CN107229906A - A kind of automobile overtaking's method for early warning based on units of variance model algorithm - Google Patents

A kind of automobile overtaking's method for early warning based on units of variance model algorithm Download PDF

Info

Publication number
CN107229906A
CN107229906A CN201710318142.6A CN201710318142A CN107229906A CN 107229906 A CN107229906 A CN 107229906A CN 201710318142 A CN201710318142 A CN 201710318142A CN 107229906 A CN107229906 A CN 107229906A
Authority
CN
China
Prior art keywords
automobile
vehicles
early warning
vehicle
msub
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201710318142.6A
Other languages
Chinese (zh)
Inventor
刘绪
张伟伟
吴训成
王慧敏
王鑫琛
张世平
刘东旭
戚文强
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 University of Engineering Science
Original Assignee
Shanghai University of Engineering Science
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 Shanghai University of Engineering Science filed Critical Shanghai University of Engineering Science
Priority to CN201710318142.6A priority Critical patent/CN107229906A/en
Publication of CN107229906A publication Critical patent/CN107229906A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
    • G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
    • G06V20/584—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of vehicle lights or traffic lights
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
    • B60W50/08—Interaction between the driver and the control system
    • B60W50/14—Means for informing the driver, warning the driver or prompting a driver intervention
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/48—Extraction of image or video features by mapping characteristic values of the pattern into a parameter space, e.g. Hough transformation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
    • G06V10/507—Summing image-intensity values; Histogram projection analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
    • G06V20/588—Recognition of the road, e.g. of lane markings; Recognition of the vehicle driving pattern in relation to the road
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
    • B60W50/08—Interaction between the driver and the control system
    • B60W50/14—Means for informing the driver, warning the driver or prompting a driver intervention
    • B60W2050/143—Alarm means
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/08—Detecting or categorising vehicles

Abstract

The present invention relates to a kind of automobile overtaking's method for early warning based on units of variance model algorithm, comprise the following steps：Front vehicle is overtaken other vehicles warning step, front vehicle is identified by rearview camera combination units of variance model algorithm for automobile, by recognizing the position relationship between front vehicle and Ben Che and lane line, judge whether front vehicle is overtaken other vehicles and send early warning to driver；Front vehicles are overtaken other vehicles warning step, front vehicles are identified by forward sight camera combination units of variance model algorithm for automobile, by recognizing the position relationship between front vehicles and Ben Che and lane line, judge whether this car needs to be overtaken other vehicles and send early warning to driver when overtaking other vehicles dangerous.Compared with prior art, the present invention has the advantages that early warning is accurate, early warning speed is fast and it is convenient to realize.

