CN110298320A - A kind of vision positioning method, device and storage medium - Google Patents
A kind of vision positioning method, device and storage medium Download PDFInfo
- Publication number
- CN110298320A CN110298320A CN201910586511.9A CN201910586511A CN110298320A CN 110298320 A CN110298320 A CN 110298320A CN 201910586511 A CN201910586511 A CN 201910586511A CN 110298320 A CN110298320 A CN 110298320A
- Authority
- CN
- China
- Prior art keywords
- region
- positioning
- semantic
- image
- target object
- 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
- 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
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/35—Categorising the entire scene, e.g. birthday party or wedding scene
Abstract
The embodiment of the present invention proposes vision positioning method, device and storage medium, wherein the described method includes: acquisition panoramic view data;Using the panoramic view data as classifying in training sample input disaggregated model, classification results are obtained;The positioning map based on semantic feature is obtained according to the classification results;At least one image data to be processed of current target object acquisition is inputted into the disaggregated model, in conjunction with the positioning map, positioning obtains the direction of the target object.Using the embodiment of the present invention, it can realize that accurate direction positions with existing magnetometer, while reducing the hardware cost of upgrading magnetometer.
Description
Technical field
The present invention relates to the technical fields of computer vision more particularly to a kind of vision positioning method, device and storage to be situated between
Matter.
Background technique
A kind of application scenarios of vision positioning processing are: (such as visual angle from left to right, view from right to left with different view
Angle, the visual angle overlooked from top to bottom etc.) see the same target object (such as building, vehicle, mobile phone terminal, in surrounding enviroment
A tree or the street lamp of curbside etc.), if checking that result is similar, it is difficult to determine the target object direction (or
Direction), it needs to position the target object.For example, can be looked by magnetometer under current scene at the parting of the ways
See the direction of certain target object.Can be more due to crossroad vehicle, the facilities such as traffic lights are also more, and these can all bring very greatly
Electromagnetic interference so that biggish error can be brought after electromagnetic interference for the magnetometer in detected target object direction, thus
Leading to the direction of target object can not accurately be determined.At present if it is intended to accurately determining direction under this scene, only
The more advanced magnetometer of energy, this certainly will increase cost.However, the problem does not obtain effective solution.
Summary of the invention
The embodiment of the present invention provides a kind of vision positioning method, is asked with solving one or more technologies in the prior art
Topic.
In a first aspect, the embodiment of the invention provides a kind of vision positioning methods, which comprises
Acquire panoramic view data;
Using the panoramic view data as classifying in training sample input disaggregated model, classification results are obtained;
The positioning map based on semantic feature is obtained according to the classification results;
At least one image data to be processed of current target object acquisition is inputted into the disaggregated model, in conjunction with described fixed
Position map, positioning obtain the direction of the target object.
It is described using the panoramic view data as classifying in training sample input disaggregated model in one embodiment, it obtains
To classification results, comprising:
In the disaggregated model, according to semantic segmentation strategy at least one image data in the panoramic view data into
Row image preprocessing, obtains pre-processed results, and the pre-processed results are the parts of images at least one described image data
Region;
Classify to the pre-processed results, obtains described in the semantic feature for corresponding to the partial image region and correspondence
The coordinate information of partial image region;
The semantic feature and the coordinate information are determined as the classification results.
In one embodiment, it is described according to semantic segmentation strategy at least one image data in the panoramic view data into
Row image preprocessing, obtains pre-processed results, comprising:
From at least one described image data, the object to remain static at the appointed time section is identified;
Using the corresponding image-region of the object as static information;
Using the static information as the pre-processed results.
It is described to obtain the positioning map based on semantic feature according to the classification results in one embodiment, comprising:
Obtain the semantic feature and the coordinate information;
Semantic chunk region according to the semantic feature and the coordinate information, in corresponding description map;
According to the coordinate information, observation visual angle of the configuration pin to the semantic chunk region;
According to the semantic feature, the coordinate information and the observation visual angle, obtain being made of multiple semantic chunk regions
The positioning map.
In one embodiment, observation visual angle of the configuration pin to the semantic chunk region, comprising:
According to the corresponding different positioning accuracies of different objects direction of observation in the panoramic view data, different observation views is configured
Angle;
The observation visual angle includes at least: the visual angle at least two directions in east, south, west, north.
In one embodiment, the method also includes: it is directed to the panoramic view data, divides the observation in the horizontal direction
Direction;Alternatively,
For the panoramic view data, the direction of observation is divided in the pitch direction.
