CN109829403A - A kind of vehicle collision avoidance method for early warning and system based on deep learning - Google Patents
A kind of vehicle collision avoidance method for early warning and system based on deep learning Download PDFInfo
- Publication number
- CN109829403A CN109829403A CN201910057691.1A CN201910057691A CN109829403A CN 109829403 A CN109829403 A CN 109829403A CN 201910057691 A CN201910057691 A CN 201910057691A CN 109829403 A CN109829403 A CN 109829403A
- Authority
- CN
- China
- Prior art keywords
- vehicle
- image
- speed
- early warning
- driver
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
The present invention provides a kind of vehicle collision avoidance method for early warning and system based on deep learning, this method carry out recognition detection to the vehicle in video by using improved YOLOv3 algorithm, obtain the location information of vehicle in the picture；Lane detection technology is detected and is tracked to the lane line in video image, is stablized, accurate lane line；Bus- Speed Monitoring technology calculates the travel speed of current vehicle using image processing techniques；According to speed, the behavioral trait of lane detection result and driver and consciousness response characteristic, one piece of safety arrestment prewarning area with speed real-time change is drawn in this front side；Location information of the collision accident early warning technology according to vehicle in the picture predicts the collision accident that may occur on highway in conjunction with the calculated result of actual time safety prewarning area.Using the present invention can the driving decision to driver help is provided, guarantee the traffic safety of driver and reduce the probability of collision to greatest extent.
Description
Technical field
The invention belongs to computer visions, intelligent driving field, and in particular to a kind of preventing collision of vehicles based on deep learning
Hit method for early warning and system.
Background technique
With being continuously increased for China's highway mileage and car ownership, automobile not only increases the modernization of people
Living standard, and the development for having pushed national economy healthy and stable to a certain extent.But this but also highway thing
Therefore incidence is higher and higher, on the one hand causes national economy loss, on the other hand seriously threatens people life property safety, because
This carries out targetedly analysis to the traffic accident that highway occurs and effectively preapres for an unfavorable turn of events in advance particularly important.
Vehicle on highway collision early warning system based on deep learning remembers driving with computer vision technique
It is analyzed and processed provided by record instrument with the video of the first viewing angles of driver.The system includes vehicle identification detection technique,
Lane detection tracking technique, the current vehicle front big core technology of the real-time computing technique three in safety zone.Three big core technologies
Domestic and international Survey on Research is as follows:
(1) the vehicle identification detection technique based on YOLOv3
In order to which the vehicle in image is identified and detected, scholars propose different algorithms, these algorithms can
To be attributed to two classes, the first kind is traditional vehicle identification algorithm, and vehicle identification process is divided into feature extraction and target point by it
In two stages of class, feature extraction needs are artificial constructed, and construct inefficiency, especially in complicated traffic environment, identification knot
Influence of the fruit vulnerable to factors such as illumination, weather, environment, recognition result are undesirable；Second class is based on deep neural network
Target identification detection algorithm, main advantage are that it is unified to a depth network frame by feature extraction, classification, frame positioning
In frame, the efficiency and accuracy of target detection are substantially increased, there is invariance simultaneously for target rotation, displacement, to complexity
Target detection scene have stronger adaptability.Currently, the target identification detection algorithm based on deep neural network is the most
Accurate method, it has surmounted the performance of conventional target detection algorithm, wherein representative deep neural network has R-
CNN, Fast R-CNN, Mask R-CNN, GoogleNet, SSD etc., but these methods slow, Wu Fada that still has calculating speed
To the requirement of real-time.
(2) lane detection technology
There are many current method for Lane detection, wherein using straight line model combination Hough transformation (Hough
Transform) method and apply its improved method, such as improved probability Hough transformation PPHT (Progressive
Probabilistic Hough Transform) it is relatively conventional, in addition there are what Menthon was proposed to utilize road opposite on both sides of the road
The marginal point in direction rebuilds 3D model, and Kluge carries out binaryzation to edge image first, then using least mean-square error come
Estimate that the spline curve model in lane, Broggi detect lane line with Hough transformation using in inverse perspective mapping.These methods
It is all achieved good results in intelligent vehicle field, but these methods based on edge detection are difficult to cope with complicated city road
Road, the influence vulnerable to other edge noises.
(3) the real-time computing technique in safety zone in front of current vehicle
Safety distance model the representative are MAZDA model, HONDA model, Berkeley model, Jaguar model and
NHSTA model is put forward one after another.University of Michigan is based on ICC FOT (Inter Company Case Fill on Time) number
It is evaluated according to rate of failing to report and rate of false alarm of the library to above-mentioned 5 kinds of models, wherein the NHSTA of United States highways traffic safety administration
The performance of model performance is best, but accuracy of alarming only has 23%, therefore existing model algorithm is still significantly improved space.
