S'Q'=u4-u2 (9)
As shown in Figure 3:
Convolution (5) (6) (8) (11) can obtain:
Wherein target is d in vehicle left side3Negative value, on the contrary it is positive value.
Step4: the location error compensation model based on YOLOv3 error function and neural network, neural network model are established
See Fig. 4.YOLOv3 single target loss function is defined as follows formula:
Wherein x, y, h, w are respectively object to be detected true top left co-ordinate in the picture and length and width.
For measured target image top left co-ordinate and length and width predicted value.C is the true probability that measured target is not background,For
Measured target is not the prediction probability of background.P is target prediction probability true value,For target prediction probability.Divided in target
In the correct situation of class,Discovery center point tolerance is close with length and width error value in an experiment, therefore will close in formula
It is merged into coordinate predicted portionsAnd in most samplesIt can be derived from:
Using error compensation, compensated testee bottom ordinate can get:
The Loss of formula (13) is the mean error of single target, and penalty coefficient γ is added in compensation formula:
In order to obtain the value of γ, author establishes neural network prediction model.It, will after training neural network model
The length of measured target and wide predicted value and n2As the training input of network, γ is exported as neural network forecast.Nerve net
The loss function of network is formula (18):
Cost=(dIt is vertical real-d2)2 (18)
The n' that finally compensation model is obtained2Bringing (17) into can be obtained compensated fore-and-aft distance.Lateral distance by
(12) it obtains, without compensation, the reason is that lateral distance error is smaller.
Claims (3)
1. a kind of monocular vision distance measuring method for assisting driving, the specific scheme is that being obtained using YOLOv3 target detection model
The position of road surface target in the picture；Establish the transverse and longitudinal ranging model based on road surface end point；Utilize the mistake for improving YOLOv3
Difference function, target position in the picture and classification information establish neural network compensation model, this model is for compensating ranging
The Systematic Errors of model and the position error of target detection model；Compensated data are inputted into ranging model, obtain ranging
As a result.
2. the transverse and longitudinal ranging model according to claim 1 based on road surface end point, it is characterised in that: the ranging model
Only need corresponding pixel points ordinate, road surface end point in the actual range and its image of known road surface point range image sensor
Ordinate in the picture, image midpoint transverse and longitudinal coordinate, image sensor pixel focal length, target transverse and longitudinal coordinate can be obtained mesh
The transverse and longitudinal distance of subject distance imaging sensor.
3. neural network compensation model according to claim 1, it is characterised in that: determine YOLOv3 target detection model
Position error and the Systematic Errors of transverse and longitudinal ranging model carry out error compensation, the study by neural network to data, prediction
A more accurate location information out.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910425229.2A CN110398226A (en) | 2019-05-21 | 2019-05-21 | A kind of monocular vision distance measuring method for advanced DAS (Driver Assistant System) |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910425229.2A CN110398226A (en) | 2019-05-21 | 2019-05-21 | A kind of monocular vision distance measuring method for advanced DAS (Driver Assistant System) |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110398226A true CN110398226A (en) | 2019-11-01 |
Family
ID=68323638
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910425229.2A Pending CN110398226A (en) | 2019-05-21 | 2019-05-21 | A kind of monocular vision distance measuring method for advanced DAS (Driver Assistant System) |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110398226A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111784657A (en) * | 2020-06-29 | 2020-10-16 | 福建中航赛凡信息科技有限公司 | Digital image-based system and method for automatically identifying cement pavement diseases |
| CN112525147A (en) * | 2020-12-08 | 2021-03-19 | 北京嘀嘀无限科技发展有限公司 | Distance measurement method for automatic driving equipment and related device |
| CN113375687A (en) * | 2021-05-12 | 2021-09-10 | 武汉极目智能技术有限公司 | Method, system and device for compensating vanishing points of lane lines based on parallel constraint |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107389026A (en) * | 2017-06-12 | 2017-11-24 | 江苏大学 | A kind of monocular vision distance-finding method based on fixing point projective transformation |
