CN107796373A - A kind of distance-finding method of the front vehicles monocular vision based on track plane geometry model-driven - Google Patents

A kind of distance-finding method of the front vehicles monocular vision based on track plane geometry model-driven Download PDF

Info

Publication number
CN107796373A
CN107796373A CN201710930473.5A CN201710930473A CN107796373A CN 107796373 A CN107796373 A CN 107796373A CN 201710930473 A CN201710930473 A CN 201710930473A CN 107796373 A CN107796373 A CN 107796373A
Authority
CN
China
Prior art keywords
vehicle
objects ahead
mrow
ahead vehicle
longitudinal
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.)
Granted
Application number
CN201710930473.5A
Other languages
Chinese (zh)
Other versions
CN107796373B (en
Inventor
刘永涛
赵俊玮
赖延年
乔洁
华珺
周文财
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Yingdong Technology Development Co ltd
Original Assignee
Changan University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Changan University filed Critical Changan University
Priority to CN201710930473.5A priority Critical patent/CN107796373B/en
Publication of CN107796373A publication Critical patent/CN107796373A/en
Application granted granted Critical
Publication of CN107796373B publication Critical patent/CN107796373B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C11/00—Photogrammetry or videogrammetry, e.g. stereogrammetry; Photographic surveying
    • G01C11/04—Interpretation of pictures
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/285—Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/277—Analysis of motion involving stochastic approaches, e.g. using Kalman filters
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/70—Determining position or orientation of objects or cameras
    • G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
    • 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
    • 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
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30248—Vehicle exterior or interior
    • G06T2207/30252—Vehicle exterior; Vicinity of vehicle

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Multimedia (AREA)
  • Remote Sensing (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Electromagnetism (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)
  • Traffic Control Systems (AREA)
  • Image Processing (AREA)

Abstract

A kind of distance-finding method of front vehicles monocular vision based on track plane geometry model-driven provided by the invention,Objects ahead vehicle image is gathered using CCD camera,Objects ahead vehicle in the objects ahead vehicle image obtained using merging in Haar like features and Adaboost algorithm identification step 2,Objects ahead vehicle acquired in step 3 is tracked using particle filter method,Longitudinal distance survey model of track plane geometry in the two field picture is built according to objects ahead vehicle in each two field picture of above-mentioned gained,Obtain target source longitudinal direction perceived distance y in the two field picture,Vehicle odometry error dynamic compensation model is built according to objects ahead vehicle in each two field picture of above-mentioned gained,Obtain longitudinal measurement error value z,Longitudinal spacing Y in the two field picture between objects ahead vehicle and our vehicle is calculated according to the target source longitudinal direction perceived distance y and longitudinal measurement error value z of gained in step 5W(P)。