Description

A kind of automobile overtaking's method for early warning based on units of variance model algorithm
Technical field
It is pre- more particularly, to a kind of automobile overtaking based on units of variance model algorithm the present invention relates to early warning field of overtaking other vehicles Alarm method.
Background technology
It is one of most commonly seen driving behavior of driver to overtake other vehicles.According to statistics, on a highway driver with 90km/h About 50 passing behaviors will be carried out by travelling in 100km distance, way.In recent years, China is because of the traffic thing for improper initiation of overtaking other vehicles Therefore in obvious ascendant trend, especially on a highway, more than 60% traffic accident is all with overtaking other vehicles relevant.Implementation is overtaken other vehicles When, driver must according to surrounding enviroment information such as current speed, vehicle spacing, wagon flow state and road traffic facilities, Adjustment driving strategy realizes passing behavior in real time.The vehicle collision for avoiding overtaking process from triggering, can be by controlling the phase between vehicle Speed and increase longitudinal direction of car spacing are realized.
Publication No. CN105216797A patent document discloses the technical scheme of entitled method of overtaking and system, should Scheme includes detection module, processing module, passes through the travel speed and positional information of video camera and preposition radar acquired disturbance thing To judge that can target vehicle perform passing maneuver；Publication No. CN101326511 patent document discloses entitled be used for The technical scheme of the method for detection or predicting vehicle cut-ins, in the embodiment of the program, other are read using radio communication The navigation system of vehicle, to determine whether another vehicle can overtake other vehicles.
At present, at home with overtake other vehicles and the correlation technique of lane change in, majority use radar or short distance information communication systems The generation of traffic accident when system is to prevent from overtaking other vehicles, this scheme is huge because of use multiple sensors cost, and in real-time field Vehicle and lane line information can not be fast and effectively recognized in scape, or is only the distance between prediction barrier and vehicle Information.
The content of the invention
The purpose of the present invention is to provide a kind of automobile overtaking's early warning based on units of variance model algorithm regarding to the issue above Method.
The purpose of the present invention can be achieved through the following technical solutions：
A kind of automobile overtaking's method for early warning based on units of variance model algorithm, methods described comprises the following steps：
Front vehicle is overtaken other vehicles warning step, and automobile is by rearview camera combination units of variance model algorithm to front vehicle It is identified, by recognizing the position relationship between front vehicle and Ben Che and lane line, judges whether front vehicle is carried out Overtake other vehicles and send early warning to driver；
Front vehicles are overtaken other vehicles warning step, and automobile is by forward sight camera combination units of variance model algorithm to front vehicles It is identified, by recognizing the position relationship between front vehicles and Ben Che and lane line, judges whether this car needs progress Overtake other vehicles and send early warning to driver when overtaking other vehicles dangerous.
Front vehicle warning step of overtaking other vehicles is specially：
A1) whether automobile there are other cars according to units of variance model algorithm in the shooting image for detecting rearview camera , if then entering step A2), if otherwise continuing to detect；
A2) automobile judges whether the distance between the vehicle that detects and this car reduce, if then into step A3), if Otherwise return to step A1)；
A3) lane line is identified automobile, and whether the vehicle for judging this car and detecting is located in same lane line, If then sending early warning to driver, if otherwise carrying out step A4)；
A4) automobile is overtaken other vehicles warning step after the vehicle detected disappears out of rearview camera into front vehicles.
Front vehicles warning step of overtaking other vehicles is specially：
B1) whether automobile there are other cars according to units of variance model algorithm in the shooting image for detecting forward sight camera , if then entering step B2), if otherwise continuing to detect；
B2) automobile judges whether the distance between the vehicle that detects and this car reduce, if then into step B3), if Otherwise return to step B1)；
B3) lane line is identified automobile, and whether the vehicle for judging this car and detecting is located in same lane line, If then sending early warning to driver, if otherwise carrying out step B4)；
B4) automobile is overtaken other vehicles warning step after the vehicle detected disappears out of forward sight camera into front vehicle.
The units of variance model algorithm is specially：
A11) according to the shooting image of rearview camera, yardstick pyramid is built；
A12) in each scale layer in yardstick pyramid, by sliding window detection method by image with by variable Model after partial model Algorithm for Training is matched, and calculates matching fraction；
A13 the corresponding scale layer of highest matching fraction) is chosen, shows to detect vehicle if the matching fraction exceedes threshold value And vehicle is appeared in matching the position corresponding to fraction highest in the scale layer.
It is described matching fraction be specially：
Wherein, score is matching fraction, (x0,y0) it is anchor point, l0For the yardstick number of plies,For the detection fraction of root model,For the response of i-th of partial model, viOffset for i-th of partial model relative to anchor point, b is deviation ratio.
It is described by units of variance model algorithm train after model be specially：By using PCA methods to training picture The processing of HOG Feature Dimension Reductions after carry out feature extraction, then model training is carried out by hidden variable support vector machine classifier, obtained Model after being trained by units of variance model algorithm.
The automobile judges that the distance between vehicle and this car detected is specially：Pass through the laser being installed on automobile Radar carries out ranging, the distance between the vehicle detected and this car.
Lane line is identified specially the automobile：Automobile is by Hough transform, the collinear points in scan image, Realize the identification to lane line.
The early warning includes realizing early warning by LED early warning panel flash and realizing early warning by buzzer rings.
The LED early warning panel is provided with the LED early-warning lamps for being at least distributed in four direction all around.
Compared with prior art, the invention has the advantages that：