| CN108830131A (en) * | 2018-04-10 | 2018-11-16 | 中科院微电子研究所昆山分所 | Traffic target detection and distance measuring method based on deep learning |
| CN109035322A (en) * | 2018-07-17 | 2018-12-18 | 重庆大学 | A kind of detection of obstacles and recognition methods based on binocular vision |
| CN109117794A (en) * | 2018-08-16 | 2019-01-01 | 广东工业大学 | A kind of moving target behavior tracking method, apparatus, equipment and readable storage medium storing program for executing |
| US20190012551A1 (en) * | 2017-03-06 | 2019-01-10 | Honda Motor Co., Ltd. | System and method for vehicle control based on object and color detection |
| CN109489620A (en) * | 2019-01-12 | 2019-03-19 | 内蒙古农业大学 | A kind of monocular vision distance measuring method |
-
2019
- 2019-05-21 CN CN201910425229.2A patent/CN110398226A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190012551A1 (en) * | 2017-03-06 | 2019-01-10 | Honda Motor Co., Ltd. | System and method for vehicle control based on object and color detection |
| CN107389026A (en) * | 2017-06-12 | 2017-11-24 | 江苏大学 | A kind of monocular vision distance-finding method based on fixing point projective transformation |
| CN108830131A (en) * | 2018-04-10 | 2018-11-16 | 中科院微电子研究所昆山分所 | Traffic target detection and distance measuring method based on deep learning |
| CN109035322A (en) * | 2018-07-17 | 2018-12-18 | 重庆大学 | A kind of detection of obstacles and recognition methods based on binocular vision |
| CN109117794A (en) * | 2018-08-16 | 2019-01-01 | 广东工业大学 | A kind of moving target behavior tracking method, apparatus, equipment and readable storage medium storing program for executing |
| CN109489620A (en) * | 2019-01-12 | 2019-03-19 | 内蒙古农业大学 | A kind of monocular vision distance measuring method |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111784657A (en) * | 2020-06-29 | 2020-10-16 | 福建中航赛凡信息科技有限公司 | Digital image-based system and method for automatically identifying cement pavement diseases |
| CN112525147A (en) * | 2020-12-08 | 2021-03-19 | 北京嘀嘀无限科技发展有限公司 | Distance measurement method for automatic driving equipment and related device |
| CN112525147B (en) * | 2020-12-08 | 2022-11-08 | 北京嘀嘀无限科技发展有限公司 | Distance measurement method for automatic driving equipment and related device |
| CN113375687A (en) * | 2021-05-12 | 2021-09-10 | 武汉极目智能技术有限公司 | Method, system and device for compensating vanishing points of lane lines based on parallel constraint |
| CN113375687B (en) * | 2021-05-12 | 2023-06-02 | 武汉极目智能技术有限公司 | Method, system and device for lane line vanishing point compensation based on parallel constraint |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11348266B2 (en) | Estimating distance to an object using a sequence of images recorded by a monocular camera | |
| CN106919915B (en) | Map road marking and road quality acquisition device and method based on ADAS system | |
| CN105835880B (en) | Lane following system | |
| CN102806913B (en) | Novel lane line deviation detection method and device | |
| JP5631581B2 (en) | Road recognition device | |
| CN110398226A (en) | A kind of monocular vision distance measuring method for advanced DAS (Driver Assistant System) | |
| JP2002175535A (en) | Road white line recognizing device | |
| CN104616502B (en) | Car license recognition and alignment system based on combination type bus or train route video network | |
| CN106256606A (en) | A kind of lane departure warning method based on vehicle-mounted binocular camera | |
| CN110031829A (en) | A kind of targeting accuracy distance measuring method based on monocular vision | |
| CN109829365A (en) | More scenes based on machine vision adapt to drive the method for early warning that deviates and turn | |
| CN103196418A (en) | Measuring method of vehicle distance at curves | |
| JP2003016430A (en) | Travel path detecting device | |
| CN104751151A (en) | Method for identifying and tracing multiple lanes in real time | |
| CN107097794B (en) | System and method for detecting lane line of road | |
| Liu et al. | Development of a vision-based driver assistance system with lane departure warning and forward collision warning functions | |
| CN111694011A (en) | Road edge detection method based on data fusion of camera and three-dimensional laser radar | |
| CN111967360A (en) | Target vehicle attitude detection method based on wheels | |
| CN114719873B (en) | Low-cost fine map automatic generation method and device and readable medium | |
| CN115273005A (en) | Visual navigation vehicle environment perception method based on improved YOLO algorithm | |
| CN108846363A (en) | A kind of subregion vehicle bottom shadow detection method based on divergence expression scanning | |
| JP2004355139A (en) | Vehicle recognition system | |
| CN114663859A (en) | Sensitive and accurate complex road condition lane deviation real-time early warning system | |
| CN105320928A (en) | Object recognition apparatus | |
| JP3925285B2 (en) | Road environment detection 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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20191101 |