CN109334563A - A kind of anticollision method for early warning based on road ahead pedestrian and bicyclist - Google Patents
A kind of anticollision method for early warning based on road ahead pedestrian and bicyclist Download PDFInfo
- Publication number
- CN109334563A CN109334563A CN201811013119.7A CN201811013119A CN109334563A CN 109334563 A CN109334563 A CN 109334563A CN 201811013119 A CN201811013119 A CN 201811013119A CN 109334563 A CN109334563 A CN 109334563A
- Authority
- CN
- China
- Prior art keywords
- early warning
- bicyclist
- target
- ipm
- pedestrian
- 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
Links
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60Q—ARRANGEMENT OF SIGNALLING OR LIGHTING DEVICES, THE MOUNTING OR SUPPORTING THEREOF OR CIRCUITS THEREFOR, FOR VEHICLES IN GENERAL
- B60Q9/00—Arrangement or adaptation of signal devices not provided for in one of main groups B60Q1/00 - B60Q7/00, e.g. haptic signalling
- B60Q9/008—Arrangement or adaptation of signal devices not provided for in one of main groups B60Q1/00 - B60Q7/00, e.g. haptic signalling for anti-collision purposes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
Landscapes
- Engineering & Computer Science (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Mechanical Engineering (AREA)
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a kind of anticollision method for early warning based on road ahead pedestrian and bicyclist belongs to driving assistance system field, including environment sensing, information are interpreted and dbjective state judge in terms of three.The training set of collection is sent into YOLO-R network and is trained by the present invention, carries out target detection and classification, realizes multiple target tracking using Kalman filter;Inverse perspective mapping is carried out to image, obtains IPM image, and fits the relation curve of original image Yu IPM image pixel coordinates by data regression modeling, then distance is estimated by the linear relationship of IPM image pixel coordinates and world coordinates;According to from vehicle speed and braking distance; it determines early warning active region, target then is judged with the presence or absence of danger, if dangerous timely prompting driver using fuzzy warning model algorithm to the target in active region; to effectively reduce the generation of accident, the safety of pedestrian and bicyclist are protected.
Description
Technical field
The invention belongs to field of automotive active safety, are related to the knowledge of image procossing and collision early warning system, tool
Body is related to a kind of anticollision method for early warning based on road ahead pedestrian and bicyclist.
Background technique
Under this typical mixed traffic environment in China, pedestrian, bicyclist, vehicle etc. are all road traffic participants,
Wherein pedestrian, bicyclist are disadvantaged group, they are exposed to outside, and without safeguard procedures, when traffic accident occurs, the person is pacified
It is more difficult to ensure entirely, therefore protects the safety of pedestrian and bicyclist particularly important.
Traditional pedestrian and bicyclist's detection algorithm is by manually extracting the features such as HOG, Haar, LUV, training classification
Device completes target detection.This method can obtain preferable detection effect, but the feature of engineer under given conditions
It is not so good in detection effect of the rather dark, target carriage change obviously and under complex scene.In contrast, deep learning can
To extract feature in the picture by convolutional layer, detection effect is substantially better than conventional machines learning method.As hardware calculates
The enhancing of ability and the foundation of a large amount of training datasets, deep learning are flourished.In terms of target detection, from
RCNN, Fast-RCNN, Faster RCNN to YOLO, SSD, YOLOv2, the speed and accuracy rate of target detection have very big
Breakthrough.
Chinese patent (CN1O2765365A) discloses pedestrian detection method based on machine vision and pedestrian anti-collision is pre-
Alert system has been obscured the personal feature between pedestrian, it is poor to be reduced individuality using the pedestrian on pedestrian's detection of classifier road
The different influence to testing result judges a possibility that accident occurs by anti-collision early warning system, but its detection accuracy is low, has ignored
The safety of bicyclist.Chinese patent (CN204870868U) discloses vehicle anticollision based on multisensor and pedestrian protecting is pre-
Alert system realizes the early warning to vehicle front and front vehicle, barrier or pedestrian with laser ranging and ultrasonic distance measurement,
This method warning index is single, and the reliability of early warning system is poor.
Summary of the invention
In view of the above-mentioned problems, the present invention proposes a kind of anticollision method for early warning based on road ahead pedestrian and bicyclist,
It constructs and is melted based on multi information according to the position of comprehensive analysis target, transverse and longitudinal distance, from the speed and collision time TTC of vehicle
The collision early warning system of conjunction judges the degree of danger of target, can effectively detect to be exposed to the pedestrian of outside and rides
Person, according to the setting of early warning active region, only target enters active region and carries out dbjective state judgement again, protection pedestrian and rides
The safety of passerby.