(1) by units of variance model algorithm, the image photographed to forward sight camera and rearview camera carries out vehicle Identification, then again by judging that the position relationship between the distance between vehicle and this car relation and vehicle and lane line determines car With this car whether occur in overtaking process dangerous and this car whether occur in overtaking process it is dangerous, this method with it is existing The method being identified by radar or short haul connection having is compared, and the accuracy recognized while recognition speed is fast also compares It is higher, and the method being identified by camera Costco Wholesale compared with radar etc. is relatively low.
(2) vehicle overtakes other vehicles warning step by front vehicle and front vehicles two steps of warning step of overtaking other vehicles are combined, Realize tracking control of full process to the vehicle in this car safe range, i.e., this car rear vehicle after overtaking other vehicles again by preceding Square vehicle cut-ins warning step is monitored, this car continue after overtaking other vehicles to more than vehicle overtaken other vehicles early warning by front vehicle Step is monitored, it is ensured that the integrality of early warning.
(3) when detection is identified to vehicle by units of variance model algorithm, ensured by building yardstick pyramid Shooting picture under each yardstick is identified, and the picture for choosing detection highest scoring is protected as the identification position of vehicle The integrality of detection is demonstrate,proved, so as to improve the levels of precision of early warning.
(4) training picture is carried out feature extraction and trained by units of variance model algorithm to obtain model, this method The model levels of precision of training is higher, so as to improve the order of accuarcy of final early warning.
(5) whether automobile recognizes lane line by Hough transform, then may determine that Ben Che with early warning vehicle in same In lane line, thus the two same lane line and apart from reduce when early warning is sent to driver in time, accident is avoided in time Generation.
(6) by LED flicker and buzzer rings simultaneously come to driver's progress early warning, farthest prompting drives The generation of member's dangerous situation, improves the security of early warning.
(7) LED early warning panel is provided with the LED early-warning lamps for being at least distributed in four direction all around, will can detect To early warning situation occur position showed by LED early-warning lamps, be easy to driver understand in time early warning generation position Put, improve the promptness of early warning.
Brief description of the drawings
Fig. 1 is that front vehicle is overtaken other vehicles the method flow diagram of warning step；
Fig. 2 is that front vehicles are overtaken other vehicles the method flow diagram of warning step；
Fig. 3 is each module placement figure of automobile for realizing automobile overtaking's method for early warning based on units of variance model algorithm；
Fig. 4 is the angular field of view schematic diagram of camera；
Fig. 5 is the holistic approach flow chart of early warning of overtaking other vehicles in embodiment；
Fig. 6 is the illustraton of model of units of variance model algorithm, wherein, (6a) is the root mould of one of component under the first visual angle Type figure, (6b) is the partial model figure of one of component under the first visual angle, and (6c) is the model deformation loss under the first visual angle, (6d) is the root illustraton of model of one of component under the second visual angle, and (6e) is the partial model of one of component under the second visual angle Figure, (6f) is the model deformation loss under the second visual angle；
Fig. 7 is the matching process figure of model；
Fig. 8 is LED early warning panel schematic diagrames；
Wherein, 111 be vehicle, and 301 be warning module, and 302 be forward sight camera, and 303 be data processing module, and 304 are Rearview camera, 401~406 be LED.
Embodiment
The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.The present embodiment is with technical solution of the present invention Premised on implemented, give detailed embodiment and specific operating process, but protection scope of the present invention is not limited to Following embodiments.
The present embodiment, which provides a kind of automobile overtaking's method for early warning based on units of variance model algorithm, includes following step Suddenly：
Front vehicle is overtaken other vehicles warning step, and automobile is by rearview camera combination units of variance model algorithm to front vehicle It is identified, by recognizing the position relationship between front vehicle and Ben Che and lane line, judges whether front vehicle is carried out Overtake other vehicles and send early warning to driver, be specially：
A1) whether automobile there are other cars according to units of variance model algorithm in the shooting image for detecting rearview camera , if then entering step A2), if otherwise continuing to detect：
A11) according to the shooting image of rearview camera, yardstick pyramid is built；
A12) in each scale layer in yardstick pyramid, by sliding window detection method by image with by variable Model after partial model Algorithm for Training is matched, and calculates matching fraction；
A13 the corresponding scale layer of highest matching fraction) is chosen, shows to detect vehicle if the matching fraction exceedes threshold value And vehicle is appeared in matching the position corresponding to fraction highest in the scale layer；
A2) automobile judges whether the distance between the vehicle that detects and this car reduce, if then into step A3), if Otherwise return to step A1)；
A3) lane line is identified automobile, and whether the vehicle for judging this car and detecting is located in same lane line, If then sending early warning to driver, if otherwise carrying out step A4)；
A4) automobile is overtaken other vehicles warning step after the vehicle detected disappears out of rearview camera into front vehicles；
Front vehicles are overtaken other vehicles warning step, and automobile is by forward sight camera combination units of variance model algorithm to front vehicles It is identified, by recognizing the position relationship between front vehicles and Ben Che and lane line, judges whether this car needs progress Overtake other vehicles and send early warning to driver when overtaking other vehicles dangerous, be specially：
B1) whether automobile there are other cars according to units of variance model algorithm in the shooting image for detecting forward sight camera , if then entering step B2), if otherwise continuing to detect；
B2) automobile judges whether the distance between the vehicle that detects and this car reduce, if then into step B3), if Otherwise return to step B1)；
B3) lane line is identified automobile, and whether the vehicle for judging this car and detecting is located in same lane line, If then sending early warning to driver, if otherwise carrying out step B4)；
B4) automobile is overtaken other vehicles warning step after the vehicle detected disappears out of forward sight camera into front vehicle.
The detailed process of above-mentioned steps is as follows：