Description

A kind of ranging of the front vehicles monocular vision based on track plane geometry model-driven Method
Technical field
The invention belongs to longitudinal direction of car safety assistant driving technical field, is related to a kind of based on the drive of track plane geometry model The distance-finding method of dynamic front vehicles monocular vision.
Background technology
Vehicle follow gallop is that driver a kind of most basic driving behavior, vehicle in traffic activity are faced with the stage of speeding Chief threat essentially from longitudinal vehicle rear-end impact, from car and front truck do not keep certain safe distance between vehicles and to from car and before The speed of car judges inaccurate and causes vehicle rear-end collision to be collided.Accurate research is from the spacing value of car and front vehicles for keeping car Spacing and vehicle collision prewarning are significant.
The mode that distance survey has been studied at present mainly have ultrasonic ranging, laser ranging, millimetre-wave radar ranging with And machine vision ranging.Ultrasonic ranging is only applicable to short distance ranging, and laser ranging and millimetre-wave radar ranging use into This is too high, and by contrast, machine vision distance measuring method hardware configuration is simple, cost is also low and obtains abundant information and easily, Therefore there is more preferable practical value and application prospect using machine vision metrology spacing.
Comprehensive every metering system is compared in machine vision, because monocular vision measurement processing data time is shorter, energy Meet ranging real-time, so measuring being in the great majority for spacing using monocular vision.
Based on monocular vision carry out spacing automatic measurement when, the positioning for front vehicles is extremely important, positioning it is accurate Property directly influences the accuracy of distance survey.Be currently based on the recognition methods of vehicle shadow by ambient factor influenceed compared with Greatly；The depth information of image is obtained by corresponding points demarcation based on tailstock mathematical model method, but because equipment limits and marks The reason such as fixed, can not obtain the transition matrix of phase co-conversion between degree of precision coordinate system, and applicability is restricted；Based on monocular When the processing of visual pattern gray processing carries out front vehicles identification, front vehicles afterbody profile can only be often identified, but due to car The presence of rear overhang terrain clearance, will certainly cause very big range error.
The content of the invention
It is an object of the invention to provide a kind of front vehicles monocular vision based on track plane geometry model-driven Distance-finding method, solve the influence of light line present in prior art, the conversion of high-precision coordinate system and the liftoff height of vehicle rear overhang The problem of spending range error caused by existing and defect.
In order to achieve the above object, the technical solution adopted by the present invention is：
A kind of distance-finding method of front vehicles monocular vision based on track plane geometry model-driven provided by the invention, Comprise the following steps：
Step 1, CCD camera is demarcated, obtains effective focal length f, CCD camera height h and CCD camera pitching Angle θ；
Step 2, objects ahead vehicle image is gathered using CCD camera；
Step 3, the objects ahead car for merging Haar-like features with being obtained in Adaboost algorithm identification step 2 is utilized Objects ahead vehicle in image；
Step 4, objects ahead vehicle acquired in step 3 is tracked using particle filter method；
Step 5, to build track plane in the two field picture according to objects ahead vehicle in each two field picture of above-mentioned gained several What longitudinal distance survey model, obtain target source longitudinal direction perceived distance y in the two field picture；
Step 6, vehicle odometry error dynamic compensation is built according to objects ahead vehicle in each two field picture of above-mentioned gained Model, obtain longitudinal measurement error value z；
Step 7, the frame is calculated according to the target source longitudinal direction perceived distance y and longitudinal measurement error value z of gained in step 5 Longitudinal spacing Y in image between objects ahead vehicle and our vehicleW(P)。
Preferably, in step 3, using before merging Haar-like features and being obtained in Adaboost algorithm identification step 2 The specific method of objects ahead vehicle in square mesh mark vehicle image：
S1, sample set is established according to the objects ahead vehicle image of gained in step 2, and sample is chosen using Adaboost algorithm Effective Haar-like features of the vehicle training sample of this concentration, weak typing corresponding to each effectively Haar-like features generation Device, Weak Classifier weighted array is become into strong classifier, is finally cascaded using cascade classifier, obtains feature samples Cascade classifier；
S2, take the vehicle training sample of magnanimity and effective Haar-like feature extractions are carried out to the vehicle training sample, it The cascade classifier that effective Haar-like features are input to feature samples afterwards carries out vehicle Detection of Existence, obtains Adaboost cascade classifiers；
S3, according to the fixed position of CCD camera, the area-of-interest in objects ahead vehicle image is determined, using S2 The Adaboost cascade classifiers of middle gained carry out vehicle Detection of Existence to area-of-interest, final to obtain objects ahead vehicle Objects ahead vehicle in image.
