CN109186625A - Intelligent vehicle carries out pinpoint method and system using mixing sampling filter - Google Patents
Intelligent vehicle carries out pinpoint method and system using mixing sampling filter Download PDFInfo
- Publication number
- CN109186625A CN109186625A CN201811243970.9A CN201811243970A CN109186625A CN 109186625 A CN109186625 A CN 109186625A CN 201811243970 A CN201811243970 A CN 201811243970A CN 109186625 A CN109186625 A CN 109186625A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- carries out
- filter
- positioning
- mixing sampling
- 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
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
- G01C21/34—Route searching; Route guidance
- G01C21/3446—Details of route searching algorithms, e.g. Dijkstra, A*, arc-flags, using precalculated routes
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/48—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00
- G01S7/4802—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S17/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
Abstract
The invention discloses a kind of intelligent vehicles to carry out pinpoint method and system using mixing sampling filter, utilize mixing sampling filter, the real-time point cloud data acquired in conjunction with the accurately diagram data and laser radar or camera constructed in advance, vehicle is realized without the accurate positioning in/weak satellite-signal environment, the auxiliary that can be used for vehicle drives and unmanned.Using a new probabilistic localization frame, stochastical sampling consistency algorithm and particle filter/histogram filtering algorithm are efficiently fused together using the method for Density Estimator, it makes full use of respective advantage to make up the deficiency of other side, the Probabilistic Cell map based on gauss hybrid models is then used to carry out positioning more robust, accurately and fast.
Description
Technical field
The invention belongs to intelligent driving technical fields more particularly to a kind of intelligent vehicle to carry out essence using mixing sampling filter
Determine the method and system of position.
Background technique
Intelligent driving technology is related to cognitive engineering, Vehicle Engineering, electric and electronic engineering, Control Science and Engineering, artificial intelligence
The subjects such as energy, intelligent vehicle are the important symbols for measuring national a research strength and industrial level.Automatic driving car
Occur, fundamentally changes the vehicle drive mode in traditional " people-Che-road " closed-loop system, driver is driven from fatigue
Freed in sailing, using advanced sensor and information technology control vehicle driving, allow in driving-activity it is conventional, persistently,
Rudimentary, duplicate operation is automatically performed, and can greatly improve the efficiency and safety of traffic system, improves mankind's movable living
Quality has extensive social application value.Meanwhile the research of intelligent driving technology will greatly enhance China in automobile active
The core competitiveness of safety etc. has promotion China's automobile electronics and the automobile industry capability of independent innovation great
Strategic importance.
Intelligent vehicle further includes environment sensing, accurate positioning, information merges, intelligence is determined on the basis of dynamics of vehicle
The parts such as plan, vehicle control.Wherein, location technology is a very crucial basic work in automatic Pilot field.Base
It can only achieve 10m or so in the positioning accuracy of Global Navigation Satellite System (GNSS), be unable to satisfy automatic Pilot demand.Based on load
The satellite positioning of wave phase difference (RTK) can keep the precision of Centimeter Level under the good environment of signal, but its signal is easily
Multipath effect is blocked or is generated by tall and big object around, so as to cause the fluctuation of positioning result and the decline of positioning accuracy.
Combined orientation technology of the RTK-GNSS in conjunction with Inertial Measurement Unit (IMU) or odometer can be improved this to a certain extent
Problem.But since inertial navigation and odometer have accumulated error, in the environment of satellite positioning is long-term unstable or fails
High-precision positioning is kept, then inertial navigation is needed to have very high precision, and this cost is very expensive.
In order to realize high-precision positioning in relatively low cost-range, the method for mainstream is usually used at present
LiDAR or vision, in conjunction with the high-precision map constructed in advance, by the way that the method for real time data and map data matching will be positioned
Precision controlling is in usable range.Currently, the cost of view-based access control model will be lower than LiDAR, but its range accuracy is low, and vulnerable to light
Influence.In contrast, the precision and reliability of LiDAR will be better than vision.
Laser radar can provide the three dimensional local information and Reflection intensity information of the body surface in ambient enviroment, the two
It can be used to position.It is accurate that positioning based on reflectivity uses the planes static informations such as mark and lane line on road surface to realize
Positioning, the positioning accuracy of this method is high, is influenced by dynamic objects such as vehicle/pedestrians small, but this method application scenarios compare
It is limited.Rain and snow weather will lead to pavement marker and become blurred, to influence positioning accuracy；In non-crossing section, road surface is lacked
Weary longitudinal direction feature, longitudinal register precision are difficult to ensure；And in the unstructured roads such as rural, cross-country, road surface may be absolutely not
Road sign information, this localization method will be entirely ineffective at this time.
Positioning application scenarios based on elevation information are more extensive, influenced by weather it is small, and in structuring and unstructured
Road is applicable in.But the critical issue faced is the moving target (vehicle and pedestrian of traveling) overcome in environment, changeable
Static object (urban renewal, the vehicle of stop, the growth of plant leaf) and it is resulting block to LiDAR positioning
It influences.