Summary of the invention
Goal of the invention: the present invention provides a kind of vehicle collision avoidance method for early warning and system based on deep learning, to driving
The driving decision of member provides help, guarantees the traffic safety of driver and reduces the probability of collision to greatest extent.
Technical solution: a kind of vehicle collision avoidance method for early warning based on deep learning of the present invention the following steps are included:
(1) Vehicular video image is pre-processed, the image after obtaining noise reduction；
(2) Bus- Speed Monitoring is carried out, using lane line as object of reference, multiple continuous images is analyzed, current vehicle is obtained
Travel speed, the behavioral trait of the speed and driver, consciousness response characteristic are combined, obtain vehicle with the speed traveling
When emergency stopping distance；
(3) lane detection is carried out, is input with pretreated image, obtains lane line with edge detection algorithm
Edge obtains the set of lane line point followed by Hough line detection algorithm, separates followed by the slope using set midpoint
The two set put are fitted to two optimal straight lines respectively later, realize lane line by the set of left and right lane line point
Detection, finally constantly updates set a little, realizes the tracking of lane line；
(4) by the pixel distance and actual range progress equation model in image, in conjunction with the knot of step (2) and step (3)
Fruit draws out the safety zone changed according to speed in vehicle front, and is visualized on the image；
(5) on the basis of YOLOv3, Optimal Neural Network Architectures model acquires vehicle data, re -training nerve net
Network is realized the real-time identification and detection to vehicle, is obtained the position of vehicle in the picture using the YOLOv3 algorithm after optimization,
And result is stored in information of vehicles queue；
(6) information of vehicles queue is traversed, in conjunction with the safety arrestment area of drafting, the Traffic Collision event that may occur is carried out
Prediction, by human-computer interaction interface, prompts driver.
Pre-process main to video image described in step (1) includes defining area-of-interest, image binaryzation, shape
State operation, Gaussian Blur.
Vehicle Speed described in step (3) is realized by following formula:
Wherein, FPS is the current broadcasting frame number of video, and frameCount (n) is that lane line continuously goes out in image ROI region
Existing number.
The distance of emergency brake of vehicle described in step (2) is crossed following formula and is realized:
Wherein, when driver takes emergency braking measure, t1For the reaction time of driver, t '2For the judgement time of driver,
t″2Braking action the time it takes is made for driver, For attachment coefficient, g=9.8 (m/s2) add for gravity
Speed, V are the travel speed of current vehicle.
The fit equation of pixel distance in actual range described in step (4) and image, is realized by following formula:
Ph=AL3-BL2+CL+D
Wherein, L is the pixel distance in image, PhFor actual range, A, B, C and D are multinomial coefficient.
Optimization data network structure model described in the step (5) the following steps are included:
(51) feature extraction is carried out to full figure by DarkNet-53 network；
(52) input picture is divided into the network unit of 19*19 and generates three predicted boundary frames and original in each cell
The IOU of indicia framing is handed over and ratio, chooses maximum frame and is predicted, is adapted to using the multiple receptive fields of the SPP network integration different
The target of size；
(53) the characteristic pattern progress of both different scales of 19*19,38*38 is used as classifier using logistic regression
Prediction.
A kind of vehicle collision avoidance early warning system based on deep learning of the present invention, including image obtains and pretreatment
Module, multi-threaded parallel computing module, Bus- Speed Monitoring module, lane detection module, safety arrestment area drafting module, vehicle are known
Other detection module, anti-collision warning module；The Bus- Speed Monitoring module carries out multiple continuous images with image processing techniques
Analysis, calculates the travel speed of current vehicle, combines with the behavioral trait of driver and consciousness response characteristic, obtain vehicle
Emergency stopping distance with the speed when driving；The lane detection module, with edge detection algorithm and Hough straight line
Detection algorithm obtains the set of lane line point, is fitted to two optimal straight lines, and constantly updates the set of these points, right
Lane line is tracked；Safety arrestment area drafting module, in image pixel distance and actual range to carry out equation quasi-
It closes, in conjunction with lane detection result and calculated emergency brake of vehicle distance, draws a root tuber in vehicle front and become according to speed
The safety zone of change, and visualized on the image；The vehicle identification detection module, uses improved YOLOv3
Algorithm is identified and is detected to the vehicle in video, obtains the location information of vehicle in the picture, and result is stored in vehicle
In message queue；The anti-collision warning module traverses information of vehicles queue, in conjunction with the safety arrestment area of drafting, sends out possible
Raw Traffic Collision event is predicted.
The utility model has the advantages that compared with prior art, beneficial effects of the present invention: 1, base of the present invention in original YOLOv3 algorithm
On plinth, it is improved, single classification detection is realized, in the case where guaranteeing precision further by the detection frame number of video
It is increased to 43FPS, has better met the demand of system real time；2, method for detecting lane lines effective solution proposed in this paper
These problems, still achieve good detection effect under the interference of edge noise；3, early warning security proposed by the present invention
Region considers the behavioral trait and consciousness response characteristic of driver, can preferably meet driver and knock into the back characteristic, can improve
The early warning rate of system.
