Automatic vertical parking system and method based on multisection type planning and machine learningTechnical field
The present invention relates to the automatic parking planning fields of intelligent automobile, more particularly, to one kind based on multisection type planning and machineThe automatic vertical parking system and method for device study.
Background technique
The unbalanced development of domestic automobile ownership and Transportation Infrastructure Construction in recent years, so that parking space is increasingly narrowSmall, technical requirements of parking are higher and higher.The limitation of parking space, it is desirable that vehicle should stop according to parking stall line gauge model, can make to parkSpace resources obtains maximized rationally utilization, is also beneficial to the overall planning of parking lot and the appearance of the city.On the other hand, parking spaceLimitation, be equally the great challenge to driver's technology, problem of parking expends the great time and efforts of people, produces when seriousThe accidents such as raw scraping, collision.Thus, fan and expectation of the automatic parking technology by market.
Traditional automated parking system, method for planning track generally consider to use R-S curve, helix, spline curveDeng one of carry out Global motion planning, and only consider the constraint of theoretic vehicle kinematics.Planning side based on R-S curveMethod, track generate clear thinking, calculate simply, to the adaptable of environment, the parking space needed is small.But R-S curve is depositedIn the discontinuous problem of curvature so that vehicle need to repeatedly stop after adjustment direction disk corner, it is serious to tire wear.Phase therewithInstead, use the curves such as helix, spline curve as track of parking, the discontinuous problem of curvature is not present, but be needed poolVehicle space is larger, and high to initial pose requirement of parking, and is not suitable for directly applying in present environment of parking.Although in addition,Speed is lower when parking, and tyre slip angle is small, and vehicle lateral sliding is few.But, it is understood that there may be roadside parking stall inclination turns toThere is very big tracking error often in the situations such as motor tracking delay, the simple track for controlling vehicle tracking planning.
Therefore, how to provide a kind of parking strategy to solve the above problems and system is that those skilled in the art are urgently to be resolvedThe problem of.
Summary of the invention
It is an object of the present invention to overcome the above-mentioned drawbacks of the prior art and provide one kind to be advised based on multisection typeDraw the automatic vertical parking system and method with machine learning.
The purpose of the present invention can be achieved through the following technical solutions:
A kind of automatic vertical parking system based on multisection type planning and machine learning, the system include:
Sensing module: including looking around camera and ultrasonic radar, described looks around camera for acquiring image, identifiesWarehouse compartment line simultaneously judges opposite warehouse compartment posture information using binocular ranging method, and the ultrasonic radar is for detecting barrier letterBreath, judges whether warehouse compartment is occupied；
Decision-making module: according to the fuse information of sensing module and instruction record, judge warehouse compartment validity and current institutePlace parks the stage, and provides posture information for planning parking path for planning module；
Planning module: the pose according to warehouse compartment information and from vehicle with respect to warehouse compartment plans the track R-S or sends out posture informationMachine study module is given, after obtaining planning path, motion-control module is sent to and is tracked；
Machine learning module: for industrial computer, determine that starting point pose of vertically parking is defeated by machine learning methodEnter the relationship with secondary spiral line parameter, and the helix planned track of parking is sent to planning module；
Motion-control module: include ECU and line control system, receive the track scatterplot that planning module transmits, pass through controllerSteering wheel, gear, throttle and brake pedal are controlled, and controls vehicle tracking planning path.
The camera of looking around is CMOS camera, and screen pixels are having a size of 640:480.
It is described look around camera and ultrasonic sensor be equipped with it is multiple, including being arranged in vehicle body all around fourCamera and and vehicle body around 12 ultrasonic sensors, left and right sides looks around camera and is mounted below rearview mirror, vehicle body4 ultrasonic radars are respectively arranged in front rear, and 2 ultrasonic radars are each side arranged.