Specific technical solution of the present invention is as follows:
A kind of anticollision method for early warning based on road ahead pedestrian and bicyclist, comprising the following steps:
S1 builds the collision early warning system based on road ahead pedestrian and bicyclist;
S2, off-line training carry out model training using the improved network YOLO-R of YOLOv2;
One frame image is input in trained YOLO-R network by S3, is detected pedestrian and bicyclist and is classified,
Realize multiple target tracking;
S4 calculates the transverse and longitudinal distance of vehicle and objects ahead;
S5 sets early warning active region, using transverse and longitudinal Distance Judgment target whether in active region according to from vehicle speed
Domain calculates warning index to the target in early warning active region, and index is substituted into early warning system, determines warning grade.
Further, the collision early warning system overall structure based on road ahead pedestrian and bicyclist includes environment sense
Know, information is interpreted and dbjective state judges three modules, environment sensing includes object detecting and tracking, vanishing Point Detection Method and from vehicle
Speed data collection;Object detecting and tracking, vanishing Point Detection Method are obtained relevant information, and are led to by forward sight camera, video frequency collection card
It crosses video frequency collection card and image information is passed into PC machine, speed is acquired by GPS module, and information is interpreted and dbjective state judgment module
It is all realized in PC machine, early warning result is finally shown by software interface.
Further, the build process of the YOLO-R network are as follows: on the basis of YOLOv2 network structure, cluster is chosen
Anchor boxes removes passthrough layers, increases residual error network and constitutes YOLO-R network.
Further, model training process is divided into propagated forward and backpropagation two parts in the S2, specifically: it will train
Sample carries out propagated forward calculating, finally exports relative position, the confidence level comprising target and the class probability letter of candidate frame
Breath;Using back-propagation algorithm and small lot gradient descent method, each layer weight of network is constantly updated, the value of cost function is reduced;
Constantly repeat the above process, when all sample trainings finish, i.e. completion an iteration.
Further, algorithm includes off-line procedure and in line process two parts, off-line procedure in the S4 specifically: is established former
The regression model of beginning image and IPM image pixel coordinates;In line process specifically: obtain the pixel at target rectangle frame bottom edge midpoint
Coordinate;Camera pitch angle is calculated by road Vanishing Point Detecting Algorithm again, according to pitch angle changing value Δ θ, corrects target picture
Plain coordinate;Then the corresponding IPM image pixel coordinates of original image pixels coordinate are found out by regression model;Finally according to IPM
The linear relationship of image pixel coordinates and world coordinates estimates the horizontal and vertical distance of target.
Further, the fore-and-aft distance X=(hIPM-v′)·σ2+Xmin, wherein hIPMIndicate the height of image, v ' expression
IPM image pixel coordinates, σ2Indicate actual physics distance value represented by unit pixel, X in vertical directionminFor IPM image institute
Corresponding practical front minimum range.
Further, the lateral distance Y=(u '-wIPM/2)·σ1, wherein wIPMIndicate the width of image, u ' expression IPM
Image pixel coordinates, σ1Indicate actual physics distance value represented by unit pixel in horizontal direction.
Further, using transverse and longitudinal Distance Judgment target whether active region formula are as follows:
dfminFor vehicle-to-target minimum distance, dfmaxFor the farthest fore-and-aft distance of vehicle-to-target.
Further, it determines warning grade using fuzzy warning model algorithm: determining warning grade, warning index set, pre-
Alert weight sets, then determine degree of membership of the weight sets to warning grade of each index in warning index, so that it is determined that fuzzy evaluation square
Battle array, determines the warning grade state current as target.
The invention has the benefit that
1, the present invention detects pedestrian and bicyclist using the improved network YOLO-R of YOLOv2, not only can be automatic
The higher level feature of characterization target is extracted, and by the Fusion Features between level, improves pedestrian and bicyclist's detection
Performance.
2, the present invention is based on the monocular ranging of the dynamic inverse perspective mapping of road disappearance point estimation and data regression modeling calculations
Method reduces influence of the camera pitch angle variation to range accuracy, improves the precision of ranging.
3, the present invention carries out anticollision early warning using fuzzy synthetic appraisement method, using multiple warning indexs to warning grade
Degree of membership, obtain fuzzy evaluating matrix, then overall merit determines final result, assigning degrees of hazard, and driver is reminded to subtract
The generation of few accident, guarantees the safety of pedestrian and bicyclist.