In one embodiment, at least one described point of image data input to be processed by current target object acquisition
Class model, in conjunction with the positioning map, positioning obtains the direction of the target object, comprising:
In the disaggregated model, image is carried out at least one image data to be processed according to semantic segmentation strategy and is located in advance
Reason retains the static information at least one described image data to be processed;
It is fixed by the positioning map according to the corresponding semantic feature of the static information, coordinate information and observation visual angle
Position obtains the direction of the target object.
It is described according to the corresponding semantic feature of the static information, coordinate information and observation visual angle in one embodiment, lead to
The positioning map is crossed to position to obtain the direction of the target object, comprising:
Semantic chunk region in the static information and the positioning map is subjected to images match, is obtained and the static state
Information has at least one target semantic chunk region of image similarity, at least one described target semantic chunk region corresponds to same
A coordinate information;
When at least one described target semantic chunk region is there are when the overlapping of multiple observation visual angles, according to multi-angle of view overlay region
Domain obtains positional parameter;
According to the positional parameter, positioning obtains the direction of the target object.
Second aspect, the embodiment of the invention provides a kind of vision positioning device, described device includes:
Acquisition unit, for acquiring panoramic view data;
Taxon, for being divided using the panoramic view data as classifying in training sample input disaggregated model
Class result;
Map generation unit, for obtaining the positioning map based on semantic feature according to the classification results;
Positioning unit, at least one image data to be processed for acquiring current target object input the classification mould
Type, in conjunction with the positioning map, positioning obtains the direction of the target object.
In one embodiment, the taxon further comprises:
Subelement is pre-processed, is used in the disaggregated model, according to semantic segmentation strategy in the panoramic view data
At least one image data carries out image preprocessing, obtains pre-processed results, and the pre-processed results are at least one described figure
As the partial image region in data;
Classification subelement obtains the language for corresponding to the partial image region for classifying to the pre-processed results
The coordinate information of adopted feature and the corresponding partial image region;
The semantic feature and the coordinate information are determined as the classification results.
In one embodiment, the pretreatment subelement is further used for:
From at least one described image data, the object to remain static at the appointed time section is identified;
Using the corresponding image-region of the object as static information;
Using the static information as the pre-processed results.
In one embodiment, the map generation unit further comprises:
Acquisition of information subelement, for obtaining the semantic feature and the coordinate information;
Region description subelement, for according to the semantic feature and the coordinate information, the corresponding language described in map
Adopted block region;
Visual angle configures subelement, for according to the coordinate information, observation visual angle of the configuration pin to the semantic chunk region;
Map generates subelement, for according to the semantic feature, the coordinate information and the observation visual angle, obtain by
The positioning map that multiple semantic chunk regions are constituted.
In one embodiment, the visual angle configures subelement, is further used for:
According to the corresponding different positioning accuracies of different objects direction of observation in the panoramic view data, different observation views is configured
Angle;
The observation visual angle includes at least: the visual angle at least two directions in east, south, west, north.
In one embodiment, described device further includes direction division unit, is used for:
For the panoramic view data, the direction of observation is divided in the horizontal direction;Alternatively,
For the panoramic view data, the direction of observation is divided in the pitch direction.
In one embodiment, the positioning unit further comprises:
Image preprocessing subelement, for waiting locating at least one according to semantic segmentation strategy in the disaggregated model
It manages image data and carries out image preprocessing, retain the static information at least one described image data to be processed;
Object locator unit is used for according to the corresponding semantic feature of the static information, coordinate information and observation visual angle,
It positions to obtain the direction of the target object by the positioning map.
In one embodiment, the object locator unit is used for:
Semantic chunk region in the static information and the positioning map is subjected to images match, is obtained and the static state
Information has at least one target semantic chunk region of image similarity, at least one described target semantic chunk region corresponds to same
A coordinate information;
When at least one described target semantic chunk region is there are when the overlapping of multiple observation visual angles, according to multi-angle of view overlay region
Domain obtains positional parameter;
According to the positional parameter, positioning obtains the direction of the target object.
The third aspect, the embodiment of the invention provides a kind of vision positioning device, the function of described device can be by hard
Part is realized, corresponding software realization can also be executed by hardware.The hardware or software include one or more and above-mentioned function
It can corresponding module.
It include processor and memory in the structure of described device in a possible design, the memory is used for
Storage supports described device to execute the program of any above-mentioned vision positioning method, the processor is configured to described for executing
The program stored in memory.Described device can also include communication interface, be used for and other equipment or communication.
Fourth aspect, the embodiment of the invention provides a kind of computer readable storage mediums, for storing information processing apparatus
Set computer software instructions used comprising for executing program involved in any above-mentioned vision positioning method.