Preferably, in step 5, as obtained by being demarcated in step 1 to CCD camera, based on track plane geometry It is world coordinate system that subpoint of the photocentre of CCD camera as C points, photocentre C points on road surface is set in longitudinal distance survey model Origin O points, the X that the ranging characteristic point of front vehicles is P, the direction of forward travel is world coordinate systemWAxle, world coordinates The Z of systemWAxle perpendicular to road surface down, CCD camera imaging plane be A ' B ' F ' E ', long sight angle plane is CEF, optical axis center institute Plane be CMN and ranging characteristic point where plane be CC2D, wherein, the optical axis CC of CCD camera1With imaging plane A ' B ' F ' E ' intersection point is C0Point, then CC0For the focal length of ccd sensor, i.e. CC0=f；The near-sighted field picture of objects ahead vehicle in image Down contour point A and imaging plane A ' B ' F ' E ' meet at G ' points, and far visual field down contour point B and imaging plane A ' B ' F ' E ' meet at H points； The ranging characteristic point P of front vehicles projects to world coordinate system XWThe point of axle is P', wherein, target source longitudinal direction perceived distance is For OP ' length.
Preferably, in step 5, target source longitudinal direction perceived distance y calculation formula is：
In formula, v0For photocentre longitudinal direction image coordinate, v (P0) a longitudinal image coordinate is characterized, dy is the longitudinal direction of unit pixel Length.
Preferably, in step 6, the structure of objects ahead vehicle odometry error dynamic compensation model, following step is specifically included Suddenly：
The first step, fixed CCD vision sensors, with the inside and outside ginseng of the scaling method progress CCD camera described in step 1 Number is demarcated and recorded；
Second step, the terrain clearance in fixed target source, target source is moved every 5m along road longitudinal direction, makes it in distance Change in the range of CCD vision sensors [10m, 100m], and image of the target source in each position is recorded with CCD camera；
3rd step, the terrain clearance in adjustment target source, makes it change in [0.2m, 1m], repeats second step；
4th step, image obtained by the 3rd step is handled using matlab, analysis target source is in different terrain clearances and difference The measurement error of lengthwise position.
Preferably, in step 6, longitudinal measurement error value z calculation formula is：
Z=118-1124x-3.133y+3522x2+34.6xy-0.0399y2-4795x3-86.22x2y-0.1845xy2+ 0.0017y3+2676x4+98.62x3y+0.1428x2y2+0.001313xy3-2.077e-5y4+8.835e-8y5-7.658e-6xy4+ 0.0004x2y3-0.1114x3y2-38.23x4y-391.1x5
In formula, x is target source terrain clearance.
Preferably, in step 7, longitudinal spacing Y between objects ahead vehicle and our vehicleW(P) calculation formula is：
Compared with prior art, the beneficial effects of the invention are as follows：
A kind of distance-finding method of front vehicles monocular vision based on track plane geometry model-driven provided by the invention, Objects ahead vehicle image is gathered using CCD camera, using merging Haar-like features and Adaboost algorithm identification step Objects ahead vehicle in the objects ahead vehicle image obtained in 2, using particle filter method to front acquired in step 3 Target vehicle is tracked, and track plane in the two field picture is built according to objects ahead vehicle in each two field picture of above-mentioned gained Longitudinal distance survey model of geometry, target source longitudinal direction perceived distance y in the two field picture is obtained, according to each of above-mentioned gained Objects ahead vehicle builds vehicle odometry error dynamic compensation model in two field picture, longitudinal measurement error value z is obtained, according to step The target source longitudinal direction perceived distance y and longitudinal measurement error value z of gained calculate objects ahead vehicle and sheet in the two field picture in 5 Longitudinal spacing Y between square vehicleW(P)；The present invention is with hardware configuration is simple, cost is low and software algorithm is flexible big and surveys The characteristics of accuracy of measurement is higher, and can avoid the interference such as runway both sides shadows on the road, vehicle in this other non-track because The influence of element, improves the detection robustness of system, and the present invention efficiently solves the presence of vehicle rear overhang terrain clearance, improved The identification efficiency of front vehicles lengthwise position relation.
Brief description of the drawings
Fig. 1 is ccd video camera scheme of installation；
Fig. 2 is the flow and method of distance survey device of the present invention；
Fig. 3 is ccd video camera calibration of camera schematic diagram；
Fig. 4 is ccd video camera calibrating external parameters schematic diagram；
Fig. 5 is the vehicle identification algorithm structure chart based on Haar-like features and Adaboost；
Fig. 6 is the structure schematic diagram of feature samples cascade classifier；
Fig. 7 is target image vehicle's contour extraction schematic diagram；
Fig. 8 is ccd video camera imaging space geometrical-restriction relation figure；
Fig. 9 is track plane restriction ranging model side view；
Figure 10 is longitudinal distance survey Analysis of error source schematic diagram.
Embodiment
Below in conjunction with the accompanying drawings, the present invention is described in more detail.
A kind of as shown in figure 1, front vehicles monocular vision based on track plane geometry model-driven provided by the invention Range unit, including CCD camera 1, CCD camera 1 are fixed on position top among vehicle front windshield 2 using sucker Put.Wherein, CCD camera 1 is attached by BNC video lines and video frequency collection card with host computer.
A kind of as shown in Fig. 2 front vehicles monocular vision based on track plane geometry model-driven provided by the invention Distance-finding method, comprise the following steps that：
The demarcation of step 1, CCD camera：
The demarcation of vision sensor is the key issue for needing to solve in machine vision, and the purpose of demarcation is to obtain CCD The inner parameter and external parameter of camera, to the conversion of 3 D stereo scene, have to complete two dimensional image in subsequent step Body：