Detailed description of the invention
Fig. 1 is a kind of overall construction drawing of the anticollision method for early warning based on road ahead pedestrian and bicyclist of the present invention;
Fig. 2 is that the present invention is based on the structure charts of the YOLO-R network of YOLOv2 network improvement;
Fig. 3 is target detection of the present invention and track algorithm figure;
Fig. 4 is that the present invention is based on the monocular location algorithm flow charts of inverse perspective mapping and data regression modeling;
Fig. 5 is the regression model of original image of the present invention and IPM image pixel ordinate;
Fig. 6 is the regression model of original image pixels ordinate v and Δ u of the present invention;
Fig. 7 is invention activation area schematic.
Specific embodiment
With reference to the accompanying drawing and the present invention is specifically described in specific embodiment.
Overall construction drawing of the invention is as shown in Figure 1, include environment sensing, information is interpreted and dbjective state judges three sides
Face first passes through camera and obtains outside vehicle environmental information, including road end point position, objects ahead position and tracking mesh
Cursor position, while being obtained from vehicle speed using GPS module, then by target detection tracking result, disappearance dot position information with
And camera calibration result calculates the transverse and longitudinal distance of target, further according to from vehicle speed, sets early warning active region, and utilize
Whether transverse and longitudinal Distance Judgment target, finally for the target in early warning active region, calculates pre- in early warning active region
Alert index, and bring these indexs into fuzzy warning model algorithm, determine warning grade.
S1 builds a kind of collision early warning system based on road ahead pedestrian and bicyclist, and overall structure includes environment
Perception, information are interpreted and dbjective state judge three big modules, environment sensing include object detecting and tracking, vanishing Point Detection Method and oneself
Vehicle speed data collection, object detecting and tracking, vanishing Point Detection Method obtain relevant information by forward sight camera, and pass through video acquisition
Image information is passed to PC machine by card, completes Image Acquisition, and by the power supply power supply of 12V, speed acquires camera by GPS module,
Information is interpreted and dbjective state judgment module is all realized in PC machine, finally shows early warning as a result, software architecture by software interface
Be under visual studio2015 development platform combine CUDA8.0 framework, deep learning accelerate library cuDNN and
What the library OpenCV2.4.10 was realized.
S2, off-line training are trained using the improved network YOLO-R of YOLOv2, and the training process of entire model is divided into
Propagated forward and backpropagation two parts;
The target rectangle frame marked in sample set is carried out to contain cluster using the method that k-means is clustered, is determined
The initialization specification and quantity of anchor boxes is not suitable for training detection pedestrian since to be related to type too many for anchor parameter
With the model of bicyclist, therefore the present invention clusters again in homemade pedestrian and bicyclist's sample database, obtains anchor number;And
On the basis of YOLOv2 network structure, remove passthrough layers, increases residual network (ResNet, Residual
Network YOLO-R network) is constituted;As shown in Figure 2.
Model training process is as follows:
(1) training sample is upset into sequence, storage in a vessel, and has used a variety of data extending methods, wraps
Include rotation image, adjust tone, saturation degree etc..Sample is divided into much in small batches, every time instructs a collection of sample feeding network
Practice, when GPU Out of Memory, it is possible to reduce the quantity of a collection of sample.
(2) these samples and label information are admitted to network, carry out propagated forward calculating, finally export the phase of candidate frame
To position, the confidence level comprising target and class probability information.The picture of input network is normalized to n × n-pixel, then
It is divided into a × a cell, each cell places b anchor boxes, and each target in sample presses its central point
Position is assigned in corresponding cell, and according to the IOU of anchor box and ground truth, selects IOU maximum
Anchor box is responsible for the prediction of the target.Sample exports the predicted value of each candidate frame: (t after YOLO-R networkx,ty,
tw,th,t0,p)。
(3) back-propagation algorithm and small lot gradient descent method are utilized, each layer weight of network is constantly updated, reduces cost letter
Several values.
(4) it constantly repeats the above process, when all sample trainings finish, i.e. completion an iteration.It is every to pass through 10 iteration
With regard to the size of adjusting training sample, so that the network trained can preferably predict various sizes of picture.Work as the number of iterations
When reaching maximum value or training error and no longer reducing for a long time, deconditioning.
S3, as shown in figure 3, object detecting and tracking: a frame image being input in trained YOLO-R network, is detected
Pedestrian and bicyclist and classify out, multiple target tracking is realized using Kalman filter;Detailed process is as follows:
(1) image and label information being input in trained network model, image is divided into a × a cell,
Each cell predicts b candidate frame, predicts a × a × b candidate frame altogether;Then pass through network forwards algorithms, predict every
The relative position of a candidate frame: tx、ty、tw、th, confidence level t0And the posterior probability p of generic.