As shown in figure 3, being each module cloth of automobile for realizing automobile overtaking's method for early warning based on units of variance model algorithm In the middle of above office's figure, the installation of forward sight camera 302 and front windshield, the road conditions for collecting the 160 ° of visual angles in the front of vehicle 111 Information, rearview camera 304 is installed on middle above rear seat windscreen, the road conditions for collecting the 160 ° of visual angles in rear of vehicle 111 Information, the analysis of data processing module 303 calculates forward sight camera 302, and the image information of rearview camera 304 is computed identification Other vehicles and the lane line gone out in image, follows the trail of the vehicle in the visual field, due to overtaking other vehicles and lane change pair for other proximate vehicles When the safety traffic of vehicle 111 is impacted, data processing module 303 sends information, warning module 301 to warning module 301 Alarm is sent, and corresponding situation is reflected on LED early warning panels, warning module 301 is by buzzer and LED early warning panel groups Into being placed on console.The angular field of view of the camera is as shown in figure 4, in the present embodiment, forward sight camera and rear-camera The angular field of view of head is all 160 °, and collected video can pick out the vehicle and car in adjacent two track in effective range Diatom information.
Based on above-mentioned part, the automobile overtaking's method for early warning such as Fig. 5 institutes based on units of variance model algorithm finally proposed Show, mainly comprise the following steps：Data processing module 303 has detected vehicle and has sailed to come from rear by the information of rearview camera 304, To sailing the vehicle tracking come, then judge whether the distance of the vehicle and this vehicle shortens, if the vehicle is away from this vehicle, Abandon following the trail of.If the vehicle is shortened within safe distance with this vehicle distances, data processing module 303 is by recognizing lane line And the vehicle, judge the vehicle whether with this vehicle be in same lane line within, if this vehicle with the vehicle same Within lane line, then alarm module sends alarm to driver, reminds driver to be likely to occur front vehicle rear-end collision.If The vehicle not within same lane line, then continues to follow the trail of, when the vehicle disappears in rearview camera with this vehicle to the vehicle When in the visual field, illustrate that the vehicle is located within the blind area of this vehicle, warning module sends alarm, remind driver to there is vehicle to be located at , should not in this case lane change again within the blind area of this vehicle.The vehicle continues to move forward, and appears in forward sight camera 302 Within the visual field, digital signal processing module 303 detects the vehicle by forward sight camera 302, the vehicle is continued to track, if the car Away from this vehicle in adjacent lane, then stop following the trail of, if vehicle track where from adjacent lane to this car, in advance Alert module sends alarm to driver, reminds driver's front vehicles lane change, and noting slowing down avoids.
Above-mentioned is the scene that vehicle is surmounted by other vehicles, surmounts step under other vehicle scenes and is：Data processing mould Block 303 detects this vehicle front by forward sight camera 302 vehicle, to the vehicle tracking, if the vehicle and this vehicle away from It is big from becoming, stop to the vehicle tracking.If the vehicle shortens with this vehicle distances, data processing module 303 judges the vehicle Whether it is in this vehicle within same lane line.If within same lane line, warning module 301 is sent to driver Early warning, rear-end collision may be occurred by reminding in front of driver, and noting slowing down avoids, if the vehicle is located at adjacent lane, after Continue to the vehicle tracking, when the vehicle is disappeared in the visual field of forward sight camera 302, illustrate that the vehicle is located at the blind area of this vehicle Interior, warning module sends early warning to driver, and reminding in driver blind area has vehicle, and lane change is unable in the case where this asks money.This vehicle Surmount the vehicle to continue to move forward, when the vehicle is appeared in the visual field of rearview camera 304, to the car tracing, the car Away from after, stop to the car tracing.
In order to which camera acquisition to environmental data is corresponding with the real-world object in vehicle running environment, video camera is found Transformational relation in point coordinates and camera environment coordinate system in the image pixel coordinates system generated between object point coordinate, is needed Video camera is demarcated.This method uses video camera and laser radar combined calibrating, is swashed by extracting demarcation thing in single line Corresponding characteristic point carries out the demarcation of video camera external parameter on optical radar and image, so as to complete single line laser coordinate, take the photograph The unification of camera coordinate, image pixel coordinates sensor coordinates, realizes the spacial alignment of laser radar and video camera.
Video camera and laser radar combined calibrating, laser radar and video camera are to be rigidly connected all with automobile, each of which Relative attitude and displacement immobilize, in the same space, the scan data point of each laser radar exists in the picture One corresponding points of displacement.Therefore, by setting up rational laser radar coordinate system and camera coordinate system, laser radar is utilized The space constraint relation of scanning element and camera review, you can the spatial transform relation of Two coordinate system is solved, so as to complete laser The spacial alignment of radar and video camera, realizes associating for laser radar data and visible images.The external parameter of video camera leads to After Planar Mechanisms equation solution, laser radar, video camera, the relativeness just determination completely of image and versus environmental coordinate system, because This Laser Radar Scanning point can be projected to image pixel coordinates by camera model head.Its image element level data fusion can be by Following equation is completed：Wherein,For in video camera Portion's parameter matrix.As video camera and laser radar observation station P simultaneously, its coordinate in video camera itself environment coordinate system is Pvc(xvc,yvc,zvc), the coordinate of subpoint is U=(u v 1) in visible imagesTSeat in radar itself world coordinates It is designated as Plc(xlc,ylc,zlc).Because video camera and laser radar have used same environment coordinate system, then haveWherein H is the setting height(from bottom) of laser radar.On The formula simultaneous of face two, can be obtained：Wherein,By the demarcation of outer ginseng and shooting of laser radar The internal reference demarcation of machine can be obtainedAnd Xlc.To sum up, it is linear by solving by extracting enough image radar corresponding points pair Equation can obtain the coordinate spin matrix of correlationWith coordinate translation matrixAnd then it is right with its to can obtain laser radar data Answer the transformation relation between image pixel.