First, using the Camera Calibration Toolbox modules and plane target drone scaling board in MATLAB softwares Demarcation obtains the inner parameter of CCD camera；Embodiment is following (such as Fig. 3)：The angle of conversion demarcation plane target drone, profit The plane target drone image under 20 frame different azimuths, the Camera put it into MATLAB softwares are gathered with CCD camera CalibrationToolbox modules carry out the resolving of CCD camera inner parameter, and the inner parameter for obtaining CCD camera is effective Focal length f；
Secondly, it is following (such as Fig. 4) using the method based on road image end point, embodiment：In CCD camera On the plane target drone image of acquisition, left-lane line and right-lane line are marked, while records left-lane line and right-lane line intersection point O Pixel coordinate, and respectively take on the lane line of left and right a point from intersection point farther out as far as possible respectively, record 2 points of A, B pixel Coordinate, according to the actual range of CCD camera calibration of camera result and two parallel lines in three-dimensional world.Pass through HALCON Calibration modules demarcation in calibration tool case in software obtains the external parameter of CCD camera 1.It can resolve and obtain Obtain the external parameter CCD camera height h and CCD camera pitching angle theta of CCD camera 1.
The collection and transmission of step 2, target vehicle image：
CCD camera 1 gathers objects ahead vehicle image, and the objects ahead vehicle image of collection is passed through into BNC videos Line, video frequency collection card transmit the image processing software into master system, obtain the objects ahead car that can be analyzed and processed Image.
Step 3, utilize the objects ahead car for merging Haar-like features with being obtained in Adaboost algorithm identification step 2 Objects ahead vehicle in image, its idiographic flow (such as Fig. 5)：
First, the objects ahead vehicle image according to obtained by step 2 establishes sample set, and sample is trained to the vehicle in sample set This and non-vehicle training sample are pre-processed, i.e.,：Effective Haar- in vehicle training sample is chosen using Adaboost algorithm Like features, Weak Classifier corresponding to each validity feature generation, become strong classifier, last structure by Weak Classifier weighted array The cascade classifier of feature samples is built, specific method is as follows:
The first step, objects ahead vehicle image is converted into integrogram：Using integrogram method by objects ahead vehicle image The integrogram of the target vehicle image is obtained after conversion, after conversion, each point represents the image upper left corner after conversion in integrogram Point in the rectangular area pixel and, as shown in Equation 1.
Wherein, pixel integration values of the i (x, y) for (x, y) point on integral image, and i (x ', y ') it is (x, y) point in artwork Interior pixel value.
Second step, fast and effectively class Haar-like features are carried out to the integrogram after conversion using Adaboost algorithm Extraction, strong classifier of the generation with feature samples, specifically：Adaboost algorithm is corresponding by circulating extraction one every time Effective class Haar-like features, Weak Classifier corresponding to each validity feature generation, become strong by Weak Classifier weighted array Grader；
3rd step, the cascade classifier of construction feature sample：Most of region does not all include mesh in usual image to be detected Vehicle is marked, therefore non-vehicle region is quickly excluded using cascade classifier, improves target detection speed.The present invention is using classics Cascade classifier is cascaded, and each layer is to train obtained Adaboost strong classifiers with Adaboost algorithm, each Adaboost strong classifiers include several Weak Classifiers again, and sample graph to be identified is cascaded grader and examined in layer Survey, if being judged to negative sample, i.e. non-vehicle target image at any one layer, strong classifier below all can't pass, so may be used So that grader below has more times to identify positive sample window, i.e. target vehicle image, idiographic flow such as Fig. 6.Examine Consider the confidence level of vehicle target detection and the real-time of algorithm, cascade classifier used in the present invention use 8 layers of grader, finally Verification and measurement ratio more than 0.9, each layer of false drop rate is 0.5, then each layer of verification and measurement ratio is more than 0.99.
Secondly, Haar-like feature extractions are carried out to magnanimity test sample, inputs the feature into Adaboost cascade sorts Device carries out vehicle Detection of Existence, ensures Detection accuracy, detection rates and algorithm real-time, foregoing two big step construct Merge Haar-like features and Adaboost front vehicles recognizer.
Finally, according to the fixed position of CCD camera, it is determined that the ROI region of objects ahead vehicle image of the present invention is figure The latter half of picture, the area-of-interest in objects ahead vehicle image is handled, obtain vehicle identification result, specific side Method is：First with fusion Haar-like features and Adaboost algorithm to area-of-interest (ROI) with above-mentioned training process one Sample is handled, i.e.,：Image preprocessing is carried out to the area-of-interest on target image and calculates integrogram；Secondly with above-mentioned instruction The Haar-like characteristic informations selected during white silk, the Haar- of area-of-interest is extracted comprising structure, position, type etc. Like characteristic values, composition characteristic vector；The Adaboost cascade classifiers finally obtained using magnanimity off-line training are to interested Region carries out vehicle Detection of Existence, and exports vehicle identification result, as shown in Figure 7.This method ensure that Detection accuracy While with algorithm real-time, it is dry that runway both sides shadows on the road, vehicle in this other non-track etc. can be effectively prevented from The influence of factor is disturbed, improves the detection robustness of system.
Step 4, using particle filter method objects ahead vehicle is tracked：