(2) to the t of predictionx、ty、tw、thAnd t0Mapping transformation is done, is obtained and the closer window conduct of anchor box
Detection block.
(3) by the threshold value T (the present embodiment T=c) of setting confidence level, the lesser detection block of possibility is removed, is specifically done
Method is: σ (t0) being multiplied with max (p), obtains the confidence level that detection block belongs to certain classification;If result is greater than threshold value T, retain
The detection block, otherwise removes.
(4) non-maxima suppression processing is carried out to each classification respectively, removes redundancy window, the specific steps are as follows: to every
The detection block of a classification presses the big minispread of confidence level;The highest detection block of confidence level is found out, then successively calculates IOU with other frames
(Intersection Over Union) deletes this frame when IOU is greater than threshold values d, otherwise retains this frame;From untreated inspection
It is highest that confidence level is selected in survey frame, is repeated the above steps, until all windows are disposed;Export the position of the detection block left
It sets, classification and confidence level.
(5) further fusion treatment is done using the result that matching algorithm exports detection algorithm, completes pedestrian and bicyclist
Classification;Multiple target is tracked using Kalman filter.
S4 calculates transverse and longitudinal distance by the monocular location algorithm of inverse perspective mapping and data regression modeling, entire to calculate
Method includes off-line procedure and in line process two parts, as shown in Figure 4;
Off-line procedure: first obtaining a vehicle-mounted image, then passes through camera calibration and inverse perspective mapping, obtains IPM
(Image Perspective Mapping) image, and the regression model of original image Yu IPM image pixel coordinates is established, such as
Fig. 5.
In line process: reading in video image in real time, obtain target rectangle frame bottom edge midpoint by detection and tracking algorithm
Pixel coordinate;Camera pitch angle is calculated by road Vanishing Point Detecting Algorithm again, according to pitch angle changing value Δ θ, corrects mesh
Mark pixel coordinate;Then the corresponding IPM image pixel coordinates of original image pixels coordinate are found out by regression model;Further root
According to the linear relationship of IPM image pixel coordinates and world coordinates, the horizontal and vertical distance of target is estimated.
By the discribed mapping curve of observation data, the relationship of v Yu v ' are fitted, then fore-and-aft distance X is found out by formula (1).
X=(hIPM-v′)·σ2+Xmin (1)
Wherein, hIPMIndicate the height of image, v ' expression IPM image pixel coordinates, σ2Indicate unit pixel in vertical direction
Represented actual physics distance value, XminFor practical front minimum range corresponding to IPM image;
From the regression model of Fig. 6 original image pixels ordinate v and Δ u can be seen that original image pixels ordinate ν with
There are apparent linear relationships by Δ u, on the original image that wherein Δ u indicates for unit pixel in horizontal direction in IPM image
Pixel value;And ROI region is the bilateral symmetry using the straight line where camera principal point as symmetry axis, i.e. IPM figure in inverse perspective mapping
Abscissa pixel value at inconocenter line corresponds to the μ of original image0(camera internal reference number).According to this condition and fitting
Linear equation out can find out the relationship of original image pixels coordinate (μ, ν) Yu IPM image pixel abscissa u ', recycle formula
(2) the lateral distance Y of target is found out.
Y=(u '-wIPM/2)·σ1 (2)
Wherein, wIPMIndicate the width of image, u ' expression IPM image pixel coordinates, σ1Indicate unit pixel in horizontal direction
Represented actual physics distance value;
S5 sets early warning active region, using transverse and longitudinal Distance Judgment target whether in active region according to from vehicle speed
Domain calculates warning index to the target in early warning active region, and brings these indexs into early warning system, determines early warning etc.
Grade;
The shape of early warning active region is set as trapezoidal by the setting of early warning active region, the present invention, as shown in Figure 7.In Fig. 7
Active region in, calculate vehicle-to-target minimum distance dfminAnd farthest fore-and-aft distance dfmax。
The horizontal boundary of active region and speed, the speed of target are all related;Assuming that moving with uniform velocity from vehicle, calculating is come from
Vehicle travels dfminAnd dfmaxRequired time is t (dfmin) and t (dfmax)。
The speed u of vehicle can be obtained by vehicle-mounted GPS module, find out the boundary of active region;Determine activation
Behind region, judge target whether in active region by following formula.
