A kind of intelligent barrier avoiding system and method for automatic driving vehicle
Technical field
The present invention relates to a kind of intelligent barrier avoiding system and methods of automatic driving vehicle.
Background technique
With the development of human society, for vehicle at means of transport important in human production life, vehicle is very big
Accelerate social development.Recent years, with the development of artificial intelligence technology and sensor technology, automatic Pilot technology is increasingly
Maturation, increasingly towards commercialized development.But due to the limitation of technology and sensor, to realize it is completely obstructed it is remarkable it is complete from
Up to the present dynamic driving can not be achieved substantially.In fact, to realize so-called automatic Pilot be expected to closing, it is semi-enclosed
Scene takes the lead in realizing.Port environment belongs to closing, semiclosed scene, drives using the guiding vehicle of automatic Pilot technology compared to manpower
It sails and higher efficiency may be implemented, and cost can be reduced.But the automatic guided vehicle under current harbour scene can not achieve
Complete automatic obstacle avoiding, there is the not high situations of conevying efficiency.
Summary of the invention
The purpose of the present invention is to solve the automatic guided vehicles under existing harbour scene can not achieve completely
Automatic obstacle avoiding has that conevying efficiency is not high, and proposes a kind of intelligent barrier avoiding system and method for automatic driving vehicle.
A kind of intelligent barrier avoiding system of automatic driving vehicle, composition includes: the obstacle being mounted on automatic driving vehicle
Analyte detection device, barrier trajectory predictions subsystem, Vehicle Decision Method subsystem, dataset acquisition device;
Obstacle detector, for detecting other automatic driving vehicles, using the automatic driving vehicle detected as barrier
Hinder object, and the data detected are handled, obtains coordinate of the barrier vehicle relative to sensor;
Dataset acquisition device, for acquiring the information of barrier vehicle driving trace, and by vehicle under harbour scene
The track of operation keeps a record and stores;Using collected track of vehicle as the training set of trajectory predictions algorithm, train several times,
Trained barrier trajectory predictions algorithm model is saved;
Barrier trajectory predictions subsystem, the barrier vehicle coordinate for detecting obstacle detector are passed to number
According to the barrier trajectory predictions algorithm model that collection acquisition device trains, and barrier is predicted according to the historical track of barrier
Future Trajectory;
Vehicle Decision Method subsystem, for according to barrier trajectory predictions algorithm model predict come Future Trajectory determine
Plan is demarcated according to the position that the Future Trajectory of vehicle is likely to occur, and may occupied position around this.
A kind of intelligent barrier avoiding method of automatic driving vehicle, the described method comprises the following steps:
Step 1: acquisition data:
Each sensor of application data set acquisition device obtains in barrier vehicle travel process, and barrier is in harbour scene
Under relative to collecting work vehicle left and right offset x and front and back offset y driving trace information, by the barrier track number of acquisition
According to being stored;
Step 2: to the pond layer of neural networkInput data is handled, whereinFor letter
Number, W is weight;
Step 3: the data after the processing of pond layer are passed in Recognition with Recurrent Neural Network,
Step 4: being again introduced into the processing of pond layer, formula is handled are as follows:
Step 5: incoming data to be changed into the form of Two dimension normal distribution parameter:
Step 6: setting loss function, loss function areIt is calculated with optimization
Method minimizes loss function;
Step 7: carrying out the training of Recognition with Recurrent Neural Network:
The parameter of Recognition with Recurrent Neural Network is optimized, training result is saved;
Step 8: trajectory predictions are carried out using trained Recognition with Recurrent Neural Network, according to avoidance principle, by the traveling of prediction
Track is added in trajectory planning, specifically:
1), in computer storage unitOperationTrajectory predictions algorithm;
2) then, the trained training result of load step seven inputs data into trained circulation mind by sensor
Through network, the pond layer of centre experience and the formula in training process and principle are same as above, it may be assumed that
3), it is by the new track of Two dimension normal distribution output predictionIt indicates are as follows:
Wherein, step 1 dataset acquisition method specifically:
The vehicle for carrying out data acquisition is equipped with visual sensor and laser sensor, and application target detection algorithm identifies vehicle
And vehicle is tracked;It is driven at a constant speed by driver driving vehicle in harbour with the speed of 20-60km/h, by several hours
Afterwards, sensor gets the track data of vehicle times a series of, is stored for the use of trajectory predictions algorithm.