Detailed description of the invention
Fig. 1 is that the present invention is based on the vehicle collision avoidance method for early warning flow charts of deep learning；
Fig. 2 is the effect for the lane line that the present invention is extracted using the adaptive threshold fuzziness method based on Gaussian distribution model
Fruit figure；
Fig. 3 is after the present invention screens straight line using perspective transform, to realize the effect picture of lane line tracking；
Fig. 4 is the safe early warning area when speed of the invention obtained using this real-time computing technique in front side safety zone is slower
Domain effect picture；
Fig. 5 is the safe early warning area when speed of the invention obtained using this real-time computing technique in front side safety zone is very fast
Domain effect picture；
Fig. 6 is the effect picture that YOL0v3 carries out after vehicle identification detection in the present invention；
Fig. 7 is that the present invention implements vehicle detection and effect picture is drawn in safe early warning region；
Fig. 8 is the present invention in highway for preventing anti-collision warning effect picture.
Specific embodiment
The present invention will be further described below with reference to the drawings.
As shown in Figure 1, a kind of vehicle collision avoidance method for early warning based on deep learning disclosed by the embodiments of the present invention, including
Following steps:
(1) video image got is pre-processed, provides support for subsequent image procossing.
Video data used in the present embodiment be motor bus in highway driving vehicle-mounted vidicon acquire with department
The video data of the first viewing angles of machine.Since raw video image contains information much unrelated with subsequent image processing operation
And noise, so first having to carry out image preprocessing, processing step mainly includes image gray processing and removal picture noise.Due to
RGB (Red Green Blue) triple channel image divides lane line, feature extraction and related operation have no help, so first
Gray processing processing is first carried out, the image of RGB triple channel is converted into single pass gray level image.Later using morphologic filtering
Method removes the salt-pepper noise in image, obtains smoother image, this is conducive to extract accurately and reliably lane line.
(2) it is input with pretreatment image, uses the adaptive threshold fuzziness method based on Gaussian distribution model first, it will
Lane line is separated from complex background.It is illustrated in figure 2 the effect picture for separating lane line from complicated image,
Wherein background parts are ater, and white dashed line is lane line, this is conducive to the mathematical model for next constructing lane line.
Then selection Canny edge detection operator detects lane line edge, is then mentioned using the method for randomized hough transform
The characteristic point in pick-up road, then construct mathematical model to lane line feature point carry out straight line fitting, followed by utilize perspective transform
Straight line is screened, realizes lane line tracking, last optimization algorithm guarantees the stability and validity of lane detection.As Fig. 3 be
On the basis of Fig. 2, the lane line to isolate constructs mathematical model, lane line is fitted to optimal straight line, and separate
Go out left and right lane line, realizes the effect picture of lane line tracking.
(3) using pretreatment image as data source, continuous multiple image is analyzed and processed, realizes the calculating of speed.
One piece of ROI (Region of Interest) area-of-interest is defined first, is calculated with the Corner Detection in image processing techniques
Method counts the number frameCount (n) that lane line continuously occurs in ROI region predetermined, calculates vehicle velocity V with following formula
(km/h):
In formula, FPS indicates the current broadcasting frame number of video, and here according to the broadcasting frame number of video, to select FPS be 25,
FrameCount (n) is the number that lane line continuously occurs in image ROI region.
According to calculated vehicle speed V (km/h), in conjunction with the behavioral trait and consciousness response characteristic of driver, under use
Formula calculates the emergency stopping distance L of vehicle when driving with present speed:
In formula, takes and judge the time 1.5 seconds, action time 1 second, i.e. t1+t′2=1.5, t "2=1.This project is in highway
It is tested, i.e., road surface is bituminous pavement, takes attachment coefficientAcceleration of gravity is g=9.8 (m/s2), speed V is to work as
The travel speed of vehicle in front.