Units of variance model algorithm, is a kind of algorithm of target detection, and units of variance model uses PCA (PCA) Feature extraction is carried out after carrying out dimension-reduction treatment to HOG (gradient orientation histogram) feature；In model training, using Latent SVM (hidden variable SVMs) grader, in target detection, using the detection thought of sliding window；For many of target Viewing angle problem, using multicompartment strategy, sets up different models under different visual angles respectively；For the problem on deformation of target, use Partial model strategy based on graph structure.Specifically, units of variance model algorithm, using a kind of Star Model, the Star Model It is made up of the high-resolution partial model of widget in a root model and coverage goal substantially over whole target.Root Model definition detection window, partial model is placed on during resolution ratio is twice of characteristic layer of layer where root.
The feature of units of variance model algorithm employs HOG features, and some improvement have been carried out to HOG features.Variable portion Part model algorithm eliminates the block in former HOG features, retains cell, during normalization directly by current cell and around it 4 The region normalization that individual cell is constituted.When calculating gradient direction, symbol gradient (0-360 °) has been employed and without symbol The strategy that gradient (0-180 °) is combined.
The model of units of variance model algorithm, each component filters model by a root and multiple partial models are constituted.Fig. 6 (a) it is the root model of one of component and the effect of visualization of partial model with Fig. 6 (b).As Fig. 6 (a) root models are thicker It is rough, the side of a vehicle is substantially presented, such as Fig. 6 (b) partial models are the part in rectangle frame, totally 6 parts, partial model Resolution ratio be twice of root model.Fig. 6 (c) is the deformation loss of model, and brighter region representation deformation loss spends bigger, Circle center is the rationality position of partial model, if the position of the partial model detected is just here, that deformation is spent It is just 0, the more remote deformation of deviation spends bigger.For various visual angles problem, using multicompartment strategy, Fig. 6 (a) (b) (c) and Fig. 6 (d) (e) (f) is respectively the component of two different visual angles.
Units of variance model algorithm, in the context of detection of algorithm, units of variance model uses sliding window detection mode, leads to Structure yardstick pyramid is crossed to search in each yardstick.It is illustrated in figure 7 the matching of detection process under a certain yardstick, i.e. auto model Process.Regard model as a filter operator, response score is characterized the similarity degree with model to be matched, more similar then score It is higher.As shown in fig. 7, left side is the testing process of root model, brighter Regional Representative's response score is higher.Right side is each part The detection process of model.First, characteristic image and model match obtaining filtered image；Then, mutually strained Change：With anchor point (i.e. top left co-ordinate) for reference position, comprehensive partial model is relative with the matching degree and partial model of feature to be managed Think that the deformation of position spends (deviation loss), obtain optimal partial model position and phase reserved portion.It is l in yardstick0Layer, with (x0, y0) be anchor point detection score, shown in equation below： Wherein,For the detection fraction of root model.Because same target has multiple components, and the inspection of different component models Surveying fraction needs alignment, so needing deviation ratio b.For the response of i-th of partial model, due to The resolution ratio of partial model is one times of root model, therefore partial model needs are in scale layer l0- λ is matched.Therefore, the seat of anchor point Mark is also required to be remapped to scale layer l0- λ, that is, be exaggerated one times of 2 (x0,y0), partial model i is relative to (the x of anchor point 20,y0) Offset vi, so in scale layer l0- λ, partial model i ideal position are 2 (x0,y0)+vi。
Hough is converted in straight-line detection, and each pixel coordinate point is for conversion into the collection of series of discrete point by hough Close.By the discrete polar coordinates formula of a straight line, the discrete point geometric equality that can give expression to straight line is as follows：R=x cos θ+ Y sin θs, wherein angle, θ refer to the angle between r and x-axis, and r is to rectilinear geometry vertical range.We are according to pixel point coordinates The value of (x, y) is drawn each (r, θ), then be transformed into polar coordinates hough transformation systems, this bear from image cartesian coordinate system The change of point to curve is referred to as the hough conversion of straight line.Conversion is limited value interval decile by quantifying hough parameter spaces Or cumulative grid.Hough transform algorithm starts, and each pixel coordinate point (x, y) is switched to above the curve point of (r, θ), tires out Corresponding grid data point is added to, when a crest occurs, is illustrated with the presence of straight line.
The detection process that hough becomes scaling method is as follows：First a Two-dimensional Counting device R (r, θ), r are set up in parameter space Scope be 0 length for arriving image diagonal, θ scope is that all values in 0 to 2 π, array are initialized as 0；Then, scan All pixels point (x, y) in image space, hough transforms carry out image space to the conversion (r, θ) of parameter space, and count Number device R (r, θ) Jia 1；3rd step, given threshold thr (r, θ) judges how many point collinearly just thinks exist directly in image Line, R (r, θ) is more than thr (r, θ), then constitutes the image in image.
Fig. 8 is that have 6 LEDs on LED early warning panel schematic diagrames, panel, when other vehicles appear in blind area on the left of this car When LED 401 flash；When other vehicles appear in this car right side blind area, LED 406 flashes；When having other vehicles and this car In same track and positioned at this tailstock portion, when the vehicle and this vehicle distances are too small, LED 404 flashes；When have other vehicles with This car is in same track and positioned at this front side, and when the vehicle and this vehicle distances are too small, LED 403 flashes；Left forward side car Vehicle lane change with this car to when unifying track in road, and LED 402 flashes；Vehicle lane change is united to this car in the track of right forward side During one track, LED 405 flashes.
In summary, a kind of automobile overtaking's method for early warning based on units of variance model algorithm is present embodiments provided, it is first The image of forward sight camera and rearview camera collection is first passed through, data processing module is trained, it is used and trains The model gone out can identify vehicle and lane line in actual scene, and vehicle is tracked, so as to judge other vehicles In the lane change of this vehicle periphery and overtake other vehicles etc. the correlation circumstance of danger may be caused to this vehicle, and correspondence scene is sent to driver Early warning, so as to enable a driver to evade ahead of time the generation of related accidents.