Wherein, X and Y indicates target and the horizontal and vertical distance from vehicle;
If target, in active region, early warning system carries out early warning using fuzzy algorithmic approach, and detailed process is as follows:
(1) warning grade, warning index collection, early warning weight sets are determined;Warning grade is divided into three grades, is peace respectively
Entirely, pay attention to, be dangerous, warning index integrate for E=the position (E1) of target, lateral distance (E2), from vehicle speed (E3), longitudinally away from
From/TTC (E4) }, weight sets is determined using Fuzzy AHP (FAHP);FAHP method determines the concrete operations packet of weight sets
It includes: constructing Fuzzy Complementary Judgment Matrices S and seek weight vectors W;
1. Fuzzy Complementary Judgment Matrices S can be obtained by following formula:
Wherein, S (i), S (j) respectively indicate the relative importance of index i, j, i, and j ∈ (1 ..., m), m indicate warning index
Number.
2. it is as follows to obtain weight vectors W process:
First find out the sum of the every row of matrix S:
Wherein, riBe the i-th row of matrix S and.
Obtain vector R1=(r1..., rk..., rm)
Then capable transformation is done to vector R1, obtains Fuzzy consistent matrix R2:
The expression formula of each element in Fuzzy consistent matrix R2 are as follows:
To other elements summation of the every row in R2 in addition to the elements in a main diagonal:
In Fuzzy consistent matrix R2 in addition to leading diagonal other elements sum are as follows:
Wherein, liIndex i is indicated to the significance level of index i-1, to liNormalization operation can find out each warning grade
Weight;Weight wiIt is represented by shown in formula (9), shown in final weight result W such as formula (10).
W=(1/8,5/24,3/8,7/24) (10)
(2) Comment gathers of each index are obtained according to fuzz method, the position Comment gathers of target are { close, holding, separate }, horizontal
It is { short, in, long } to the Comment gathers of distance, is { low speed, middling speed, high speed }, fore-and-aft distance/TTC from the Comment gathers of vehicle speed
Comment gathers are { short/small, in, long/big }.
(3) after establishing warning index collection and its Comment gathers, it is necessary to determine the Comment gathers of each index in warning index to pre-
The degree of membership of alert grade, so that it is determined that fuzzy evaluating matrix.Subordinating degree function uses discrete quantized versions, according to expertise
Method obtains degree of membership table shown in table 1, by tabling look-up, obtains fuzzy evaluating matrix R.
1 degree of membership table of table
(4) it selects M (,+) operator to synthesize weight vector and fuzzy evaluating matrix, obtains dbjective state to early warning
The degree of membership vector of gradeAccording to maximum membership grade principle, S is selected1Warning grade corresponding to middle maximum value is made
For the current state of target.
Specific embodiment described above is used to illustrate the present invention, but the present invention is not limited only to this, any ability
Field technique personnel are not departing from change and modification in spirit and scope of the invention, should all be included in protection scope of the present invention,
The scope of the present invention should be defined by the scope defined by the claims..
Claims (9)
1. a kind of anticollision method for early warning based on road ahead pedestrian and bicyclist, which comprises the following steps:
S1 builds the collision early warning system based on road ahead pedestrian and bicyclist;
S2, off-line training carry out model training using the improved network YOLO-R of YOLOv2;
One frame image is input in trained YOLO-R network by S3, is detected pedestrian and bicyclist and is classified, and is realized
Multiple target tracking;
S4 calculates the transverse and longitudinal distance of vehicle and objects ahead;
Whether S5 sets early warning active region according to from vehicle speed, right using transverse and longitudinal Distance Judgment target in active region
Target in early warning active region calculates warning index, and index is substituted into early warning system, determines warning grade.
2. a kind of anticollision method for early warning based on road ahead pedestrian and bicyclist according to claim 1, feature
It is, the collision early warning system overall structure based on road ahead pedestrian and bicyclist includes environment sensing, information solution
It reads and dbjective state judges three modules, environment sensing includes object detecting and tracking, vanishing Point Detection Method and from vehicle speed data collection;
Object detecting and tracking, vanishing Point Detection Method obtain relevant information by forward sight camera, and are believed image by video frequency collection card
Breath passes to computer, and vehicle speed information is acquired by GPS module, and information is interpreted and dbjective state judgment module is all real in a computer
It is existing, early warning result is finally shown by software interface.