(4) actual range is reflected on the image, needs to fit an equation, i.e., converts actual range
Method for the pixel distance on image, use is: according to the specification (6 off-white lines, 9 meters of intervals) of China's lane line, in image
Multiple characteristic points are demarcated on middle lane line vertical direction, according to concrete application scene, the data of the characteristic point of acquisition such as 1 institute of table
Show:
The characteristic point data that table 1 acquires
| Actual range (Ph) | 15 | 21 | 30 | 36 | 45 |
| Pixel distance (L) | 215 | 240 | 262 | 274 | 283 |
By researching and analysing the correlation between these data, due to being non-linear relation between these data, so with
Unitary cubic polynomial Ph=AL3-BL2+ CL+D constructs the mathematical models of these data, uses least square method (model of fit
Reach minimum in the weighted sum of squares of the residual error of each point with actual observed value) find out the expression of the mathematical model to get
To pixels tall Ph(px) and the fit equation of actual range L (m) is as follows:
Ph=0.0012L3-0.1674L2+8.882L+115.72
Wherein, L is the pixel distance in image, PhFor actual range.It brings the calculated result L in 4 into above formula, can obtain
To actual range corresponding pixels tall P in the pictureh。
(5) this real-time computing technique in front side safety zone is then used, realizes the real-time calculating in safe early warning region, method
Are as follows: according to pixels tall Ph, in conjunction with the mathematical model that lane detection is tracked, calculate with PhThe straight line of height and left and right
The intersecting point coordinate of two lane lines draws one piece of safe early warning region changed with speed in this front side according to coordinate.Such as Fig. 4
It is respectively the safe early warning regional effect figure that is obtained using this real-time computing technique in front side safety zone with Fig. 5, wherein shadow part
The height divided is the calculated maximum emergency stopping distance of travel speed according to current vehicle, and Fig. 4 is that speed is drawn when slower
Safe early warning area, at this time dash area height it is lower；Fig. 5 is the safe early warning area drawn when speed is very fast, at this time shadow part
Divide height higher.The height in especially early warning security region is the travel speed with vehicle and dynamic change.
(6) algorithm of target detection-YOLOv3 algorithm by a kind of based on deep neural network is to the vehicle in video image
It is identified and is detected in real time, to obtain the location information of front truck in the picture.This project is based on YOLOv3 algorithm, for vehicle
Recognition detection demand is improved on the basis of original YOLOv3 algorithm, its network model is optimized, from COCO (Common
Objects in Context) automobile is picked out in data set image as training data, batch normalization is carried out to image
And gray processing is finally obtained the network model file after training, is reduced using the neural network structure training data after optimization
The parameter amount of YOLOv3 realizes single classification detection, further improves the detection frame number of video in the case where guaranteeing precision
To 43FPS, to further increase the speed and precision of vehicle recognition detection under complex background.If Fig. 6 is after application enhancements
YOLOv3 carries out the effect picture after vehicle identification detection, wherein the vehicle being detected can be marked with a box, and
The box upper left corner indicates the classification of detected target.
(7) the real-time computing technique in safety zone in front of current vehicle and vehicle identification detection technique based on YOLOv3 are carried out
It is comprehensive, while drawing safe early warning region, recognition detection is carried out to the vehicle in video, Fig. 7 is actual time safety precautionary areas
Draw identification and detection effect figure with vehicle in domain.According to the location information of vehicle in the picture, in conjunction with the actual time safety of drafting
Prewarning area predicts the collision accident that may occur.Fig. 8 is highway for preventing anti-collision warning effect picture, when detecting
Front truck target when being in left (right side) side in early warning security region, system gives the early warning that vehicle is carried out in a driver left side (right side) side, when detecting
Front truck target when being in inside early warning security region, system gives driver's early warning to drive with caution.
A kind of vehicle collision avoidance early warning system based on deep learning disclosed by the invention, comprising: image obtains and pre- place
It manages module and provides data source for obtaining vehicle-mounted vidicon video data for system, while being located in advance to the video of acquisition
Reason provides input for next operation.Multi-threaded parallel computing module makes full use of for realizing multi-threaded parallel operation
The resource of computer accelerates calculating speed, guarantees the real-time of system.Bus- Speed Monitoring module, with image processing techniques, to even
Continue multiple images to be analyzed, calculate the travel speed of current vehicle, then by the behavioral trait of the speed and driver and
Consciousness response characteristic combines, and obtains the emergency stopping distance of vehicle when driving with the speed.Lane detection module, with side
Edge detection algorithm and Hough line detection algorithm obtain the set of lane line point, are then fitted to these set put optimal
Two straight lines, realize the detection of lane line, finally constantly update the set of these points, are tracked to lane line.Safety arrestment
Area's drafting module, to the pixel distance and actual range progress equation model in image, in conjunction with lane detection result and calculating
Emergency brake of vehicle distance out is drawn the safety zone that a root tuber changes according to speed in vehicle front, and is carried out on the image
It visualizes.Vehicle identification detection module is identified and is examined to the vehicle in video using improved YOLOv3 algorithm
It surveys, obtains the location information of vehicle in the picture, and result is stored in information of vehicles queue.Anti-collision warning module, traversal
Information of vehicles queue predicts the Traffic Collision event that may occur in conjunction with the safety arrestment area of drafting.I.e. if vehicle
Position in left and right lane line two sides, then to driver issue pay attention to left and right Lai Che voice broadcast, if the position of vehicle is in
In left and right lane line (safety arrestment area), then the voice broadcast for paying attention to slowing down is issued to driver.