Baidu uses the positioning for merging two kinds of information, according to the confidence level of positioning result, adaptive adjustment
The weight of the two.But when surface mark is contaminated and ambient enviroment there are when a large amount of dynamic objects, the above method would become hard to
To stabilization, accurate location structure.
The key problem of LiDAR positioning is exactly the registration problems of real time data and map datum, and type is not according to the map
It is same to be divided into a registration for the registration of cloud map and grating map.
It is usually used in scene rebuilding and SLAM based on a registration for cloud map, more representational two classes method has repeatedly
NDT is converted for closest approach ICP and Gaussian Profile.Many researchers are made that contribution to the development of two class method for registering, propose
A series of outstanding innovatory algorithms, such as GICP, ML-NDT, IRON etc. improve the robustness of algorithm.
Registration based on these methods usually requires to define the cost function of two frame point cloud registerings, and crosses calculating functional gradient
Carry out the minimum registration cost of iteration.Due to the property of gradient decline, these methods are highly dependent on initial position, and by
The influence of local minimum.It is larger in initial error, it is easily trapped into local optimum in the biggish situation of ambient noise, often can not
Reach ideal registration effect.Therefore, they are normally used as local accuracy registration algorithm and come using before this, usually
It needs to carry out rough registration to filter out noise, and more satisfactory initial position is provided for accuracy registration.
Method based on stochastical sampling consistency (Random Sample Consensus, RANSAC) is presently the most normal
One of rough registration method, many scholars are improved from feature extraction and RANSAC algorithm itself, 4 points of matching (4-
Point Congruent Set, 4PCS) and its innovatory algorithm utilize the coplanar 4 points of progress RANSAC for meeting affine constraint independent of time
Search, even if the overlapping region of two frame point clouds is smaller, initial positional deviation is larger, can still obtain good registration result,
It achieves to compare and be widely applied.But in the unmanned urban environment faced, it is frequently present of a large amount of dynamic object, to be made
With the mode of random sampling simultaneously find four static object points be it is very difficult, efficiency of algorithm can have a greatly reduced quality.ML-
RANSAC algorithm solves the point cloud under high dynamic scene by the multi-layer framework of the sample-iteration-sample established and matches
Quasi- problem.The present invention is extended the algorithm, is incorporated in the probabilistic localization frame the present invention is based on grating map.
In the registration based on grating map, particle filter be the algorithm that is most widely used of mobile robot field it
One.The algorithm when handling the extremely complex or unknown problem of prior distribution with very outstanding performance, when population abundance
And in the case where being reasonably distributed, even if search space is larger, still there can be very maximum probability to find optimal solution after successive ignition.
With iterations going on, particle can gradually restrain, so that arithmetic accuracy is gradually promoted, but also have lost the multiplicity of particle simultaneously
Property, cause particle dilution, algorithm robustness decline in large-scale high dynamic scene, increases the wind for falling into local optimum
Danger.
In view of unpiloted application scenarios, position error can be usually limited in several meters of model by global position system
In enclosing, Google proposes histogram filtering (Histogram Filter, HF) algorithm.This method allows do not have to concern particle
The remarkable advantage lost near correct position also becomes one of the main stream approach of LiDAR positioning in recent years.Histogram filtering
Inefficiency when there are problems that search space is larger with particle filter, in higher-dimension situation, this problem is more prominent.Needle
To the problem, Baidu uses Lucas-kanade (LK) algorithm to determine the course of vehicle first, then with histogram filtering come into
The location efficiency in big search space can be improved in the positioning of two dimensions in row horizontal position, but LK algorithm is inclined in course angle
Failure is easy when difference is excessive.
Although in most cases, the error of satellite positioning result all will not be especially big, some ratios are still remained
More extreme situation.For example, blocking for tunnel will lead to satellite positioning failure, and the longitudinal feature list in tunnel in tunnel
One will lead to LiDAR positioning failure, relative positioning can only be carried out by dead reckoning, vehicle travels certain in this environment
After discrete time, biggish accumulated error can be generated, as soon as at this time if vehicle driving is needed to the fork in the road in tunnel
It to be positioned under biggish initial deviation using LiDAR and map.In addition, unmanned vehicle can also encounter asking for similar " kidnapping "
Topic, for example be switched in underground garage, initial position error at this time is usually very big, it there may come a time when to have rice up to a hundred,
And the course of vehicle is also that completion is unknown.In this case positioning is also what automatic Pilot needed to solve.
Summary of the invention
The present invention provides a kind of intelligent vehicle and carries out pinpoint method and system using mixing sampling filter, using mixed
Sampling filter is closed, it is real in conjunction with the real-time point cloud data that the accurately diagram data and laser radar or camera constructed in advance acquires
Without the accurate positioning in/weak satellite-signal environment, the auxiliary that can be used for vehicle drives and unmanned existing vehicle.
The present invention to achieve the above object, using following technical scheme: first using gauss hybrid models construct probability grid
Then lattice map uses a new probabilistic localization frame, will be based on stochastical sampling consistency using the method for Density Estimator
Method and method based on Bayesian filter be efficiently fused together, make full use of respective advantage to make up the deficiency of other side, into
Row positioning more robust, accurately and fast.