Claims (10)

1. a kind of automobile overtaking's method for early warning based on units of variance model algorithm, it is characterised in that methods described includes following Step：
Front vehicle is overtaken other vehicles warning step, and automobile is carried out by rearview camera combination units of variance model algorithm to front vehicle Identification, by recognizing the position relationship between front vehicle and Ben Che and lane line, judges whether front vehicle is overtaken other vehicles And send early warning to driver；
Front vehicles are overtaken other vehicles warning step, and automobile is carried out by forward sight camera combination units of variance model algorithm to front vehicles Identification, by recognizing the position relationship between front vehicles and Ben Che and lane line, judges whether this car needs to be overtaken other vehicles And send early warning to driver when overtaking other vehicles dangerous.
2. automobile overtaking's method for early warning according to claim 1 based on units of variance model algorithm, it is characterised in that institute Stating front vehicle warning step of overtaking other vehicles is specially：
A1) whether automobile there are other vehicles according to units of variance model algorithm in the shooting image for detecting rearview camera, if It is then to enter step A2), if otherwise continuing to detect；
A2) automobile judges whether the distance between the vehicle that detects and this car reduce, if then into step A3), if otherwise Return to step A1)；
A3) lane line is identified automobile, and whether the vehicle for judging this car and detecting is located in same lane line, if Early warning then is sent to driver, if otherwise carrying out step A4)；
A4) automobile is overtaken other vehicles warning step after the vehicle detected disappears out of rearview camera into front vehicles.
3. automobile overtaking's method for early warning according to claim 1 based on units of variance model algorithm, it is characterised in that institute Stating front vehicles warning step of overtaking other vehicles is specially：
B1) whether automobile there are other vehicles according to units of variance model algorithm in the shooting image for detecting forward sight camera, if It is then to enter step B2), if otherwise continuing to detect；
B2) automobile judges whether the distance between the vehicle that detects and this car reduce, if then into step B3), if otherwise Return to step B1)；
B3) lane line is identified automobile, and whether the vehicle for judging this car and detecting is located in same lane line, if Early warning then is sent to driver, if otherwise carrying out step B4)；
B4) automobile is overtaken other vehicles warning step after the vehicle detected disappears out of forward sight camera into front vehicle.
4. automobile overtaking's method for early warning based on units of variance model algorithm according to Claims 2 or 3, its feature exists In the units of variance model algorithm is specially：
A11) according to the shooting image of rearview camera, yardstick pyramid is built；
A12) in each scale layer in yardstick pyramid, by the way that sliding window detection method is by image and passes through units of variance Model after model algorithm training is matched, and calculates matching fraction；
A13 the corresponding scale layer of highest matching fraction) is chosen, shows to detect vehicle and car if the matching fraction exceedes threshold value Appear in matching the position corresponding to fraction highest in the scale layer.
5. automobile overtaking's method for early warning according to claim 4 based on units of variance model algorithm, it is characterised in that institute Stating matching fraction is specially：
<mrow> <mi>s</mi> <mi>c</mi> <mi>o</mi> <mi>r</mi> <mi>e</mi> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>,</mo> <msub> <mi>y</mi> <mn>0</mn> </msub> <mo>,</mo> <msub> <mi>l</mi> <mn>0</mn> </msub> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>R</mi> <mrow> <mn>0</mn> <mo>,</mo> <msub> <mi>l</mi> <mn>0</mn> </msub> </mrow> </msub> <mrow> <mo>(</mo> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>,</mo> <msub> <mi>y</mi> <mn>0</mn> </msub> <mo>)</mo> </mrow> <mo>+</mo> <munderover> <mo>&amp;Sigma;</mo> <mrow> <mi>i</mi> <mo>=</mo> <mn>1</mn> </mrow> <mi>n</mi> </munderover> <msub> <mi>D</mi> <mrow> <mi>i</mi> <mo>,</mo> <msub> <mi>l</mi> <mn>0</mn> </msub> <mo>-</mo> <mi>&amp;lambda;</mi> </mrow> </msub> <mrow> <mo>(</mo> <mn>2</mn> <mo>(</mo> <mrow> <msub> <mi>x</mi> <mn>0</mn> </msub> <mo>,</mo> <msub> <mi>y</mi> <mn>0</mn> </msub> </mrow> <mo>)</mo> <mo>+</mo> <msub> <mi>v</mi> <mi>i</mi> </msub> <mo>)</mo> </mrow> <mo>+</mo> <mi>b</mi> </mrow>
Wherein, score is matching fraction, (x0,y0) it is anchor point, l0For the yardstick number of plies,For the detection fraction of root model,For the response of i-th of partial model, viOffset for i-th of partial model relative to anchor point, b is deviation ratio.
6. automobile overtaking's method for early warning according to claim 4 based on units of variance model algorithm, it is characterised in that institute Stating the model after being trained by units of variance model algorithm is specially：The HOG features for training picture are dropped by using PCA methods Feature extraction is carried out after dimension processing, then model training is carried out by hidden variable support vector machine classifier, is obtained by variable portion Model after the training of part model algorithm.
7. automobile overtaking's method for early warning based on units of variance model algorithm according to Claims 2 or 3, its feature exists In the automobile judges that the distance between vehicle and this car detected is specially：Pass through the laser radar being installed on automobile Carry out ranging, the distance between the vehicle detected and this car.
8. automobile overtaking's method for early warning based on units of variance model algorithm according to Claims 2 or 3, its feature exists In lane line is identified specially the automobile：Automobile is by Hough transform, the collinear points in scan image, realizes pair The identification of lane line.
9. automobile overtaking's method for early warning according to claim 1 based on units of variance model algorithm, it is characterised in that institute Stating early warning includes realizing early warning by LED early warning panel flash and realizing early warning by buzzer rings.
10. automobile overtaking's method for early warning according to claim 9 based on units of variance model algorithm, it is characterised in that The LED early warning panel is provided with the LED early-warning lamps for being at least distributed in four direction all around.
CN201710318142.6A 2017-05-08 2017-05-08 A kind of automobile overtaking's method for early warning based on units of variance model algorithm Pending CN107229906A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710318142.6A CN107229906A (en) 2017-05-08 2017-05-08 A kind of automobile overtaking's method for early warning based on units of variance model algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710318142.6A CN107229906A (en) 2017-05-08 2017-05-08 A kind of automobile overtaking's method for early warning based on units of variance model algorithm