Claims (7)
1. a kind of vehicle collision avoidance method for early warning based on deep learning, which comprises the following steps:
(1) Vehicular video image is pre-processed, the image after obtaining noise reduction；
(2) Bus- Speed Monitoring is carried out, using lane line as object of reference, multiple continuous images is analyzed, the row of current vehicle is obtained
Speed is sailed, the behavioral trait of the speed and driver, consciousness response characteristic are combined, obtains vehicle with the speed when driving
Emergency stopping distance；
(3) lane detection is carried out, is input with pretreated image, obtains the side of lane line with edge detection algorithm
Edge obtains the set of lane line point followed by Hough line detection algorithm, separates a left side followed by the slope using set midpoint
The two set put are fitted to two optimal straight lines respectively later, realize the inspection of lane line by the set of right-lane line point
It surveys, finally constantly updates set a little, realize the tracking of lane line；
(4) by image pixel distance and actual range carry out equation model, in conjunction with step (2) and step (3) as a result,
Vehicle front draws out the safety zone changed according to speed, and is visualized on the image；
(5) on the basis of YOLOv3, Optimal Neural Network Architectures model acquires vehicle data, and re -training neural network makes
With the YOLOv3 algorithm after optimization, real-time identification and detection to vehicle are realized, obtain the position of vehicle in the picture, and will knot
Fruit is stored in information of vehicles queue；
(6) information of vehicles queue is traversed, in conjunction with the safety arrestment area of drafting, the Traffic Collision event that may occur is carried out pre-
It surveys, by human-computer interaction interface, driver is prompted.
2. a kind of vehicle collision avoidance method for early warning based on deep learning according to claim 1, which is characterized in that step
(1) video image pre-process mainly including defining area-of-interest, image binaryzation, morphological operation, height described in
This is fuzzy.
3. a kind of vehicle collision avoidance method for early warning based on deep learning according to claim 1, which is characterized in that step
(3) Vehicle Speed described in is realized by following formula:
Wherein, FPS is the current broadcasting frame number of video, and frameCount (n) is that lane line continuously occurs in image ROI region
Number.
4. a kind of vehicle collision avoidance method for early warning based on deep learning according to claim 1, which is characterized in that step
(2) the emergency brake of vehicle distance described in is crossed following formula and is realized:
Wherein, when driver takes emergency braking measure, t1For the reaction time of driver, t '2For the judgement time of driver, t "2For
Driver makes braking action the time it takes, For attachment coefficient, g=9.8 (m/s2) it is acceleration of gravity, V
For the travel speed of current vehicle.
5. a kind of vehicle collision avoidance method for early warning based on deep learning according to claim 1, which is characterized in that step
(4) fit equation of pixel distance in the actual range and image described in, is realized by following formula:
Ph=AL3-BL2+CL+D
Wherein, L is the pixel distance in image, PhFor actual range, A, B, C and D are multinomial coefficient.
6. a kind of vehicle collision avoidance method for early warning based on deep learning according to claim 1, which is characterized in that described
Optimization data network structure model described in step (5) the following steps are included:
(51) feature extraction is carried out to full figure by DarkNet-53 network；
(52) input picture is divided into the network unit of 19*19 and generates three predicted boundary frames in each cell and marked with former
The IOU of frame is handed over and ratio, chooses maximum frame and is predicted, adapts to different size using the multiple receptive fields of the SPP network integration
Target；
(53) it using logistic regression as classifier, is predicted using the characteristic pattern of 19*19,38*38 both different scales.
7. a kind of vehicle collision avoidance early warning system based on deep learning, including image are obtained with preprocessing module, multithreading simultaneously
Row computing module, lane detection module, safety arrestment area drafting module, vehicle identification detection module, touches at Bus- Speed Monitoring module
Hit warning module, which is characterized in that the Bus- Speed Monitoring module divides multiple continuous images with image processing techniques
Analysis, calculates the travel speed of current vehicle, combines with the behavioral trait of driver and consciousness response characteristic, obtain vehicle