All sampled results in stochastical sampling coherence method are subjected to core rather than just interior ratio maximum result
Density estimation, and in this, as the update of prior distribution progress Bayesian filter, then the prediction distribution of Bayesian filter is made
For the priori of subsequent time new round stochastical sampling.
Specific step is as follows:
A. the point cloud data acquired in advance is projected in two-dimensional grid, often according to its corresponding accurate world coordinates
A grid contains elevation information and reflectivity information, and two kinds of information are indicated by gauss hybrid models, in gauss hybrid models
Component number be not less than 1, grid is square, side length be not less than 5cm.
B. the real-time cloud and map datum of acquisition are denoised respectively, segmentation and feature extraction, then to extraction
Characteristic point carries out the registration based on RANSAC.
C. the interior ratio sampling for meeting constraint condition in RANSAC each time obtained as the weight of the position or
Confidence level is clustered and is fitted to whole results in this RANSAC using the method for Density Estimator, obtained each
The parameter of core.Distance threshold d is put in whereininlierDetermine the error range of rough registration, and the bandwidth h of Density EstimatorKDEWith
The sample space size w of filtering algorithm in accurate positioningspaceIt is determined by the error range.Therefore by the frame, we are just
Can be naturally by the key parameter of these three parts: interior distance threshold in RANSAC, Gaussian Kernel Density estimation in band
Width, the initial samples space filtered in positioning are unified.
hKDE=k1·dinlier
wspace=k2·dinlier
D. it is filtered, is obtained several accurate using such as histogram of the filtering method based on Bayes near the peak value of distribution
High probability position is filtered each peak value, and according to kernel density function and Ben Shi when peak value number is more than 1 using histogram
The prior probability at quarter calculates these positions in the posterior probability at current time.Use the maximum position of posterior probability fixed as this
The final position of position, according to probability calculation, these position variances obtain the confidence level of this positioning.
E. according to markov principle and bayes rule, the motion state exported using Inertial Measurement Unit or odometer
Information calculates separately the prior distribution of the corresponding subsequent time positioning in these positions.
Wherein, xtIndicate vehicle in the position of t moment,Indicate predicted value of the vehicle at the t+1 moment, ut+1Indicate vehicle
From t moment to the motion state at t+1 moment, it is (μ that t+1 moment car status information, which obeys parameter,t+1,σt+1) Gaussian Profile,
Δxt+1,tIt indicatesWith xtVariable quantity.
Correspondingly, it additionally provides a kind of intelligent vehicle and carries out pinpoint system using mixing sampling filter, including is mixed
The real-time point cloud data unit of sampling filtering unit, the high-precision map unit constructed in advance and radar or camera acquisition is closed,
Using the mixing sampling filtering unit, in conjunction with the high-precision map cell data constructed in advance and radar or
The real-time point cloud data unit of camera acquisition realizes accurate positioning of the vehicle in nothing/weak satellite-signal environment.
Wherein, the mixing sampling filtering unit is a kind of consistent based on stochastical sampling using the fusion of Density Estimator method
Property method and the probabilistic localization frame based on Bayesian filter method, by all sampled results in stochastical sampling coherence method
Density Estimator is carried out rather than just interior ratio maximum result, and carries out Bayesian filter more in this, as prior distribution
Newly, and using the prediction distribution of Bayesian filter as the priori of subsequent time new round stochastical sampling.
Wherein, the high-precision map unit constructed in advance is to project to point cloud data in two-dimensional grid, each grid
Lattice contain elevation information and reflectivity information, and two kinds of information are indicated by gauss hybrid models, point in gauss hybrid models
It measures number and is not less than 1, grid is square, and side length is not less than 5cm.
Wherein, the radar or camera acquisition real-time point cloud data unit be include directly parsing to obtain from sensor
Original point cloud and by algorithm process extract characteristic point cloud.
Compared with prior art, beneficial effects of the present invention are to get rid of automatic Pilot positioning to be used to expensive high-precision
Property navigation equipment demand, it is fixed when can realize the high-precision real for meeting automatic Pilot demand in the case where cost is greatly reduced
Position；
With the more wide scope of application and scene, it can be not only used for urban structure road, can be also used for township
The semi-structured road in village and cross-country unstructured road；It can be not only used for static or low dynamic scene, can be also used for vehicle
The high dynamic scene that pedestrian is various, environmental change is big；Can not only work under good weather condition, can be also used for rain,
Snow, greasy weather gas；
The initial pose deviation that can be tolerated can reach rice up to a hundred, can solve the starting of vehicle underground garage, long-term tunnel
Accurate positioning in the extreme scenes such as road traveling, provides important technology for the realization and volume production of the automatic Pilot of L3, L4, L5 rank
Support.