It is a kind of to be parked method based on multisection type planning and the automatic vertical of machine learning, comprising the following steps:
1) when driver control after parking vehicles to park warehouse compartment periphery after, open automatic parking mode；
2) the slow straight-line travelling of moving control module for controlling vehicle；
3) by around vehicle body look around camera and ultrasonic radar carries out warehouse compartment angular coordinate and limited block positionDetection obtains warehouse compartment regional location and size, judges whether the warehouse compartment meets the requirement on parking stall, determine parking stall, and carry outStep 4), otherwise, return step 2)；
4) according to the coordinate and current pose from vehicle with respect to warehouse compartment, parking path is planned；
5) by first segment park curve whether be able to achieve collisionless storage, judge whether to need multistage path planning, if so,Step 6) is then carried out, if it is not, then carrying out step 7)；
6) multistage path planning adjusts to obtain using multistage R-S curve, specifically includes the following steps:
61) premised on vehicle rear axle right side is without impinging on warehouse compartment angle point, the first segment straight line rail of first segment R-S curve is determinedMark；
62) vehicle beats to the right steering wheel to extreme position, and moves backward to right back, when to be located at warehouse compartment left for vehicle left back pointStop when on side line or its extended line；
63) vehicle beats steering wheel to extreme position to the left, and advances to left front, until vehicle pose is adjusted to lead toSecondary spiral line tracking is crossed to realize the angle for storage of once parking or be located in safe distance at a distance from front obstacle；
64) basis looks around camera and ultrasonic sensor and detects warehouse compartment coordinate again, redefines warehouse compartment in conjunction with boat positionInformation and from parking stall appearance, eliminates the error of boat position, and return step 4)；
7) according to current vehicle location coordinate and course angle information, using based on the machine learning network of secondary spiral lineStorage track is sought in calculation；
8) motion-control module is according to storage track, control vehicle storage, end of parking, and exits park mode.
In the step 3), when detecting multiple warehouse compartments, the preferential selection warehouse compartment nearest apart from this vehicle, and judgement isIt is no to meet the requirements, if not meeting, selects and judge next warehouse compartment.
In the step 3), warehouse compartment meets the requirement on parking stall while meeting the following conditions:
Warehouse compartment type matching, warehouse compartment size be appropriate and warehouse compartment in barrier is not present.
The step 7) specifically includes the following steps:
71) according to the coordinate range of setting and course angular region, the secondary spiral line tracking ginseng under each initial state is obtainedNumber；
72) initial coordinate values and course angle and corresponding secondary spiral line parameter, training learning network parameter are utilized；
73) using the coordinate information of current vehicle and course angle as input, the learning network after training is inputted, storage is obtainedTrace information, be sent to motion-control module.
When driver by vehicle parking to warehouse compartment position when, open automatic driving mode:
When the warehouse compartment fuse information that sensing module is sent has not been obtained in decision-making module, planning module sends straight line and plans roadDiameter controls vehicle low speed and moves ahead, sensing module continues to detect to motion-control module；
After detecting available warehouse compartment, judge whether warehouse compartment is available and obtains warehouse compartment type, and judgement is current by decision-making moduleWhich kind of vehicle is in and parks the stage, and planning module is according to this phase paths of warehouse compartment information planning, and transmitting path scatterplot extremely movesControl module is tracked；
When automatic parking mode ends, decision-making module completes parking backed off after random automatic Pilot mould by control brake pedalFormula.
Compared with prior art, the invention has the following advantages that
One, adaptable, reliable and stable: the present invention provides the rule of parking of a kind of combination R-S curve and secondary spiral lineThe method of drawing reduces process original place adjustment direction number of parking, suitable for the narrow environment of parking in city incity, to the adaptability of environmentBy force, it is short to calculate the time, and can be adjusted when there is tracking offset, process of parking is reliable and stable.