3. a kind of anticollision method for early warning based on road ahead pedestrian and bicyclist according to claim 1, feature
It is, the build process of the YOLO-R network are as follows: on the basis of YOLOv2 network structure, cluster chooses anchor
Boxes removes passthrough layers, increases residual error network and constitutes YOLO-R network.
4. a kind of anticollision method for early warning based on road ahead pedestrian and bicyclist according to claim 1, special
Sign is that model training process is divided into propagated forward and backpropagation two parts in the S2, specifically: training sample is carried out
Propagated forward calculates, and finally exports relative position, the confidence level comprising target and the class probability information of candidate frame;Using anti-
To propagation algorithm and small lot gradient descent method, each layer weight of network is constantly updated, the value of cost function is reduced;Constantly repeat
Process is stated, when all sample trainings finish, i.e. completion an iteration.
5. a kind of anticollision method for early warning based on road ahead pedestrian and bicyclist according to claim 1, special
Sign is, algorithm includes off-line procedure and in line process two parts, off-line procedure in the S4 specifically: establish original image with
The regression model of IPM image pixel coordinates;In line process specifically: obtain the pixel coordinate at target rectangle frame bottom edge midpoint;Again
Camera pitch angle is calculated by road Vanishing Point Detecting Algorithm, according to pitch angle changing value Δ θ, corrects object pixel coordinate;
Then the corresponding IPM image pixel coordinates of original image pixels coordinate are found out by regression model;Finally according to IPM image pixel
The linear relationship of coordinate and world coordinates estimates the horizontal and vertical distance of target.
6. a kind of anticollision method for early warning based on road ahead pedestrian and bicyclist according to claim 5, special
Sign is, the fore-and-aft distance X=(hIPM-v′)·σ2+Xmin, wherein hIPMIndicate the height of image, v ' expression IPM image slices
Plain coordinate, σ2Indicate actual physics distance value represented by unit pixel, X in vertical directionminFor reality corresponding to IPM image
Minimum range in front of border.
7. a kind of anticollision method for early warning based on road ahead pedestrian and bicyclist according to claim 5, special
Sign is, the lateral distance Y=(u '-wIPM/2)·σ1, wherein wIPMIndicate the width of image, u ' expression IPM image pixel
Coordinate, σ1Indicate actual physics distance value represented by unit pixel in horizontal direction.
8. a kind of anticollision method for early warning based on road ahead pedestrian and bicyclist according to claims 6 or 7,
Be characterized in that, using transverse and longitudinal Distance Judgment target whether active region formula are as follows:dfminFor
Vehicle-to-target minimum distance, dfmaxFor the farthest fore-and-aft distance of vehicle-to-target.
9. a kind of anticollision method for early warning based on road ahead pedestrian and bicyclist according to claim 1, special
Sign is, determines warning grade using fuzzy warning model algorithm: determine warning grade, warning index collection, early warning weight sets, then
Degree of membership of the Comment gathers to warning grade of each index in warning index is determined, so that it is determined that fuzzy evaluating matrix, determines early warning
The grade state current as target.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811013119.7A CN109334563B (en) | 2018-08-31 | 2018-08-31 | Anti-collision early warning method based on pedestrians and riders in front of road |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811013119.7A CN109334563B (en) | 2018-08-31 | 2018-08-31 | Anti-collision early warning method based on pedestrians and riders in front of road |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109334563A true CN109334563A (en) | 2019-02-15 |
CN109334563B CN109334563B (en) | 2021-06-22 |
Family
ID=65296814
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811013119.7A Active CN109334563B (en) | 2018-08-31 | 2018-08-31 | Anti-collision early warning method based on pedestrians and riders in front of road |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109334563B (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109919074A (en) * | 2019-03-01 | 2019-06-21 | 中国科学院合肥物质科学研究院 | A kind of the vehicle cognitive method and device of view-based access control model cognition technology |