Detailed description of the invention
Fig. 1 the application preferred embodiments test section satellite mapping；
Fig. 2 uses the movement locus schematic diagram before and after the method for the present invention positioning；
The positioning result contrast schematic diagram of tri- kinds of distinct methods of Fig. 3；
In Fig. 3: (a) lateral error, (b) longitudinal error, (c) course angle error；
The positioning result schematic diagram at the crossing Fig. 4；
In Fig. 4: (a) top view that this example positions on high-precision map, (b) this example positions on high-precision map
3D view, (c) the forward sight camera view of positioning scene, (d) multi-level stochastical sampling one of positioning distribution (e) of histogram filtering
The positioning distribution of cause property, (f) the positioning distribution of this example；
The positioning result schematic diagram of Fig. 5 linear road；
In Fig. 5: (a) top view that this example positions on high-precision map, (b) this example positions on high-precision map
3D view, (c) the forward sight camera view of positioning scene, (d) multi-level stochastical sampling one of positioning distribution (e) of histogram filtering
The positioning distribution of cause property, (f) the positioning distribution of this example；
Horizontal position error is distributed under Fig. 6 difference initial deviation；
In Fig. 6: (a) scatter plot, (b) histogram；
Fig. 7 is the application method and step flow chart.
Specific embodiment
The present invention is described in further detail below in conjunction with the drawings and specific embodiments.It should be appreciated that described herein
Specific embodiment be only used to explain the present invention, be not intended to limit the present invention.
It should be noted that in the absence of conflict, the features in the embodiments and the embodiments of the present application can phase
Mutually combination.
This example carries out point cloud data acquisition using velodyne HDL-64E laser radar, and algorithm is in dominant frequency
It is run on the 4 core industrial personal computers of 2.40GHz, inside saves as 8GB, system uses ubuntu 16.04.
This example is distinguished using the Gauss model of the component containing there are two come the reflectivity to each grid and highly distribution
It is modeled.For elevation information, it is distributed using the height that two Gaussian components describe ground and barrier respectively；For reflection
Rate information describes the distribution of the high reflectance data and antiradar reflectivity data on ground using two components respectively.Grid size
For 20cm.
This example is used based on the algorithm of region growing of polynary geometrical constraint come to laser radar point cloud and grating map point
Denoising and segmentation are not synchronized, and the focus point for extracting each target after segmentation is based on as with mutatis mutandis characteristic point
The ML-RANSAC of " sample-iteration-sample ".
Density Estimator is a kind of nonparametric technique for estimated probability density function, x1,x2,...,xnIt is independent same
N sample point of distribution, if its probability density function is f, Density Estimator is following:
Wherein, K () is kernel function, this example is using gaussian kernel function.H > 0 is smoothing bandwidth (bandwidth).
This example estimates (x, y, θ) three parameters of vehicle, in the model, only models to horizontal position
And estimation, i.e. x=(x, y), the course angle of vehicle use the corresponding course angle of the maximum pose of weight in each Gaussian kernel.
Bayesian filter is a kind of filter based on Bayes principle, including Kalman filtering, particle filter, histogram
Filtering etc., this example are filtered using the histogram of current mainstream.In order to improve locating speed, using the method for branch-and-bound
Histogram filtering is improved.
This example has chosen true urban road scene and is tested, and test segment is as shown in Figure 1.Section overall length
7.3km, section contain urban inner major trunk roads, large-scale intersection, city expressway, overpass gate etc..The section
The acquisition time of high-precision map and the acquisition time of experimental data are spaced 1 year or more, and the data for winter, and one
For the data of summer.The environment in many places of this meaning has occurred that a degree of variation, and two groups of hairs are illustrated in Fig. 1
The scene for changing, one group of variation for road equipment at intersection, another group be the road Dong Xialiangji both sides vegetation variation,
And antifreeze fence can be arranged to vegetation in winter, and summer does not have then.The variation of these static scenes will increase the difficulty of LiDAR positioning
Degree.We are by algorithm proposed by the present invention with the localization method (hereinafter referred to as HF) based on histogram filtering HF and based on ML-
The localization method (hereinafter referred to as ML-RANSAC) of RANSAC compares.
Different according to the type of initial positional deviation, experiment is divided into 2 parts, in the road for containing a large amount of dynamic objects
It is carried out in scene.
The position deviation of 1 consecutive variations
One test of experiment is that the lower IMU of a precision is used only and carries out vehicle position without using GNSS
The reckoning of appearance, the input that the reckoning result of inertial navigation is positioned as LiDAR, testing algorithm are having existing for a large amount of vehicle and pedestrians
Positioning accuracy and stability in high dynamic urban environment.It is this to calculate that the accumulated error generated is continuous slowly varying by pure
, there is no the jumps of position.
Meanwhile the vehicle driving trace based on differential GPS is acquired, and use offline LiDAR SLAM on this basis
Truthful data of the technology as vehicle location.Experiment includes AB and two sections of CD.Fig. 2 respectively illustrates the true of vehicle driving trace
Track after data, the track that IMU is calculated and the LiDAR positioning based on the method for the present invention.
The position error as shown in Figure 3 of three kinds of methods in an experiment can be seen that the method for the present invention either mean error
Or worst error is minimum in three.HF precision is higher, and mean error and the method for the present invention error are close, but the Shandong of algorithm
Stick is slightly poor, and in some dynamic scenes, blocking and interfering due to dynamic object is easy to appear the situation of registration mistake, ML-
The stability of RANSAC located lateral is better than HF, but the scene less in some longitudinal features, because being not bound with filtering positioning
Algorithm and probabilistic framework of the invention lack reasonable confidence level estimation, and positioning result is easy to appear longitudinal fluctuation, and total
The positioning accuracy of body is poor.