During actual acquisition image, because target vehicle background image complexity is various, easily cause algorithm missing inspection or appearance The problems such as false-alarm.Therefore, in order to ensure the real-time of system and robustness, the present invention is using particle filter algorithm to carrying out front Vehicle tracking, specific tracking step are as follows：Particle initialization, time renewal, observation renewal step, resampling and state renewal. The population N of particle filter is set to be arranged to 100 herein, each iteration hits is 30.It can ensure that track algorithm has Higher tracking accuracy and stability, tracking process average take 20ms, and track algorithm is to type of vehicle, attitudes vibration, environment The non-determined factors such as interference have stronger immunocompetence, disclosure satisfy that onboard system real-time and robustness requirement.
Step 5, the longitudinal distance survey model for building track plane geometry：
Image can be handled each frame obtained above and carry out longitudinal distance survey model meter based on track plane geometry Calculate, longitudinal ranging model based on track plane geometry is established according to the installation site of CCD camera first, as shown in Figure 8.Its In to set subpoint of the photocentre of CCD camera as C points, photocentre C points on road surface be world coordinate system origin O points, front vehicles Ranging characteristic point be P, the X that the direction of forward travel is world coordinate systemWThe Z of axle, world coordinate systemWAxle is perpendicular to road Down, CCD camera imaging plane is A ' B ' F ' E ', long sight angle plane is CEF, plane where optical axis center is CMN and survey It is CC away from plane where characteristic point2D, wherein, the optical axis CC of CCD camera1Intersection point with imaging plane A ' B ' F ' E ' is C0Point, Then CC0For the focal length of ccd sensor, i.e. CC0=f；The near-sighted field picture down contour point A of objects ahead vehicle and imaging in image Plane A ' B ' F ' E ' meet at G ' points, and far visual field down contour point B and imaging plane A ' B ' F ' E ' meet at H points；The ranging of front vehicles Characteristic point P projects to world coordinate system XWThe point of axle is P', wherein, target source longitudinal direction perceived distance is OP ' length.
Secondly lower vehicle odometry model inference front vehicles are constrained in the position of world coordinate system according to above-mentioned road plane, As shown in Figure 9.Known C0、CC0, θ and imaging plane A ' B ' F ' E ' each length of side, pass through below equation and solve ranging characteristic point P's The ordinate value of world coordinates, it can be learnt from the side view of Fig. 8 tracks plane restriction ranging model：
CC0=f, ∠ OC1C=θ, OC=h (2)
In Δ C0CP′0In,
Wherein P '0C0It is desirable on the occasion of and negative value, have again：
In formula, P '0C0=y (C0)-y(P′0)=[v0-v(P′0)] × dy=[v0-v(P0)]×dy
In Δ OCP ', OP'=OC × tan ∠ OCP'
It can then derive,
In formula, v0For photocentre longitudinal direction image coordinate, v (P0) a longitudinal image coordinate is characterized, dy is the longitudinal direction of unit pixel Length, y are target source longitudinal direction perceived distance.
Step 6, structure vehicle odometry error dynamic compensation model：
As shown in Figure 10, objects ahead vehicle tail region is divided into region A and region B two by boundary of vehicle's contour lower edge Part, region A vertical height are the rear overhang height h of target vehicleα, point D is vehicle tail region lower edge midpoint, and point P is Vehicle tail region projects midpoint on road surface, because front vehicles identification algorithm can only detect objects ahead vehicle tail wheel Exterior feature, therefore selected point D make it that longitudinal distance survey value is bigger than normal as longitudinal distance survey characteristic point.
In order to reduce rear overhang height error to caused by longitudinal distance survey, longitudinal distance survey value is set accurately to force as far as possible Nearly actual value, it is 2m × 1m red objects target source as front vehicles tail region simulation object, mesh that the present invention, which uses specification, Target source is fixed on support and terrain clearance is adjustable, by a large amount of vehicle samples of analytic statistics, determines that vehicle rear overhang highly becomes It is [0.2m, 1m] to change scope.
Its specific implementation step is：
The first step, fixed CCD vision sensors, with the inside and outside ginseng of the scaling method progress CCD camera described in step 1 Number is demarcated and recorded；
Second step, the terrain clearance in fixed target source, target source is moved every 5m along road longitudinal direction, makes it in distance Change in the range of CCD vision sensors [10m, 100m], and image of the target source in each position is recorded with CCD camera；
3rd step, the terrain clearance in adjustment target source, makes it change in [0.2m, 1m], repeats second step；
4th step, image obtained by the 3rd step is handled using matlab, analysis target source is vertical in different terrain clearances, difference To the measurement error of position.According to target source in different terrain clearances, the error information of different longitudinal position, longitudinal direction can be returned Measurement error dynamic compensation model is：
According to target source in different terrain clearances, the error information of different longitudinal position, longitudinal measurement error can be returned Dynamic compensation model is：
Wherein：X is target source terrain clearance in formula, and y is target source longitudinal direction perceived distance, and z is longitudinal measurement error.
Step 7, spacing calculate：
According to vehicle odometry in the structure of longitudinal distance survey model based on track plane geometry in step 5 and step 6 Spacing is reconstructed the structure of error dynamic compensation model, then the longitudinal distance survey model reconstructed is：