The invention has the benefit that
The present invention on automatic driving vehicle by installing obstacle detector, barrier trajectory predictions subsystem, vehicle
The barrier vehicle coordinate of decision-making subsystem, dataset acquisition device, obstacle detector detection is passed to data centralized procurement
The barrier trajectory predictions algorithm model that acquisition means train, and predict according to the historical track of barrier the future of barrier
Track;Again by Vehicle Decision Method subsystem according to barrier trajectory predictions algorithm model predict come Future Trajectory carry out decision,
Decision making algorithm is demarcated according to the position that the Future Trajectory of vehicle is likely to occur, and may occupied position around this.
Design of the invention can carry out the better avoidance of effect to barrier vehicle.The system can reduce automatic Pilot vehicle
Collision probability between increases the traffic efficiency of automatic driving vehicle under harbour scene.
Detailed description of the invention
Fig. 1 is the intelligent barrier avoiding system principle diagram of automatic driving vehicle of the present invention;
Fig. 2 is the intelligent barrier avoiding method flow diagram of automatic driving vehicle of the present invention;
Fig. 3 is avoidance schematic diagram of the present invention;
Fig. 4 is trajectory predictions algorithm flow chart of the present invention.
Specific embodiment
Specific embodiment 1:
The intelligent barrier avoiding system of a kind of automatic driving vehicle of present embodiment, as shown in Figure 1, its composition includes: installation
Obstacle detector 1, barrier trajectory predictions subsystem 2, Vehicle Decision Method subsystem 3 on automatic driving vehicle, data
Collect acquisition device 4;
Obstacle detector 1, for detecting other automatic driving vehicles, using the automatic driving vehicle detected as barrier
Hinder object, and the data detected are handled, obtains coordinate of the barrier vehicle relative to sensor;
Dataset acquisition device 4, for acquiring the information of barrier vehicle driving trace, and by vehicle under harbour scene
The track of operation keeps a record and stores;Using collected track of vehicle as the training set of trajectory predictions algorithm, train several times,
Trained barrier trajectory predictions algorithm model is saved;(acquire the information of the driving trace of other barrier vehicles, purpose
It is that the trace information of other barrier vehicles is passed in neural network, then neural network can export the future of other vehicles
Track)
Barrier trajectory predictions subsystem 2, the barrier vehicle coordinate for detecting obstacle detector 1 are passed to
The barrier trajectory predictions algorithm model that dataset acquisition device 4 trains, and barrier is predicted according to the historical track of barrier
Hinder the Future Trajectory of object;
Vehicle Decision Method subsystem 3, for according to barrier trajectory predictions algorithm model predict come Future Trajectory carry out
Decision, decision making algorithm are demarcated according to the position that the Future Trajectory of vehicle is likely to occur, and may be occupied around this
Position.
Specific embodiment 2:
A kind of intelligent barrier avoiding system of automatic driving vehicle of present embodiment, the obstacle detector 1 pass through
Visual sensor and laser radar sensor are realized;
The visual sensor detects vehicle by object detection method,
The laser radar sensor detects the specific location of barrier;
Using laser radar sensor and visual sensor merge as a result, obtain whether barrier is vehicle, and obtain
Take barrier vehicle relative to the relative position coordinates of sensor.
Specific embodiment 3:
A kind of intelligent barrier avoiding system of automatic driving vehicle of present embodiment, the dataset acquisition device 4, passes through
Laser sensor and visual sensor are realized, are that laser sensor and visual sensor are mounted in common vehicle, by driver
Drive the driving information that common vehicle acquires other vehicles under port environment.
Specific embodiment 4:
A kind of intelligent barrier avoiding system of automatic driving vehicle of present embodiment, barrier trajectory predictions subsystem 2 are applied
Collected training set is passed to Recognition with Recurrent Neural Network training by Recognition with Recurrent Neural Network (RNN), and training is passed to new rail after completing
Mark information predicts following trace information by this new trace information using trajectory predictions algorithm.
Specific embodiment 5:
A kind of intelligent barrier avoiding system of automatic driving vehicle of present embodiment, Vehicle Decision Method subsystem 3 is according to predicting
Automatic driving vehicle the future may appear position, by decision making algorithm will herein position mark be occupy, carry out track
Around the position that this is occupied when planning.