Fig. 4 and Fig. 5 illustrates the positioning result of two different scenes.In this two groups of scenes, HF and ML-RANSAC's determines
Position result has gone out first deviation in various degree.Wherein (a) (b) respectively indicates LiDAR in high-precision map using the present invention
The top view of the positioning result of method, 3D view, wherein high-precision map shows that red display is for fixed with grayscale image
The real-time point cloud data of position, blue indicate that the static object obtained by registration, yellow line are the track of LiDAR positioning.It (c) is vehicle
Carry the scene figure of forward sight camera shooting.(d) (e) (f) is the probability distribution of positioning result fitting of three kinds of methods in each scene,
As can be seen that due to the interference of dynamic object in the scene of crossing, the peak value of the positioning distribution of HF has deviated considerably from correct position
It sets, and ML-RANSAC is then closer to actual position；Due to the scarcity of environment longitudinal direction feature in linear road scene, HF's determines
There is biggish variance in the longitudinal direction in bit distribution, and the positioning of ML-RANSAC distribution then exist in a vehicle longitudinal direction it is multiple
Peak value.Through the invention to the fusion of two methods, obtain that peak value is accurate and variance is lesser in above-mentioned two scene
Positioning distribution.
The position deviation of 2 Discrete Stochastics jump
Experiment two further promotes difficulty on the basis of testing one, is further added by the reckoning result there are accumulated error
The input that one random bias is positioned as LiDAR, simulation GPS is because not exclusively blocking the positional mutation feelings generated with multipath effect
Condition tests the precision and stability of algorithm in the presence of having jump error.Experiment is divided into three groups, course angle offset
The random fluctuation in ± 5 °, horizontal position offset amount is respectively in 2.5m, 5m, random fluctuation in 10m radius.
Fig. 6 shows the positioning result horizontal distance under different initial deviations of three kinds of localization methods on the test segment
Error is distributed scatter plot and error distribution histogram.For the clearer distribution situation that position error is presented, Fig. 6's (a) is vertical
Coordinate has used logarithmic coordinates, it can be seen that the positioning accuracy of the method for the present invention is to initial deviation and insensitive, positioning accuracy beginning
It is maintained at higher level eventually, and the global error of the positioning result of HF and ML-RANSAC and fluctuation are more larger.In the reality
In testing, we have only counted that LiDAR is positioned as a result, eliminating the reckoning result of IMU between LiDAR positioning twice.Due to three
The time-consuming that kind method is completed once to position under different initial errors is different, therefore in same section of test data, total time
Identical, the positioning number of different initial error ranges is different, with the increase of initial deviation, the positioning number of three kinds of methods
The apparent decline trend of effect.
As can be seen that inventive algorithm is much larger than other methods in number of the horizontal error less than 0.1m from Fig. 6 (b),
And the error integrated distribution of positioning result is within 0.5m.
Table 1 gives the statistical result that three kinds of methods position under different initial deviations, including transverse direction, longitudinal, course angle
And the root-mean-square error of horizontal position, horizontal error are less than the percentage of 0.1m and 0.4m, the mean value and standard of algorithm time-consuming
Difference.When initial deviation is 10m, positioning number of the HF in 0.4m error only accounts for the 9.09% of its total positioning number, and originally
Inventive method still has 93.18%.
Positioning statistical result comparison under the different initial positional deviations of table 1
The experimental results showed that either in the initial error continuously accumulated still in the initial error feelings of Discrete Stochastic bounce
Under condition, algorithm maintains higher positioning accuracy and faster locating speed.It filters and calculates compared to the histogram before improvement
Method and multi-level stochastical sampling consistency algorithm further improve positioning under the premise of guaranteeing positioning accuracy and speed
Robustness can adapt to more complicated positioning scene；And there are can still protect in the case where relatively large deviation in initial position
Quickly accurate positioning is held, this is that algorithm before improving can not be accomplished.
The above is only a preferred embodiment of the present invention, it is noted that for the common skill of the art
For art personnel, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications
Also it should be regarded as protection scope of the present invention.
Claims (8)
1. a kind of intelligent vehicle carries out pinpoint method using mixing sampling filter, it is characterized in that: using mixing sampling filter
Wave, in conjunction with the real-time point cloud data that the accurately diagram data and radar or camera that construct in advance acquire, realize vehicle without/it is weak
Accurate positioning in satellite-signal environment.
2. intelligent vehicle according to claim 1 carries out pinpoint method using mixing sampling filter, it is characterized in that:
The mixing sampling filter is that a kind of merged using Density Estimator method is based on stochastical sampling coherence method and based on pattra leaves
The probabilistic localization frame of this filtering method, by all sampled results in stochastical sampling coherence method rather than just interior ratio
Example maximum result carries out Density Estimator, and the update of Bayesian filter is carried out in this, as prior distribution, and Bayes is filtered
Priori of the prediction distribution of wave as subsequent time new round stochastical sampling.