Claims (7)

  1. A kind of 1. distance-finding method of the front vehicles monocular vision based on track plane geometry model-driven, it is characterised in that bag Include following steps：
    Step 1, CCD camera is demarcated, obtains effective focal length f, CCD camera height h and CCD camera pitching angle theta；
    Step 2, objects ahead vehicle image is gathered using CCD camera；
    Step 3, the objects ahead vehicle figure for merging Haar-like features with being obtained in Adaboost algorithm identification step 2 is utilized Objects ahead vehicle as in；
    Step 4, objects ahead vehicle acquired in step 3 is tracked using particle filter method；
    Step 5, track plane geometry in the two field picture is built according to objects ahead vehicle in each two field picture of above-mentioned gained Longitudinal distance survey model, obtain target source longitudinal direction perceived distance y in the two field picture；
    Step 6, vehicle odometry error dynamic compensation model is built according to objects ahead vehicle in each two field picture of above-mentioned gained, Obtain longitudinal measurement error value z；
    Step 7, the two field picture is calculated according to the target source longitudinal direction perceived distance y and longitudinal measurement error value z of gained in step 5 Longitudinal spacing Y between middle objects ahead vehicle and our vehicleW(P)。
  2. A kind of 2. ranging of front vehicles monocular vision based on track plane geometry model-driven according to claim 1 Method, it is characterised in that：In step 3, using merging what is obtained in Haar-like features and Adaboost algorithm identification step 2 The specific method of objects ahead vehicle in objects ahead vehicle image：
    S1, sample set is established according to the objects ahead vehicle image of gained in step 2, and sample set is chosen using Adaboost algorithm In vehicle training sample effective Haar-like features, each effectively Haar-like features produce corresponding to Weak Classifier, Weak Classifier weighted array is become into strong classifier, finally cascaded using cascade classifier, obtains the level of feature samples Join grader；
    S2, take the vehicle training sample of magnanimity and effective Haar-like feature extractions are carried out to the vehicle training sample, afterwards will The cascade classifier that effective Haar-like features are input to feature samples carries out vehicle Detection of Existence, obtains Adaboost levels Join grader；
    S3, according to the fixed position of CCD camera, the area-of-interest in objects ahead vehicle image is determined, using institute in S2 The Adaboost cascade classifiers obtained carry out vehicle Detection of Existence to area-of-interest, final to obtain objects ahead vehicle image In objects ahead vehicle.
  3. A kind of 3. ranging of front vehicles monocular vision based on track plane geometry model-driven according to claim 1 Method, it is characterised in that：In step 5, as obtained by being demarcated in step 1 to CCD camera, based on track plane geometry It is world coordinate system that subpoint of the photocentre of CCD camera as C points, photocentre C points on road surface is set in longitudinal distance survey model Origin O points, the X that the ranging characteristic point of front vehicles is P, the direction of forward travel is world coordinate systemWAxle, world coordinates The Z of systemWAxle perpendicular to road surface down, CCD camera imaging plane be A ' B ' F ' E ', long sight angle plane is CEF, optical axis center institute Plane be CMN and ranging characteristic point where plane be CC2D, wherein, the optical axis CC of CCD camera1With imaging plane A ' B ' F ' E ' intersection point is C0Point, then CC0For the focal length of ccd sensor, i.e. CC0=f；The near-sighted field picture of objects ahead vehicle in image Down contour point A and imaging plane A ' B ' F ' E ' meet at G ' points, and far visual field down contour point B and imaging plane A ' B ' F ' E ' meet at H points； The ranging characteristic point P of front vehicles projects to world coordinate system XWThe point of axle is P', wherein, target source longitudinal direction perceived distance is For OP ' length.
  4. A kind of 4. ranging of front vehicles monocular vision based on track plane geometry model-driven according to claim 3 Method, it is characterised in that：In step 5, target source longitudinal direction perceived distance y calculation formula is：
    <mrow> <mi>y</mi> <mo>=</mo> <mi>h</mi> <mo>&amp;times;</mo> <mi>t</mi> <mi>a</mi> <mi>n</mi> <mo>&amp;lsqb;</mo> <mfrac> <mi>&amp;pi;</mi> <mn>2</mn> </mfrac> <mo>-</mo> <mi>&amp;theta;</mi> <mo>+</mo> <mi>arctan</mi> <mrow> <mo>(</mo> <mfrac> <mrow> <mo>&amp;lsqb;</mo> <msub> <mi>v</mi> <mn>0</mn> </msub> <mo>-</mo> <mi>v</mi> <mrow> <mo>(</mo> <msub> <mi>P</mi> <mn>0</mn> </msub> <mo>)</mo> </mrow> <mo>&amp;rsqb;</mo> <mo>&amp;times;</mo> <mi>d</mi> <mi>y</mi> </mrow> <mi>f</mi> </mfrac> <mo>)</mo> </mrow> <mo>&amp;rsqb;</mo> </mrow>
    In formula, v0For photocentre longitudinal direction image coordinate, v (P0) a longitudinal image coordinate is characterized, dy is that the longitudinal direction of unit pixel is long Degree.
  5. 5. a kind of front vehicles monocular vision based on track plane geometry model-driven according to claim 3 or 4 Distance-finding method, it is characterised in that：In step 6, the structure of objects ahead vehicle odometry error dynamic compensation model, specifically include with Lower step：
    The first step, fixed CCD vision sensors, with the inside and outside parameter mark of the scaling method progress CCD camera described in step 1 Determine and record；
    Second step, the terrain clearance in fixed target source, target source is moved every 5m along road longitudinal direction, makes it in distance CCD Change in the range of vision sensor [10m, 100m], and image of the target source in each position is recorded with CCD camera；
    3rd step, the terrain clearance in adjustment target source, makes it change in [0.2m, 1m], repeats second step；
    4th step, image obtained by the 3rd step is handled using matlab, analysis target source is in different terrain clearances and different longitudinal directions The measurement error of position.
  6. A kind of 6. ranging of front vehicles monocular vision based on track plane geometry model-driven according to claim 5 Method, it is characterised in that：In step 6, longitudinal measurement error value z calculation formula is：
    Z=118-1124x-3.133y+3522x2+34.6xy-0.0399y2-4795x3-86.22x2y-0.1845xy2+ 0.0017y3+2676x4+98.62x3y+0.1428x2y2+0.001313xy3-2.077e-5y4+8.835e-8y5-7.658e-6xy4+ 0.0004x2y3-0.1114x3y2-38.23x4y-391.1x5
    In formula, x is target source terrain clearance.
  7. A kind of 7. ranging of front vehicles monocular vision based on track plane geometry model-driven according to claim 6 Method, it is characterised in that：In step 7, longitudinal spacing Y between objects ahead vehicle and our vehicleW(P) calculation formula For：
    <mrow> <msub> <mi>Y</mi> <mi>W</mi> </msub> <mrow> <mo>(</mo> <mi>P</mi> <mo>)</mo> </mrow> <mo>=</mo> <mi>h</mi> <mo>&amp;times;</mo> <mi>t</mi> <mi>a</mi> <mi>n</mi> <mo>&amp;lsqb;</mo> <mfrac> <mi>&amp;pi;</mi> <mn>2</mn> </mfrac> <mo>-</mo> <mi>&amp;theta;</mi> <mo>+</mo> <mi>arctan</mi> <mrow> <mo>(</mo> <mfrac> <mrow> <mo>&amp;lsqb;</mo> <msub> <mi>v</mi> <mn>0</mn> </msub> <mo>-</mo> <mi>v</mi> <mrow> <mo>(</mo> <msub> <mi>P</mi> <mn>0</mn> </msub> <mo>)</mo> </mrow> <mo>&amp;rsqb;</mo> <mo>&amp;times;</mo> <mi>d</mi> <mi>y</mi> </mrow> <mi>f</mi> </mfrac> <mo>)</mo> </mrow> <mo>&amp;rsqb;</mo> <mo>-</mo> <mi>z</mi> <mo>.</mo> </mrow>
CN201710930473.5A 2017-10-09 2017-10-09 Distance measurement method based on monocular vision of front vehicle driven by lane plane geometric model Expired - Fee Related CN107796373B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710930473.5A CN107796373B (en) 2017-10-09 2017-10-09 Distance measurement method based on monocular vision of front vehicle driven by lane plane geometric model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710930473.5A CN107796373B (en) 2017-10-09 2017-10-09 Distance measurement method based on monocular vision of front vehicle driven by lane plane geometric model