Specific embodiment 6:
The intelligent barrier avoiding method of a kind of automatic driving vehicle of present embodiment, as shown in Fig. 2, the method includes following
Step:
Step 1: acquisition data:
Each sensor of application data set acquisition device obtains in barrier vehicle travel process, and barrier is in harbour scene
Under relative to collecting work vehicle left and right offset x and front and back offset y driving trace information, by the barrier track number of acquisition
According to being stored;
Step 2: to the pond layer of neural networkInput data is handled, whereinFor letter
Number, W is weight;
Step 3: the data after the processing of pond layer are passed in Recognition with Recurrent Neural Network, Recognition with Recurrent Neural Network formula is
Step 4: being again introduced into the processing of pond layer, processing formula is similar with step 2, are as follows:
Step 5: incoming data to be changed into the form of Two dimension normal distribution parameter:
Step 6: setting loss function, loss function areIt is calculated with optimization
Method minimizes loss function;
Step 7: carrying out the training of Recognition with Recurrent Neural Network:
The parameter of Recognition with Recurrent Neural Network is optimized, training result is saved;
Step 8: trajectory predictions are carried out using trained Recognition with Recurrent Neural Network, according to avoidance principle, by the traveling of prediction
Track is added in trajectory planning, specifically:
1), in computer storage unitOperationTrajectory predictions algorithm;
2) then, the trained training result of load step seven inputs data into trained circulation mind by sensor
Through network, the pond layer of centre experience and the formula in training process and principle are same as above, it may be assumed that
3), it is by the new track of Two dimension normal distribution output predictionFormula indicates are as follows:
Wherein, step 1 dataset acquisition method specifically:
The vehicle for carrying out data acquisition is equipped with visual sensor and laser sensor, and application target detection algorithm identifies vehicle
And vehicle is tracked;It is driven at a constant speed by driver driving vehicle in harbour with the speed of 20-60km/h, by several hours
Afterwards, sensor gets the track data of vehicle times a series of, is stored for the use of trajectory predictions algorithm.
Specific embodiment 7:
A kind of intelligent barrier avoiding method of automatic driving vehicle of present embodiment, according to avoidance principle described in step 8,
The driving trace of prediction is added in trajectory planning specifically, being with current vehicle as shown in figure 3, two cars travel in opposite directions
Main study subject, barrier vehicle is towards the direction running where current vehicle, and the detection of obstacles of current vehicle fills at this time
Barrier vehicle can be detected by setting, and barrier trajectory predictions subsystem is by observing that the walking trend prediction of barrier vehicle goes out
Barrier vehicle the future may appear position, will predict the position mark come is the rectangle that frames of dotted line, vehicle in figure
Decision-making subsystem judges that the position that barrier vehicle is likely to occur carries out corresponding avoidance, and the new track regenerated is bent in figure
Shown in line;Wherein, by default, the new track that vehicle generates is on the right side of current vehicle running track;And it generates new
Track must satisfy the limitation of practical map, and track in the region that can not be travelled, cannot need to meet the traveling limitation of track of vehicle
Condition, in travelable region.
Specific embodiment 8:
A kind of intelligent barrier avoiding method of automatic driving vehicle of present embodiment, the traveling limit for meeting track of vehicle
Condition processed is realized by verification process, specifically:
Step 1, the restrictive condition that will likely influence track of vehicle are converted into mathematic condition, such as the limitation that may relate to
Condition includes: direction limitation, vehicle when rate limitation, acceleration limitation, turning limit at a distance from barrier etc., conversion
Mathematical formulae later are as follows: j1 (x), j2 (x), j3 (x) ...;
Step 2 takes different weight a1, a2, a3 ... to different limitations respectively, to meet vehicle determining in different situations
Plan demand;
Step 3, the problem of converting majorized function for trajectory problem: x*=argminJ (x);Wherein, J (x) is to first pass through
Each restrictive condition is multiplied with each respective weights, then the result that will respectively seize the opportunity results added;
Step 4, solving optimization functional minimum value are to get to current optimal path.
Specific embodiment 9:
The core of a kind of intelligent barrier avoiding method of automatic driving vehicle of present embodiment, the trajectory predictions method is
Using Recognition with Recurrent Neural Network, Recognition with Recurrent Neural Network can preferable processing sequence problem, and what track of vehicle data can be convenient
It is converted into sequence problem.The training process and prediction process of neural network are respectively shown including two flow charts as shown in Figure 4;
Trained model can be generated after trajectory predictions algorithm executes training process, it is defeated during neural network prediction
The track for entering barrier vehicle can be obtained by the Future Trajectory of vehicle using trained model before.
The present invention can also have other various embodiments, without deviating from the spirit and substance of the present invention, this field
Technical staff makes various corresponding changes and modifications in accordance with the present invention, but these corresponding changes and modifications all should belong to
The protection scope of the appended claims of the present invention.