3. intelligent vehicle according to claim 1 or 2 carries out pinpoint method, feature using mixing sampling filter
Be: the high-precision map constructed in advance is to project to point cloud data in two-dimensional grid, and each grid, which contains height, to be believed
Breath and reflectivity information, two kinds of information are indicated that the component number in gauss hybrid models is not less than 1 by gauss hybrid models.
4. intelligent vehicle according to claim 1 or 2 carries out pinpoint method, feature using mixing sampling filter
Be: the radar or camera acquisition real-time point cloud data be include the original point cloud and warp directly parsed from sensor
Cross the characteristic point cloud of algorithm process extraction.
5. a kind of intelligent vehicle carries out pinpoint system using mixing sampling filter, it is characterized in that: including mixing sampling filter
Wave unit, the high-precision map unit constructed in advance and radar or the real-time point cloud data unit of camera acquisition,
Using the mixing sampling filtering unit, in conjunction with the accurately diagram data unit constructed in advance and radar or camera
The real-time point cloud data unit of acquisition realizes accurate positioning of the vehicle in nothing/weak satellite-signal environment.
6. intelligent vehicle according to claim 5 carries out pinpoint system using mixing sampling filter, it is characterized in that:
The mixing sampling filtering unit is that a kind of merged using Density Estimator method is based on stochastical sampling coherence method and be based on
The probabilistic localization frame of Bayesian filter method, by all sampled results in stochastical sampling coherence method rather than just interior
Point ratio maximum result carries out Density Estimator, and the update of Bayesian filter is carried out in this, as prior distribution, and by pattra leaves
Priori of the prediction distribution of this filtering as subsequent time new round stochastical sampling.
7. intelligent vehicle according to claim 5 or 6 carries out pinpoint system, feature using mixing sampling filter
Be: the high-precision map unit constructed in advance is to project to point cloud data in two-dimensional grid, and each grid contains height
Degree information and reflectivity information, two kinds of information are indicated that the component number in gauss hybrid models is not small by gauss hybrid models
In 1.
8. intelligent vehicle according to claim 5 or 6 carries out pinpoint system, feature using mixing sampling filter
Be: the radar or camera acquisition real-time point cloud data unit be include the original point cloud directly parsed from sensor
With the characteristic point cloud extracted by algorithm process.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811243970.9A CN109186625B (en) | 2018-10-24 | 2018-10-24 | Method and system for accurately positioning intelligent vehicle by using hybrid sampling filtering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811243970.9A CN109186625B (en) | 2018-10-24 | 2018-10-24 | Method and system for accurately positioning intelligent vehicle by using hybrid sampling filtering |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109186625A true CN109186625A (en) | 2019-01-11 |
| CN109186625B CN109186625B (en) | 2020-05-05 |
Family
ID=64943128
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811243970.9A Active CN109186625B (en) | 2018-10-24 | 2018-10-24 | Method and system for accurately positioning intelligent vehicle by using hybrid sampling filtering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109186625B (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109725329A (en) * | 2019-02-20 | 2019-05-07 | 苏州风图智能科技有限公司 | A kind of unmanned vehicle localization method and device |
| CN109839112A (en) * | 2019-03-11 | 2019-06-04 | 中南大学 | Underground working apparatus localization method, device, system and storage medium |
| CN109856993A (en) * | 2019-01-29 | 2019-06-07 | 北京奥特贝睿科技有限公司 | A kind of autonomous driving emulation platform |
| CN109900298A (en) * | 2019-03-01 | 2019-06-18 | 武汉光庭科技有限公司 | A kind of vehicle location calibration method and system |
| CN109931939A (en) * | 2019-02-27 | 2019-06-25 | 杭州飞步科技有限公司 | Localization method, device, equipment and the computer readable storage medium of vehicle |
| CN110045733A (en) * | 2019-04-04 | 2019-07-23 | 肖卫国 | A kind of real-time location method and its system, computer-readable medium |
| CN110060282A (en) * | 2019-03-27 | 2019-07-26 | 东软睿驰汽车技术（沈阳）有限公司 | A kind of method and device of point cloud registering |
| CN110556012A (en) * | 2019-09-16 | 2019-12-10 | 北京百度网讯科技有限公司 | Lane positioning method and vehicle positioning system |
| CN110595492A (en) * | 2019-09-25 | 2019-12-20 | 上海交通大学 | Vehicle self-positioning system and method in park environment |