CN110929606A (en) * | 2019-11-11 | 2020-03-27 | 浙江鸿泉车联网有限公司 | Vehicle blind area pedestrian monitoring method and device |
CN111783700A (en) * | 2020-07-06 | 2020-10-16 | 中国交通通信信息中心 | Automatic recognition early warning method and system for road foreign matters |
CN111860097A (en) * | 2020-04-16 | 2020-10-30 | 西安电子科技大学 | Abnormal behavior detection method based on fuzzy theory |
CN111921199A (en) * | 2020-08-25 | 2020-11-13 | 腾讯科技(深圳)有限公司 | Virtual object state detection method, device, terminal and storage medium |
CN112249021A (en) * | 2020-10-14 | 2021-01-22 | 广州汽车集团股份有限公司 | Road pedestrian collision risk prediction method and system |
CN112349144A (en) * | 2020-11-10 | 2021-02-09 | 中科海微(北京)科技有限公司 | Monocular vision-based vehicle collision early warning method and system |
CN113031580A (en) * | 2019-12-25 | 2021-06-25 | 奥迪股份公司 | Remote assistance system, control device, corresponding server, vehicle and method |
CN113192331A (en) * | 2021-04-26 | 2021-07-30 | 吉林大学 | Intelligent early warning system and early warning method for riding safety in internet environment |
CN113449589A (en) * | 2021-05-16 | 2021-09-28 | 桂林电子科技大学 | Method for calculating driving strategy of unmanned automobile in urban traffic scene |
CN114267181A (en) * | 2022-03-03 | 2022-04-01 | 深圳市旗扬特种装备技术工程有限公司 | Pedestrian early warning method and device for urban road intersection and electronic equipment |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080172177A1 (en) * | 2005-07-19 | 2008-07-17 | Toyota Technical Center Usa, Inc. | Crash prediction network with visual input for vehicle |
CN102765365A (en) * | 2011-05-06 | 2012-11-07 | 香港生产力促进局 | Pedestrian detection method based on machine vision and pedestrian anti-collision warning system based on machine vision |
CN106156725A (en) * | 2016-06-16 | 2016-11-23 | 江苏大学 | A kind of method of work of the identification early warning system of pedestrian based on vehicle front and cyclist |
US9760806B1 (en) * | 2016-05-11 | 2017-09-12 | TCL Research America Inc. | Method and system for vision-centric deep-learning-based road situation analysis |
CN107862287A (en) * | 2017-11-08 | 2018-03-30 | 吉林大学 | A kind of front zonule object identification and vehicle early warning method |
CN107972662A (en) * | 2017-10-16 | 2018-05-01 | 华南理工大学 | To anti-collision warning method before a kind of vehicle based on deep learning |
-
2018
- 2018-08-31 CN CN201811013119.7A patent/CN109334563B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080172177A1 (en) * | 2005-07-19 | 2008-07-17 | Toyota Technical Center Usa, Inc. | Crash prediction network with visual input for vehicle |
CN102765365A (en) * | 2011-05-06 | 2012-11-07 | 香港生产力促进局 | Pedestrian detection method based on machine vision and pedestrian anti-collision warning system based on machine vision |
US9760806B1 (en) * | 2016-05-11 | 2017-09-12 | TCL Research America Inc. | Method and system for vision-centric deep-learning-based road situation analysis |
CN107368890A (en) * | 2016-05-11 | 2017-11-21 | Tcl集团股份有限公司 | A kind of road condition analyzing method and system based on deep learning centered on vision |
CN106156725A (en) * | 2016-06-16 | 2016-11-23 | 江苏大学 | A kind of method of work of the identification early warning system of pedestrian based on vehicle front and cyclist |
CN107972662A (en) * | 2017-10-16 | 2018-05-01 | 华南理工大学 | To anti-collision warning method before a kind of vehicle based on deep learning |
CN107862287A (en) * | 2017-11-08 | 2018-03-30 | 吉林大学 | A kind of front zonule object identification and vehicle early warning method |
Non-Patent Citations (4)
Title |
---|
WENBO LAN等: "Pedestrian Detection Based on YOLO Network Model", 《PROCEEDINGS OF 2018 IEEE,INTERNATIONAL CONFERENCE ON MECHATRONICS AND AUTOMATION》 * |
刘军等: "基于OpenCV的前方车辆检测和前撞预警", 《汽车技术》 * |
刘军等: "基于单目视觉车辆姿态角估计和逆透视变换的车距测量", 《农业工程学报》 * |
高宗等: "基于YOLO网络的行人检测方法", 《计算机工程》 * |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109919074B (en) * | 2019-03-01 | 2022-11-04 | 中国科学院合肥物质科学研究院 | Vehicle sensing method and device based on visual sensing technology |
CN109919074A (en) * | 2019-03-01 | 2019-06-21 | 中国科学院合肥物质科学研究院 | A kind of the vehicle cognitive method and device of view-based access control model cognition technology |
CN110929606A (en) * | 2019-11-11 | 2020-03-27 | 浙江鸿泉车联网有限公司 | Vehicle blind area pedestrian monitoring method and device |
CN113031580A (en) * | 2019-12-25 | 2021-06-25 | 奥迪股份公司 | Remote assistance system, control device, corresponding server, vehicle and method |