Publications (1)

Publication Number Publication Date
CN107229906A true CN107229906A (en) 2017-10-03

Family

ID=59933777

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710318142.6A Pending CN107229906A (en) 2017-05-08 2017-05-08 A kind of automobile overtaking's method for early warning based on units of variance model algorithm

Country Status (1)

Country Link
CN (1) CN107229906A (en)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107985189A (en) * 2017-10-26 2018-05-04 西安科技大学 Towards driver's lane change Deep Early Warning method under scorch environment
CN108714304A (en) * 2018-04-02 2018-10-30 网易（杭州）网络有限公司 Overtake other vehicles display methods and device in a kind of game
CN108847026A (en) * 2018-05-31 2018-11-20 安徽四创电子股份有限公司 A method of it is converted based on matrix coordinate and realizes that data investigation is shown
CN110509880A (en) * 2019-09-24 2019-11-29 上海为彪汽配制造有限公司 Automobile rear blind monitoring system and method, radar control box
CN110525333A (en) * 2018-05-23 2019-12-03 上海擎感智能科技有限公司 A kind of passing behavior monitoring method and system, car-mounted terminal based on car-mounted terminal
CN110588518A (en) * 2019-09-24 2019-12-20 上海为彪汽配制造有限公司 Automobile rear detection system and method and radar control box
CN110775057A (en) * 2019-08-29 2020-02-11 浙江零跑科技有限公司 Lane assisting method for analyzing and controlling steering torque based on vehicle-mounted blind zone visual scene
CN112721931A (en) * 2021-01-18 2021-04-30 智马达汽车有限公司 Vehicle meeting method, device, equipment and storage medium
CN114023109A (en) * 2021-11-03 2022-02-08 中国矿业大学 Early warning system for preventing rear-end collision of mobile tank car