| CN110766979A (en) * | 2019-11-13 | 2020-02-07 | 奥特酷智能科技(南京)有限公司 | Parking space detection method for automatic driving vehicle |
| CN111257882A (en) * | 2020-03-19 | 2020-06-09 | 北京三快在线科技有限公司 | Data fusion method and device, unmanned equipment and readable storage medium |
| CN111522020A (en) * | 2020-06-23 | 2020-08-11 | 山东亦贝数据技术有限公司 | Park activity element hybrid positioning system and method |
| CN112154429A (en) * | 2019-07-29 | 2020-12-29 | 深圳市大疆创新科技有限公司 | High-precision map positioning method, system, platform and computer readable storage medium |
| CN112455503A (en) * | 2019-09-09 | 2021-03-09 | 中车株洲电力机车研究所有限公司 | Train positioning method and device based on radar |
| CN112711027A (en) * | 2020-12-08 | 2021-04-27 | 北京市首都公路发展集团有限公司 | Tunnel internal transverse positioning method based on laser radar point cloud data |
| CN112747756A (en) * | 2019-10-30 | 2021-05-04 | 北京地平线机器人技术研发有限公司 | Map construction method and device |
| CN112824997A (en) * | 2019-11-20 | 2021-05-21 | 通用汽车环球科技运作有限责任公司 | Method and system for local lane of travel perception |
| CN113008245A (en) * | 2019-12-20 | 2021-06-22 | 北京图森智途科技有限公司 | Positioning information fusion method and device and computer server |
| CN114323035A (en) * | 2020-09-30 | 2022-04-12 | 华为技术有限公司 | Positioning method, device and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103473565A (en) * | 2013-08-23 | 2013-12-25 | 华为技术有限公司 | Image matching method and device |
| CN105467838A (en) * | 2015-11-10 | 2016-04-06 | 山西大学 | Synchronous positioning and map constructing method under random finite set framework |
| CN106534766A (en) * | 2015-09-09 | 2017-03-22 | 广州市维安电子技术有限公司 | Method and device for using PTZ camera to automatically track target |
-
2018
- 2018-10-24 CN CN201811243970.9A patent/CN109186625B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103473565A (en) * | 2013-08-23 | 2013-12-25 | 华为技术有限公司 | Image matching method and device |
| CN106534766A (en) * | 2015-09-09 | 2017-03-22 | 广州市维安电子技术有限公司 | Method and device for using PTZ camera to automatically track target |
| CN105467838A (en) * | 2015-11-10 | 2016-04-06 | 山西大学 | Synchronous positioning and map constructing method under random finite set framework |
Non-Patent Citations (8)
| Title |
|---|
| CHEN TAI，YUN-HUI LIU: "A Robust Estimator for Structure from Motion Based on Kernel Density Estimation", 《PROCEEDINGS OF THE 2006 IEEE/RSJ INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS》 * |
| GUOWEI WAN, ET AL: "Robust and Precise Vehicle Localization based on Multi-sensor Fusion in Diverse City Scenes", 《2018 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION》 * |
| HYUNGJIN KIM,ET AL: "Robust Vehicle Localization Using Entropy-Weighted Particle Filter-based Data Fusion of Vertical and Road Intensity Information for a Large Scale Urban Area", 《IEEE ROBOTICS AND AUTOMATION LETTERS》 * |
| JESSE LEVINSON, ET AL: "Robust Vehicle Localization in Urban Environments Using Probabilistic Maps", 《2010 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION ANCHORAGE CONVENTION DISTRICT》 * |
| 张明友、吕明: "《近代信号处理理论与方法》", 31 January 2005, 国防工业出版社 * |
| 王任栋等: "一种鲁棒的城市复杂动态场景点云配准方法", 《机器人》 * |
| 赵凯等: "基于均值高程图的城市环境三维LiDAＲ 点云地面分割方法", 《军事交通学院学报》 * |
| 靳中鑫: "《应用统计信息分析与例题解析》", 31 January 2006, 国防工业出版社 * |
Cited By (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109856993B (en) * | 2019-01-29 | 2022-05-31 | 北京奥特贝睿科技有限公司 | Autonomous driving simulation platform |
| CN109856993A (en) * | 2019-01-29 | 2019-06-07 | 北京奥特贝睿科技有限公司 | A kind of autonomous driving emulation platform |
| CN109725329A (en) * | 2019-02-20 | 2019-05-07 | 苏州风图智能科技有限公司 | A kind of unmanned vehicle localization method and device |
| CN109931939A (en) * | 2019-02-27 | 2019-06-25 | 杭州飞步科技有限公司 | Localization method, device, equipment and the computer readable storage medium of vehicle |
| CN109900298B (en) * | 2019-03-01 | 2023-06-30 | 武汉光庭科技有限公司 | Vehicle positioning calibration method and system |
| CN109900298A (en) * | 2019-03-01 | 2019-06-18 | 武汉光庭科技有限公司 | A kind of vehicle location calibration method and system |
| CN109839112A (en) * | 2019-03-11 | 2019-06-04 | 中南大学 | Underground working apparatus localization method, device, system and storage medium |
| CN109839112B (en) * | 2019-03-11 | 2023-04-07 | 中南大学 | Underground operation equipment positioning method, device and system and storage medium |
| CN110060282A (en) * | 2019-03-27 | 2019-07-26 | 东软睿驰汽车技术（沈阳）有限公司 | A kind of method and device of point cloud registering |
| CN110045733A (en) * | 2019-04-04 | 2019-07-23 | 肖卫国 | A kind of real-time location method and its system, computer-readable medium |
| CN110045733B (en) * | 2019-04-04 | 2022-11-01 | 肖卫国 | Real-time positioning method and system and computer readable medium |
| CN112154429B (en) * | 2019-07-29 | 2024-03-19 | 深圳市大疆创新科技有限公司 | High-precision map positioning method, system, platform and computer readable storage medium |