CN111860097A (en) * | 2020-04-16 | 2020-10-30 | 西安电子科技大学 | Abnormal behavior detection method based on fuzzy theory |
CN111860097B (en) * | 2020-04-16 | 2024-02-20 | 西安电子科技大学 | Abnormal behavior detection method based on fuzzy theory |
CN111783700A (en) * | 2020-07-06 | 2020-10-16 | 中国交通通信信息中心 | Automatic recognition early warning method and system for road foreign matters |
CN111783700B (en) * | 2020-07-06 | 2023-11-24 | 中国交通通信信息中心 | Automatic recognition and early warning method and system for pavement foreign matters |
CN111921199A (en) * | 2020-08-25 | 2020-11-13 | 腾讯科技(深圳)有限公司 | Virtual object state detection method, device, terminal and storage medium |
CN111921199B (en) * | 2020-08-25 | 2023-09-26 | 腾讯科技(深圳)有限公司 | Method, device, terminal and storage medium for detecting state of virtual object |
CN112249021B (en) * | 2020-10-14 | 2022-06-24 | 广州汽车集团股份有限公司 | Road pedestrian collision risk prediction method and system |
CN112249021A (en) * | 2020-10-14 | 2021-01-22 | 广州汽车集团股份有限公司 | Road pedestrian collision risk prediction method and system |
CN112349144A (en) * | 2020-11-10 | 2021-02-09 | 中科海微(北京)科技有限公司 | Monocular vision-based vehicle collision early warning method and system |
CN113192331A (en) * | 2021-04-26 | 2021-07-30 | 吉林大学 | Intelligent early warning system and early warning method for riding safety in internet environment |
CN113449589A (en) * | 2021-05-16 | 2021-09-28 | 桂林电子科技大学 | Method for calculating driving strategy of unmanned automobile in urban traffic scene |
CN114267181A (en) * | 2022-03-03 | 2022-04-01 | 深圳市旗扬特种装备技术工程有限公司 | Pedestrian early warning method and device for urban road intersection and electronic equipment |
Also Published As
Publication number | Publication date |
---|---|
CN109334563B (en) | 2021-06-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109334563A (en) | A kind of anticollision method for early warning based on road ahead pedestrian and bicyclist | |
US11783230B2 (en) | Automatic generation of ground truth data for training or retraining machine learning models | |
KR102109941B1 (en) | Method and Apparatus for Vehicle Detection Using Lidar Sensor and Camera | |
Keller et al. | The benefits of dense stereo for pedestrian detection | |
US7660436B2 (en) | Stereo-vision based imminent collision detection | |
CN107463890B (en) | A kind of Foregut fermenters and tracking based on monocular forward sight camera | |
CN110163904A (en) | Object marking method, control method for movement, device, equipment and storage medium | |
CN111932580A (en) | Road 3D vehicle tracking method and system based on Kalman filtering and Hungary algorithm | |
CN106503653A (en) | Area marking method, device and electronic equipment | |
CN109460709A (en) | The method of RTG dysopia analyte detection based on the fusion of RGB and D information | |
Guo et al. | A multimodal ADAS system for unmarked urban scenarios based on road context understanding | |
CN108537197A (en) | A kind of lane detection prior-warning device and method for early warning based on deep learning | |
US10576974B2 (en) | Multiple-parts based vehicle detection integrated with lane detection for improved computational efficiency and robustness | |
Kuang et al. | Feature selection based on tensor decomposition and object proposal for night-time multiclass vehicle detection | |
EP2960858A1 (en) | Sensor system for determining distance information based on stereoscopic images | |
Fan et al. | Real-time object detection for lidar based on ls-r-yolov4 neural network | |
US10984263B2 (en) | Detection and validation of objects from sequential images of a camera by using homographies | |
CN109919074A (en) | A kind of the vehicle cognitive method and device of view-based access control model cognition technology | |
Liu et al. | Vehicle detection and ranging using two different focal length cameras | |
Amanatiadis et al. | ViPED: On-road vehicle passenger detection for autonomous vehicles | |
CN116601667A (en) | System and method for 3D object detection and tracking with monocular surveillance camera | |
Xia et al. | Vehicles overtaking detection using RGB-D data | |
Dewangan et al. | Towards the design of vision-based intelligent vehicle system: methodologies and challenges | |
BOURJA et al. | Real time vehicle detection, tracking, and inter-vehicle distance estimation based on stereovision and deep learning using YOLOv3 | |
Álvarez et al. | Perception advances in outdoor vehicle detection for automatic cruise control |
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 |