With the emergency stopping distance of the speed when driving；The lane detection module is examined with edge detection algorithm and Hough straight line
Method of determining and calculating obtains the set of lane line point, is fitted to two optimal straight lines, and constantly updates the set of these points, to vehicle
Diatom is tracked；Safety arrestment area drafting module, in image pixel distance and actual range carry out equation model,
In conjunction with lane detection result and calculated emergency brake of vehicle distance, draw what a root tuber changed according to speed in vehicle front
Safety zone, and visualized on the image；The vehicle identification detection module, is calculated using improved YOLOv3
Method is identified and is detected to the vehicle in video, obtains the location information of vehicle in the picture, and result is stored in vehicle
In message queue；The anti-collision warning module traverses information of vehicles queue, in conjunction with the safety arrestment area of drafting, occurs to possible
Traffic Collision event predicted.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910057691.1A CN109829403B (en) | 2019-01-22 | 2019-01-22 | Vehicle anti-collision early warning method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910057691.1A CN109829403B (en) | 2019-01-22 | 2019-01-22 | Vehicle anti-collision early warning method and system based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109829403A true CN109829403A (en) | 2019-05-31 |
| CN109829403B CN109829403B (en) | 2020-10-16 |
Family
ID=66861075
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910057691.1A Active CN109829403B (en) | 2019-01-22 | 2019-01-22 | Vehicle anti-collision early warning method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109829403B (en) |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109377785A (en) * | 2018-10-22 | 2019-02-22 | 北京航空航天大学 | A kind of highway merging area vehicle collision method for early warning based on deep learning |
| CN110276988A (en) * | 2019-06-26 | 2019-09-24 | 重庆邮电大学 | A kind of DAS (Driver Assistant System) based on collision warning algorithm |
| CN111091061A (en) * | 2019-11-20 | 2020-05-01 | 浙江工业大学 | Vehicle scratch detection method based on video analysis |
| CN111160231A (en) * | 2019-12-27 | 2020-05-15 | 北京联合大学 | Automatic driving environment road extraction method based on Mask R-CNN |
| CN111309023A (en) * | 2020-03-03 | 2020-06-19 | 西安交通大学 | Lane keeping control system and method based on all-round vision image and machine vision |
| CN111369496A (en) * | 2020-02-18 | 2020-07-03 | 中北大学 | Pupil center positioning method based on star ray |
| CN111611942A (en) * | 2020-05-22 | 2020-09-01 | 山东大学 | Method for extracting and building database by perspective self-adaptive lane skeleton |
| CN112070039A (en) * | 2020-09-11 | 2020-12-11 | 广州亚美智造科技有限公司 | Vehicle collision detection method and system based on Hash coding |
| 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 |
| CN112419746A (en) * | 2019-08-20 | 2021-02-26 | 广州中广国科测控技术有限公司 | Vehicle speed self-monitoring method for highway traffic |
| CN112991733A (en) * | 2021-03-02 | 2021-06-18 | 南京航空航天大学 | Intelligent network connection collision optimization system and method for bus |
| CN113068000A (en) * | 2019-12-16 | 2021-07-02 | 杭州海康威视数字技术股份有限公司 | Method, device, equipment and system for monitoring video target and storage medium |
| CN113192363A (en) * | 2021-04-29 | 2021-07-30 | 杭州电子科技大学 | Video data edge calculation method based on artificial intelligence algorithm |
| CN113306566A (en) * | 2021-06-16 | 2021-08-27 | 上海大学 | Vehicle and pedestrian collision early warning method and device based on sniffing technology |
| CN113353071A (en) * | 2021-05-28 | 2021-09-07 | 云度新能源汽车有限公司 | Narrow area intersection vehicle safety auxiliary method and system based on deep learning |
| CN113370977A (en) * | 2021-05-06 | 2021-09-10 | 上海大学 | Intelligent vehicle forward collision early warning method and system based on vision |
| CN113408320A (en) * | 2020-03-16 | 2021-09-17 | 上海博泰悦臻网络技术服务有限公司 | Method, electronic device, and computer storage medium for vehicle collision avoidance |
| CN113486837A (en) * | 2021-07-19 | 2021-10-08 | 安徽江淮汽车集团股份有限公司 | Automatic driving control method for low-pass obstacle |
| CN113657265A (en) * | 2021-08-16 | 2021-11-16 | 长安大学 | Vehicle distance detection method, system, device and medium |
| CN113963531A (en) * | 2021-09-07 | 2022-01-21 | 北京动视元科技有限公司 | Detection method for keeping safe distance of running vehicle and display mode thereof |
| CN114838796A (en) * | 2022-04-29 | 2022-08-02 | 合肥市正茂科技有限公司 | Vision-assisted vehicle dynamic weighing method and weighing system |
| CN115294767A (en) * | 2022-08-02 | 2022-11-04 | 山西省智慧交通研究院有限公司 | Real-time detection and traffic safety early warning method and device for highway lane lines |