| CN112154429A (en) * | 2019-07-29 | 2020-12-29 | 深圳市大疆创新科技有限公司 | High-precision map positioning method, system, platform and computer readable storage medium |
| WO2021016806A1 (en) * | 2019-07-29 | 2021-02-04 | 深圳市大疆创新科技有限公司 | High-precision map positioning method, system and platform, and computer-readable storage medium |
| CN112455503A (en) * | 2019-09-09 | 2021-03-09 | 中车株洲电力机车研究所有限公司 | Train positioning method and device based on radar |
| CN110556012A (en) * | 2019-09-16 | 2019-12-10 | 北京百度网讯科技有限公司 | Lane positioning method and vehicle positioning system |
| CN110556012B (en) * | 2019-09-16 | 2022-03-08 | 北京百度网讯科技有限公司 | Lane positioning method and vehicle positioning system |
| CN110595492A (en) * | 2019-09-25 | 2019-12-20 | 上海交通大学 | Vehicle self-positioning system and method in park environment |
| CN112747756A (en) * | 2019-10-30 | 2021-05-04 | 北京地平线机器人技术研发有限公司 | Map construction method and device |
| CN110766979A (en) * | 2019-11-13 | 2020-02-07 | 奥特酷智能科技(南京)有限公司 | Parking space detection method for automatic driving vehicle |
| CN112824997A (en) * | 2019-11-20 | 2021-05-21 | 通用汽车环球科技运作有限责任公司 | Method and system for local lane of travel perception |
| CN113008245A (en) * | 2019-12-20 | 2021-06-22 | 北京图森智途科技有限公司 | Positioning information fusion method and device and computer server |
| CN113008245B (en) * | 2019-12-20 | 2022-12-27 | 北京图森智途科技有限公司 | Positioning information fusion method and device and computer server |
| CN111257882B (en) * | 2020-03-19 | 2021-11-19 | 北京三快在线科技有限公司 | Data fusion method and device, unmanned equipment and readable storage medium |
| CN111257882A (en) * | 2020-03-19 | 2020-06-09 | 北京三快在线科技有限公司 | Data fusion method and device, unmanned equipment and readable storage medium |
| CN111522020A (en) * | 2020-06-23 | 2020-08-11 | 山东亦贝数据技术有限公司 | Park activity element hybrid positioning system and method |
| CN114323035A (en) * | 2020-09-30 | 2022-04-12 | 华为技术有限公司 | Positioning method, device and system |
| CN112711027A (en) * | 2020-12-08 | 2021-04-27 | 北京市首都公路发展集团有限公司 | Tunnel internal transverse positioning method based on laser radar point cloud data |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109186625B (en) | 2020-05-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109186625A (en) | Intelligent vehicle carries out pinpoint method and system using mixing sampling filter | |
| CN108955702B (en) | Lane-level map creation system based on three-dimensional laser and GPS inertial navigation system | |
| CN106908775B (en) | A kind of unmanned vehicle real-time location method based on laser reflection intensity | |
| CN111551958B (en) | Mining area unmanned high-precision map manufacturing method | |
| CN109099901B (en) | Full-automatic road roller positioning method based on multi-source data fusion | |
| CN110146910B (en) | Positioning method and device based on data fusion of GPS and laser radar | |
| Tao et al. | Lane marking aided vehicle localization | |
| Ma et al. | Generation of horizontally curved driving lines in HD maps using mobile laser scanning point clouds | |
| CN106651953A (en) | Vehicle position and gesture estimation method based on traffic sign | |
| Gikas et al. | A novel geodetic engineering method for accurate and automated road/railway centerline geometry extraction based on the bearing diagram and fractal behavior | |
| CN108961811A (en) | Parking lot vehicle positioning method, system, mobile terminal and storage medium | |
| Li et al. | Road DNA based localization for autonomous vehicles | |
| CN109101743B (en) | Method for constructing high-precision road network model | |
| CN112380312A (en) | Laser map updating method based on grid detection, terminal and computer equipment | |
| CN113920198B (en) | Coarse-to-fine multi-sensor fusion positioning method based on semantic edge alignment | |
| CN110349192A (en) | A kind of tracking of the online Target Tracking System based on three-dimensional laser point cloud | |
| CN109241855B (en) | Intelligent vehicle travelable area detection method based on stereoscopic vision | |
| Li et al. | Adaptive road detection method combining lane line and obstacle boundary | |
| Burnett et al. | Building a winning self-driving car in six months | |
| He et al. | Real-time track obstacle detection from 3D LIDAR point cloud | |
| CN111089580B (en) | Unmanned war chariot simultaneous positioning and map construction method based on covariance intersection | |
| Yu et al. | Road curbs detection based on laser radar | |
| CN112985428B (en) | Safety angle-based priority reference method for image layer of high-precision sensing map | |
| Deusch | Random finite set-based localization and SLAM for highly automated vehicles | |
| Hu et al. | LiDAR-based road extraction for UGV in high definition map |
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 |