Publications (2)

Publication Number Publication Date
CN107796373A true CN107796373A (en) 2018-03-13
CN107796373B CN107796373B (en) 2020-07-28

Family

ID=61532879

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710930473.5A Expired - Fee Related CN107796373B (en) 2017-10-09 2017-10-09 Distance measurement method based on monocular vision of front vehicle driven by lane plane geometric model

Country Status (1)

Country Link
CN (1) CN107796373B (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108759667A (en) * 2018-05-29 2018-11-06 福州大学 Front truck distance measuring method based on monocular vision and image segmentation under vehicle-mounted camera
CN109272536A (en) * 2018-09-21 2019-01-25 浙江工商大学 A kind of diatom vanishing point tracking based on Kalman filter
CN109752709A (en) * 2019-01-22 2019-05-14 武汉鸿瑞达信息技术有限公司 A kind of distance measurement method and device based on image
CN112365741A (en) * 2020-10-23 2021-02-12 淮阴工学院 Safety early warning method and system based on multilane vehicle distance detection
CN112686209A (en) * 2021-01-25 2021-04-20 深圳市艾为智能有限公司 Vehicle rear blind area monitoring method based on wheel identification
CN112880642A (en) * 2021-03-01 2021-06-01 苏州挚途科技有限公司 Distance measuring system and distance measuring method
CN113221739A (en) * 2021-05-12 2021-08-06 中国科学技术大学 Monocular vision-based vehicle distance measuring method

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102661733A (en) * 2012-05-28 2012-09-12 天津工业大学 Front vehicle ranging method based on monocular vision
CN202614463U (en) * 2012-05-09 2012-12-19 齐齐哈尔大学 A temperature drift calibration device of a piezoresistive pressure sensor
CN102865824A (en) * 2012-09-18 2013-01-09 北京经纬恒润科技有限公司 Method and device for calculating relative distance between vehicles
KR20160037424A (en) * 2014-09-29 2016-04-06 동명대학교산학협력단 A Novel Multi-view Face Detection Method Based on Improved Real Adaboost Algorithm
CN105574552A (en) * 2014-10-09 2016-05-11 东北大学 Vehicle ranging and collision early warning method based on monocular vision
CN106524908A (en) * 2016-10-17 2017-03-22 湖北文理学院 Measurement method for machine tool total travel space errors
CN107194045A (en) * 2017-05-08 2017-09-22 北京航空航天大学 Ripple modeling method is disturbed before a kind of refueled aircraft for air refuelling