Two, automatic parking: driver only needs to open parking system switch in suitable position in the present invention, and system can be certainlyDynamic identification warehouse compartment sideline simultaneously judges the information such as warehouse compartment type, relative position and validity, and hereafter planning acts, participate in without the mankindJudgement；
Three, it improves efficiency: after parking system obtains warehouse compartment information in the present invention, can completely automatically carry out planning of parkingAnd motion control, will park process programming and automation, improve people's Working Life efficiency.
Four, planning is easy: the present invention generates spiral trajectory using the method for machine learning, improve track formation speed andSuccess rate keeps planning process more easy.
Detailed description of the invention
Fig. 1 is the flow chart of automatic parking method process of the invention.
Fig. 2 is the path multisection type R-S of the present invention schematic diagram.
Fig. 3 is the structural schematic diagram for the combination combined positioning method that the present invention uses.
Fig. 4 is the structural schematic diagram of automated parking system of the invention.
Fig. 5 is machine learning schematic network structure of the present invention.
Specific embodiment
The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.
Embodiment
Fig. 1 is the flow chart of automatic parking method process of the invention, and this method specific steps include:
Step 1: being controlled by driver to warehouse compartment periphery of parking, open automatic parking mode, parking system adapter tube vehicleControl；
Step 2: module of parking controls the slow straight-line travelling of vehicle；
Step 3: by looking around camera and ultrasonic sensor progress warehouse compartment angular coordinate and limited block around vehicle bodyThe detection of position calculates warehouse compartment regional location and size, and detects barrier presence or absence within warehouse compartment region；If warehouse compartment regionShape size meets the requirements, and barrier is not present, and determines parking stall, enters step 4；Otherwise, step 2 is returned to；
If selecting the warehouse compartment nearest apart from this vehicle preferably, detecting multiple warehouse compartments in step 3, judging whether to meetStandard；If not meeting, reselection judges next warehouse compartment.
Whether judgment criteria whether judging warehouse compartment properly in step 3 includes depositing in warehouse compartment type, warehouse compartment size and warehouse compartmentIn barrier.
As shown in Fig. 2, preferably, first planning the track (midpoint Fig. 21 to 4 sections of tracks of point) based on R-S curve.
Step 4: according to the coordinate and current pose from vehicle with respect to warehouse compartment, planning parking path；
Step 5: by first segment park curve whether be able to achieve collisionless storage, judge whether to need multistage path planning,If so, entering step 6；If it is not, entering step 7；
Preferably, step 6 is as shown in Fig. 2, the specific steps adjusted are as follows: is adjusted using multistage R-S curve, adjustment sideMethod (for parking to the right) is as follows:
Step 61: premised on the right side of vehicle rear axle without impinging on warehouse compartment angle point, determining first segment R-S curve first segment straight lineTrack (midpoint Fig. 21 to point 2 sections)；
Step 62: vehicle beats to the right steering wheel to extreme position, moves backward afterwards to the right, until vehicle left back point is on the left of warehouse compartmentOn line or its extended line (midpoint Fig. 22 to point 3 sections)；
Step 63: vehicle beats steering wheel to extreme position to the left, advances to left front, until vehicle pose is adjusted to be easy to logicalIt crosses secondary spiral line tracking and realizes the angle (it is stored in secondary spiral line tracking cluster) for storage of once parking or to apart from frontAt barrier safe distance (midpoint Fig. 23 to point 4 sections)；
Step 64: detecting warehouse compartment coordinate again with ultrasonic sensor using looking around, redefine library in conjunction with dead reckoningPosition information and from parking stall appearance, eliminates the error of dead reckoning, returns to step 4；
Again opposite warehouse compartment is detected in step 64 and sits calibration method, as shown in figure 3, needing to merge secondary positioning, boat position pushes awayThe location informations such as calculation, inertial navigation, such as in outdoor, also it is contemplated that fusion GPS information judges from parking stall appearance.