| CN115376082A (en) * | 2022-08-02 | 2022-11-22 | 北京理工大学 | Lane line detection method integrating traditional feature extraction and deep neural network |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101131321A (en) * | 2007-09-28 | 2008-02-27 | 深圳先进技术研究院 | Real-time safe interval measurement method and device used for vehicle anti-collision warning |
| CN101391589A (en) * | 2008-10-30 | 2009-03-25 | 上海大学 | Vehicle intelligent alarming method and device |
| KR20130067842A (en) * | 2011-12-14 | 2013-06-25 | 한국전자통신연구원 | Image based object information detecting method and apparatus |
| CN103832433A (en) * | 2012-11-21 | 2014-06-04 | 中国科学院沈阳计算技术研究所有限公司 | Lane departure and front collision warning system and achieving method thereof |
| CN105912998A (en) * | 2016-04-05 | 2016-08-31 | 辽宁工业大学 | Vehicle collision prevention early warning method based on vision |
| CN107609472A (en) * | 2017-08-04 | 2018-01-19 | 湖南星云智能科技有限公司 | A kind of pilotless automobile NI Vision Builder for Automated Inspection based on vehicle-mounted dual camera |
| CN108490950A (en) * | 2018-04-23 | 2018-09-04 | 安徽江淮汽车集团股份有限公司 | A kind of lane change method and system based on virtual rail |
| CN109064744A (en) * | 2018-08-14 | 2018-12-21 | 淮阴工学院 | A kind of biasing right-turn lane Design of length method based on Queueing Theory Method |
-
2019
- 2019-01-22 CN CN201910057691.1A patent/CN109829403B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101131321A (en) * | 2007-09-28 | 2008-02-27 | 深圳先进技术研究院 | Real-time safe interval measurement method and device used for vehicle anti-collision warning |
| CN101391589A (en) * | 2008-10-30 | 2009-03-25 | 上海大学 | Vehicle intelligent alarming method and device |
| KR20130067842A (en) * | 2011-12-14 | 2013-06-25 | 한국전자통신연구원 | Image based object information detecting method and apparatus |
| CN103832433A (en) * | 2012-11-21 | 2014-06-04 | 中国科学院沈阳计算技术研究所有限公司 | Lane departure and front collision warning system and achieving method thereof |
| CN105912998A (en) * | 2016-04-05 | 2016-08-31 | 辽宁工业大学 | Vehicle collision prevention early warning method based on vision |
| CN107609472A (en) * | 2017-08-04 | 2018-01-19 | 湖南星云智能科技有限公司 | A kind of pilotless automobile NI Vision Builder for Automated Inspection based on vehicle-mounted dual camera |
| CN108490950A (en) * | 2018-04-23 | 2018-09-04 | 安徽江淮汽车集团股份有限公司 | A kind of lane change method and system based on virtual rail |
| CN109064744A (en) * | 2018-08-14 | 2018-12-21 | 淮阴工学院 | A kind of biasing right-turn lane Design of length method based on Queueing Theory Method |
Non-Patent Citations (1)
| Title |
|---|
| KWANG-JU KIM等: "Performance Enhancement of YOLOv3 by Adding Prediction Layers with Spatial Pyramid Pooling for Vehicle Detection", 《2018 15TH IEEE INTERNATIONAL CONFERENCE ON ADVANCED VIDEO AND SIGNAL BASED SURVEILLANCE》 * |
Cited By (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109377785A (en) * | 2018-10-22 | 2019-02-22 | 北京航空航天大学 | A kind of highway merging area vehicle collision method for early warning based on deep learning |
| CN110276988A (en) * | 2019-06-26 | 2019-09-24 | 重庆邮电大学 | A kind of DAS (Driver Assistant System) based on collision warning algorithm |
| CN112419746A (en) * | 2019-08-20 | 2021-02-26 | 广州中广国科测控技术有限公司 | Vehicle speed self-monitoring method for highway traffic |
| CN111091061A (en) * | 2019-11-20 | 2020-05-01 | 浙江工业大学 | Vehicle scratch detection method based on video analysis |
| CN113068000B (en) * | 2019-12-16 | 2023-07-18 | 杭州海康威视数字技术股份有限公司 | Video target monitoring method, device, equipment, system and storage medium |
| CN113068000A (en) * | 2019-12-16 | 2021-07-02 | 杭州海康威视数字技术股份有限公司 | Method, device, equipment and system for monitoring video target and storage medium |
| CN111160231A (en) * | 2019-12-27 | 2020-05-15 | 北京联合大学 | Automatic driving environment road extraction method based on Mask R-CNN |
| CN111369496A (en) * | 2020-02-18 | 2020-07-03 | 中北大学 | Pupil center positioning method based on star ray |
| CN111369496B (en) * | 2020-02-18 | 2022-07-01 | 中北大学 | Pupil center positioning method based on star ray |
| CN111309023A (en) * | 2020-03-03 | 2020-06-19 | 西安交通大学 | Lane keeping control system and method based on all-round vision image and machine vision |
| CN113408320A (en) * | 2020-03-16 | 2021-09-17 | 上海博泰悦臻网络技术服务有限公司 | Method, electronic device, and computer storage medium for vehicle collision avoidance |
| CN111611942A (en) * | 2020-05-22 | 2020-09-01 | 山东大学 | Method for extracting and building database by perspective self-adaptive lane skeleton |
| CN111611942B (en) * | 2020-05-22 | 2022-03-22 | 山东大学 | Method for extracting and building database by perspective self-adaptive lane skeleton |
| CN112070039A (en) * | 2020-09-11 | 2020-12-11 | 广州亚美智造科技有限公司 | Vehicle collision detection method and system based on Hash coding |
| CN112070039B (en) * | 2020-09-11 | 2023-08-15 | 亚美智联数据科技有限公司 | Hash code-based vehicle collision detection method and system |
| 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 |
| CN112991733A (en) * | 2021-03-02 | 2021-06-18 | 南京航空航天大学 | Intelligent network connection collision optimization system and method for bus |