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103295021A (en) * 2012-02-24 2013-09-11 北京明日时尚信息技术有限公司 Method and system for detecting and recognizing feature of vehicle in static image
CN103587467A (en) * 2013-11-21 2014-02-19 中国科学院合肥物质科学研究院 Dangerous-overtaking early-warning prompting method and system
CN106314276A (en) * 2016-09-22 2017-01-11 西华大学 Rear car overtaking reminding system based on distance measuring

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103295021A (en) * 2012-02-24 2013-09-11 北京明日时尚信息技术有限公司 Method and system for detecting and recognizing feature of vehicle in static image
CN103587467A (en) * 2013-11-21 2014-02-19 中国科学院合肥物质科学研究院 Dangerous-overtaking early-warning prompting method and system
CN106314276A (en) * 2016-09-22 2017-01-11 西华大学 Rear car overtaking reminding system based on distance measuring

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
李春伟 等: "一种基于可变形部件模型的快速对象检测算法", 《电子与信息学报》 *

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107985189A (en) * 2017-10-26 2018-05-04 西安科技大学 Towards driver's lane change Deep Early Warning method under scorch environment
CN108714304A (en) * 2018-04-02 2018-10-30 网易（杭州）网络有限公司 Overtake other vehicles display methods and device in a kind of game
CN110525333A (en) * 2018-05-23 2019-12-03 上海擎感智能科技有限公司 A kind of passing behavior monitoring method and system, car-mounted terminal based on car-mounted terminal
CN108847026A (en) * 2018-05-31 2018-11-20 安徽四创电子股份有限公司 A method of it is converted based on matrix coordinate and realizes that data investigation is shown
CN110775057A (en) * 2019-08-29 2020-02-11 浙江零跑科技有限公司 Lane assisting method for analyzing and controlling steering torque based on vehicle-mounted blind zone visual scene
CN110509880A (en) * 2019-09-24 2019-11-29 上海为彪汽配制造有限公司 Automobile rear blind monitoring system and method, radar control box
CN110588518A (en) * 2019-09-24 2019-12-20 上海为彪汽配制造有限公司 Automobile rear detection system and method and radar control box
CN112721931A (en) * 2021-01-18 2021-04-30 智马达汽车有限公司 Vehicle meeting method, device, equipment and storage medium
CN114023109A (en) * 2021-11-03 2022-02-08 中国矿业大学 Early warning system for preventing rear-end collision of mobile tank car

Similar Documents

Publication Publication Date Title
CN107229906A (en) A kind of automobile overtaking&#39;s method for early warning based on units of variance model algorithm
CN110517521B (en) Lane departure early warning method based on road-vehicle fusion perception
CN110077399B (en) Vehicle anti-collision method based on road marking and wheel detection fusion
CN110065494B (en) Vehicle anti-collision method based on wheel detection
KR102359806B1 (en) Control of the host vehicle based on the detected parking vehicle characteristics
Wu et al. Applying a functional neurofuzzy network to real-time lane detection and front-vehicle distance measurement
CN103630122B (en) Monocular vision lane line detection method and distance measurement method thereof
CN110356325B (en) Urban traffic passenger vehicle blind area early warning system
CN106324618B (en) Realize the method based on laser radar detection lane line system
CN102685516A (en) Active safety type assistant driving method based on stereoscopic vision
CN107031623A (en) A kind of road method for early warning based on vehicle-mounted blind area camera
CN109733283B (en) AR-based shielded barrier recognition early warning system and recognition early warning method
CN105678787A (en) Heavy-duty lorry driving barrier detection and tracking method based on binocular fisheye camera
US20100100268A1 (en) Enhanced clear path detection in the presence of traffic infrastructure indicator
CN105620489A (en) Driving assistance system and real-time warning and prompting method for vehicle
CN104573646A (en) Detection method and system, based on laser radar and binocular camera, for pedestrian in front of vehicle
CN104129389A (en) Method for effectively judging and recognizing vehicle travelling conditions and device thereof
CN103455144A (en) Vehicle-mounted man-machine interaction system and method
CN110126824A (en) A kind of commercial vehicle AEBS system of integrated binocular camera and millimetre-wave radar
CN108528337B (en) Intelligent inner wheel difference early warning system and early warning method based on video projection
CN102778223A (en) License number cooperation target and monocular camera based automobile anti-collision early warning method
EP2293588A1 (en) Method for using a stereovision camera arrangement
CN104200689A (en) Road early warning method and device
CN103204104B (en) Monitored control system and method are driven in a kind of full visual angle of vehicle
CN110816527A (en) Vehicle-mounted night vision safety method and 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
RJ01 Rejection of invention patent application after publication

Application publication date: 20171003

RJ01 Rejection of invention patent application after publication