Step 7: according to current vehicle location coordinate and course angle information (coordinate and course angle at the midpoint Fig. 2 4), utilizingMachine learning network query function based on secondary spiral line seeks storage track, specific steps are as follows: learning network structure such as Fig. 5 instituteShow, machine learning network training method is as follows:
Step 71: according to expected coordinate range and course angular region, calculating suitable secondary spiral under each initial stateLine tracking parameter；
Step 72: utilizing initial state information (coordinate value and course angle) and corresponding secondary spiral line parameter, trainingLearning network parameter；
Step 73: using the coordinate information of current vehicle and course angle as input, being carried out by the learning network after trainingIt calculates, the trace information that is put in storage (midpoint Fig. 24 to point 5 sections) is sent to control module.
Step 8: control module controls vehicle storage according to expected trajectory, and park mode is exited in end of parking.
Preferably, it parks in step 8, under automatic driving mode the specific steps of control are as follows:
When driver by vehicle parking to rational position when, automatic driving mode is opened in selection；
When the warehouse compartment fuse information that sensing module is sent has not been obtained in decision-making module, planning module sends straight line and plans roadDiameter, control vehicle low speed move ahead, and sensing module continues to detect；
After detecting available warehouse compartment, judge whether warehouse compartment available and warehouse compartment type by decision-making module, and judge to work as front truckWhich kind of park the stage in, planning module is according to this phase paths of warehouse compartment information planning, and transmitting path scatterplot is to controlling mouldBlock is tracked；
When automatic parking mode ends, decision-making module completes parking backed off after random automatic Pilot mould by control brake pedalFormula.
As shown in figure 4, Fig. 4 is a kind of structural schematic diagram of automated parking system provided by the invention, including sensing module,Decision-making module, planning module, machine learning module and motion-control module:
Sensing module, including camera and ultrasonic radar are looked around, camera is looked around for acquiring image, identifies warehouse compartment lineAnd opposite warehouse compartment posture information is judged using binocular ranging method, ultrasonic radar judges warehouse compartment for detecting obstacle informationIt is whether occupied.
Decision-making module, according to the fuse information of sensing module and instruction record, judge warehouse compartment validity and current institutePlace parks the stage, is supplied to planning module posture information for planning parking path；
Planning module, the pose according to warehouse compartment information and from vehicle with respect to warehouse compartment plan the track R-S or send out posture informationMachine study module is given, after obtaining planning path, control module is sent to and is tracked；
Machine learning module is industrial computer, determines starting point pose input of vertically parking by machine learning methodWith the relationship of secondary spiral line parameter, and the helix planned track of parking is sent to planning module；
Motion-control module includes ECU and line control system, receives the track scatterplot that planning module transmits, passes through controller controlSteering wheel, gear, throttle and brake pedal processed control vehicle tracking planning path.
Preferably, machine learning module preferentially uses the learning method of neural network, and structure is as shown in figure 5, specificInclude:
Normalization unit is inputted, the vehicle posture information of input is normalized, convenient for calculating；
Neural network unit calculates output according to input according to machine learning method；
Track generation unit calculates spiral trajectory point coordinate according to the helix parameter of output, generates track scatterplot.
The present invention provides a kind of automatic vertical parking systems, look around camera acquisition figure by what vehicle body surrounding was arrangedPicture, ultrasonic sensor detect the validity information that obstacle information judges the position of opposite warehouse compartment, course information and warehouse compartment.WhenRecognize suitable size and without vehicle occupy warehouse compartment when, into automatic parking process.Automated parking system is according to currently from vehiclePose and warehouse compartment information carry out the planning of parking path, are adjusted if needed using multistage R-S curve from parking stall appearance to suitable positionIt sets, generates secondary spiral line parameter further according to learning network and calculate track of parking.Finally, by electric control gear control steering wheel,Throttle and brake pedal carry out storage of parking, and can voluntarily judge warehouse compartment type and validity, realize the vertical pool in full-automatic groundVehicle.
The above is only the preferred embodiment of the present invention, it is noted that those skilled in the art are comeIt says, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications also should be regarded asProtection scope of the present invention.