| CN113192363A (en) * | 2021-04-29 | 2021-07-30 | 杭州电子科技大学 | Video data edge calculation method based on artificial intelligence algorithm |
| CN113370977A (en) * | 2021-05-06 | 2021-09-10 | 上海大学 | Intelligent vehicle forward collision early warning method and system based on vision |
| CN113353071A (en) * | 2021-05-28 | 2021-09-07 | 云度新能源汽车有限公司 | Narrow area intersection vehicle safety auxiliary method and system based on deep learning |
| CN113353071B (en) * | 2021-05-28 | 2023-12-19 | 云度新能源汽车有限公司 | Narrow area intersection vehicle safety auxiliary method and system based on deep learning |
| CN113306566A (en) * | 2021-06-16 | 2021-08-27 | 上海大学 | Vehicle and pedestrian collision early warning method and device based on sniffing technology |
| CN113306566B (en) * | 2021-06-16 | 2023-12-12 | 上海大学 | Vehicle pedestrian collision early warning method and device based on sniffing technology |
| CN113486837A (en) * | 2021-07-19 | 2021-10-08 | 安徽江淮汽车集团股份有限公司 | Automatic driving control method for low-pass obstacle |
| CN113657265A (en) * | 2021-08-16 | 2021-11-16 | 长安大学 | Vehicle distance detection method, system, device and medium |
| CN113657265B (en) * | 2021-08-16 | 2023-10-10 | 长安大学 | Vehicle distance detection method, system, equipment and medium |
| CN113963531A (en) * | 2021-09-07 | 2022-01-21 | 北京动视元科技有限公司 | Detection method for keeping safe distance of running vehicle and display mode thereof |
| CN114838796A (en) * | 2022-04-29 | 2022-08-02 | 合肥市正茂科技有限公司 | Vision-assisted vehicle dynamic weighing method and weighing system |
| CN114838796B (en) * | 2022-04-29 | 2023-06-09 | 合肥市正茂科技有限公司 | Visual auxiliary vehicle dynamic weighing method and system |
| CN115294767A (en) * | 2022-08-02 | 2022-11-04 | 山西省智慧交通研究院有限公司 | Real-time detection and traffic safety early warning method and device for highway lane lines |
| CN115376082A (en) * | 2022-08-02 | 2022-11-22 | 北京理工大学 | Lane line detection method integrating traditional feature extraction and deep neural network |
| CN115294767B (en) * | 2022-08-02 | 2023-06-27 | 山西省智慧交通研究院有限公司 | Real-time detection and traffic safety early warning method and device for expressway lane line |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109829403B (en) | 2020-10-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109829403A (en) | A kind of vehicle collision avoidance method for early warning and system based on deep learning | |
| CN107031623B (en) | A kind of road method for early warning based on vehicle-mounted blind area camera | |
| CN105550665B (en) | A kind of pilotless automobile based on binocular vision can lead to method for detecting area | |
| CN100544446C (en) | The real time movement detection method that is used for video monitoring | |
| CN103522970B (en) | Based on vehicle driving safety method of inspection and the system of machine vision | |
| CN103400157B (en) | Road pedestrian and non-motor vehicle detection method based on video analysis | |
| CN108983219A (en) | A kind of image information of traffic scene and the fusion method and system of radar information | |
| CN106682586A (en) | Method for real-time lane line detection based on vision under complex lighting conditions | |
| CN108596129A (en) | A kind of vehicle based on intelligent video analysis technology gets over line detecting method | |
| CN107341454A (en) | The detection method and device of barrier, electronic equipment in a kind of scene | |
| CN106951879A (en) | Multiple features fusion vehicle checking method based on camera and millimetre-wave radar | |
| CN105844222A (en) | System and method for front vehicle collision early warning based on visual sense | |
| CN104778444A (en) | Method for analyzing apparent characteristic of vehicle image in road scene | |
| CN106022243B (en) | A kind of retrograde recognition methods of the car lane vehicle based on image procossing | |
| CN103927548B (en) | Novel vehicle collision avoiding brake behavior detection method | |
| CN103984950A (en) | Moving vehicle stop lamp state recognition method adaptable to day detection | |
| CN103455820A (en) | Method and system for detecting and tracking vehicle based on machine vision technology | |
| CN109948552B (en) | Method for detecting lane line in complex traffic environment | |
| CN113370977B (en) | Intelligent vehicle forward collision early warning method and system based on vision | |
| CN104182756A (en) | Method for detecting barriers in front of vehicles on basis of monocular vision | |
| CN115205796B (en) | Rail line foreign matter intrusion monitoring and risk early warning method and system | |
| CN107315998A (en) | Vehicle class division method and system based on lane line | |
| CN111553214B (en) | Method and system for detecting smoking behavior of driver | |
| CN114359876B (en) | Vehicle target identification method and storage medium | |
| CN106951820B (en) | Passenger flow statistical method based on annular template and ellipse fitting |
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 |