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN202614463U (en) * 2012-05-09 2012-12-19 齐齐哈尔大学 A temperature drift calibration device of a piezoresistive pressure sensor
CN102661733A (en) * 2012-05-28 2012-09-12 天津工业大学 Front vehicle ranging method based on monocular vision
CN102865824A (en) * 2012-09-18 2013-01-09 北京经纬恒润科技有限公司 Method and device for calculating relative distance between vehicles
KR20160037424A (en) * 2014-09-29 2016-04-06 동명대학교산학협력단 A Novel Multi-view Face Detection Method Based on Improved Real Adaboost Algorithm
CN105574552A (en) * 2014-10-09 2016-05-11 东北大学 Vehicle ranging and collision early warning method based on monocular vision
CN106524908A (en) * 2016-10-17 2017-03-22 湖北文理学院 Measurement method for machine tool total travel space errors
CN107194045A (en) * 2017-05-08 2017-09-22 北京航空航天大学 Ripple modeling method is disturbed before a kind of refueled aircraft for air refuelling

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
杨炜，魏朗: "基于单目视觉的纵向车间距检测研究", 《自动化测试技术》 *
段翔，杨炜: "前方车辆图像识别与纵向安全域控制方法研究", 《交通运输》 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108759667A (en) * 2018-05-29 2018-11-06 福州大学 Front truck distance measuring method based on monocular vision and image segmentation under vehicle-mounted camera
CN109272536A (en) * 2018-09-21 2019-01-25 浙江工商大学 A kind of diatom vanishing point tracking based on Kalman filter
CN109272536B (en) * 2018-09-21 2021-11-09 浙江工商大学 Lane line vanishing point tracking method based on Kalman filtering
CN109752709A (en) * 2019-01-22 2019-05-14 武汉鸿瑞达信息技术有限公司 A kind of distance measurement method and device based on image
CN112365741A (en) * 2020-10-23 2021-02-12 淮阴工学院 Safety early warning method and system based on multilane vehicle distance detection
CN112365741B (en) * 2020-10-23 2021-09-28 淮阴工学院 Safety early warning method and system based on multilane vehicle distance detection
CN112686209A (en) * 2021-01-25 2021-04-20 深圳市艾为智能有限公司 Vehicle rear blind area monitoring method based on wheel identification
CN112880642A (en) * 2021-03-01 2021-06-01 苏州挚途科技有限公司 Distance measuring system and distance measuring method
CN113221739A (en) * 2021-05-12 2021-08-06 中国科学技术大学 Monocular vision-based vehicle distance measuring method

Also Published As

Publication number Publication date
CN107796373B (en) 2020-07-28

Similar Documents

Publication Publication Date Title
CN107796373A (en) A kind of distance-finding method of the front vehicles monocular vision based on track plane geometry model-driven
CN110531376B (en) Obstacle detection and tracking method for port unmanned vehicle
CN107341453B (en) Lane line extraction method and device
US20200041284A1 (en) Map road marking and road quality collecting apparatus and method based on adas system
CN103559791B (en) A kind of vehicle checking method merging radar and ccd video camera signal
CN108596058A (en) Running disorder object distance measuring method based on computer vision
CN105892471A (en) Automatic automobile driving method and device
CN111563469A (en) Method and device for identifying irregular parking behaviors
CN113850102B (en) Vehicle-mounted vision detection method and system based on millimeter wave radar assistance
CN110745140A (en) Vehicle lane change early warning method based on continuous image constraint pose estimation
CN104573646A (en) Detection method and system, based on laser radar and binocular camera, for pedestrian in front of vehicle
CN107192994A (en) Multi-line laser radar mass cloud data is quickly effectively extracted and vehicle, lane line characteristic recognition method
CN104700414A (en) Rapid distance-measuring method for pedestrian on road ahead on the basis of on-board binocular camera
CN106096525A (en) A kind of compound lane recognition system and method
CN114359181B (en) Intelligent traffic target fusion detection method and system based on image and point cloud
CN106295459A (en) Based on machine vision and the vehicle detection of cascade classifier and method for early warning
CN108645375B (en) Rapid vehicle distance measurement optimization method for vehicle-mounted binocular system
CN104751119A (en) Rapid detecting and tracking method for pedestrians based on information fusion
CN107985189A (en) Towards driver&#39;s lane change Deep Early Warning method under scorch environment
CN111967360A (en) Target vehicle attitude detection method based on wheels
CN112735164B (en) Test data construction method and test method
CN106803262A (en) The method that car speed is independently resolved using binocular vision
CN115857040A (en) Dynamic visual detection device and method for foreign matters on locomotive roof
Lu et al. Forward vehicle collision warning based on quick camera calibration
Liu et al. Research on security of key algorithms in intelligent driving system

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
GR01 Patent grant
GR01 Patent grant
TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20210129

Address after: 201600 no.216 GANGYE Road, Xiaokunshan Town, Songjiang District, Shanghai

Patentee after: Shanghai Yingdong Technology Development Co.,Ltd.

Address before: 710064 No. 33, South Second Ring Road, Shaanxi, Xi'an

Patentee before: CHANG'AN University

CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20200728

Termination date: 20211009