A technical solution in above-mentioned technical proposal have the following advantages that or the utility model has the advantages that
In the embodiment of the present invention, panoramic view data is acquired;Using the panoramic view data as in training sample input disaggregated model
Classify, obtains classification results;The positioning map based on semantic feature is obtained according to the classification results;By current goal pair
As at least one described disaggregated model of image data input to be processed of acquisition, in conjunction with the positioning map, positioning obtains described
The direction of target object.Led to using the embodiment of the present invention for being difficult to the case where determining direction (or direction) of target object
It crosses using the panoramic view data of extraction as classifying in training sample input disaggregated model, is obtained according to classification results based on semanteme
Then the positioning map of feature applies the disaggregated model and positioning map, can orient the direction of current target object.Due to
The direction that target object can be oriented by disaggregated model and positioning map, haves no need to change current hardware, therefore, uses
Existing magnetometer can realize accurate direction positioning, while reduce the hardware cost of upgrading magnetometer.
Above-mentioned general introduction is merely to illustrate that the purpose of book, it is not intended to be limited in any way.Except foregoing description
Schematical aspect, except embodiment and feature, by reference to attached drawing and the following detailed description, the present invention is further
Aspect, embodiment and feature, which will be, to be readily apparent that.
Detailed description of the invention
In the accompanying drawings, unless specified otherwise herein, otherwise indicate the same or similar through the identical appended drawing reference of multiple attached drawings
Component or element.What these attached drawings were not necessarily to scale.It should be understood that these attached drawings depict only according to the present invention
Disclosed some embodiments, and should not serve to limit the scope of the present invention.
Fig. 1 shows the flow chart of vision positioning method according to an embodiment of the present invention.
Fig. 2 shows the flow charts of vision positioning method according to an embodiment of the present invention.
Fig. 3 shows vision positioning schematic diagram of a scenario according to an embodiment of the present invention.
Fig. 4 shows the structural block diagram of vision positioning device according to an embodiment of the present invention.
Fig. 5 shows the structural block diagram of vision positioning device according to an embodiment of the present invention.
Specific embodiment
Hereinafter, certain exemplary embodiments are simply just described.As one skilled in the art will recognize that
Like that, without departing from the spirit or scope of the present invention, described embodiment can be modified by various different modes.
Therefore, attached drawing and description are considered essentially illustrative rather than restrictive.
In the related technology, in an application scenarios, with different view (as visual angle from left to right, visual angle from right to left,
Visual angle overlooked from top to bottom etc.) see the same target object (such as building, vehicle, mobile phone terminal, one in surrounding enviroment
Tree or the street lamp of curbside etc.), if checking that result is similar, it is difficult to determine direction (or the court of the target object
To), it needs to position the target object.Direction is usually to be determined using magnetometer.But fields are waited at the parting of the ways
Jing Chu, since there are various interference for surrounding, such as the vehicle (metal shell will affect magnetometer), electric pole and the column that largely pass by
Bar (metal material will affect magnetometer), these can all bring very big electromagnetic interference, so that being used for detected target object direction
Magnetometer biggish error can be brought after electromagnetic interference, cause direction not determine accurately.
Magnetometer is also earth magnetism, magnetic strength device, can be used for testing magnetic field strength and direction, positioning target object is (as currently set
It is standby) orientation.The principle of magnetometer is similar with compass principle, can measure on target object and four corners of the world four direction
Angle, so, gyroscope is known " target object has turned a body ", and accelerometer knows " target object has been gone ahead several meters again ",
And magnetometer then knows " target object be westwards direction ".In practical applications, since error is modified and compensates needs, in addition to
Magnetometer, can be combined with gyroscope and accelerometer positions together, using the speciality of every kind of sensor, allow final positioning result
It is more acurrate, for example positioning result is obtained in combination with magnetic direction and direction motion conditions.
However, due to electromagnetic interference, it, can only be with more if to accurately determine the direction of target object in above-mentioned scene
Advanced magnetometer, this certainly will increase hardware cost.In this regard, reducing hardware cost, specifically while needing to realize accurate positionin
It is the vision positioning processing for proposing the embodiment of the present invention.
Fig. 1 shows the flow chart of vision positioning method according to an embodiment of the present invention.As shown in Figure 1, the process includes:
Step 101, acquisition panoramic view data.
The panoramic view data is inputted in disaggregated model as training sample and is classified by step 102, obtains classification knot
Fruit.
In one example, panoramic view data can be the image data acquired at the parting of the ways, the row including crossroad walking
People, vehicle, building, mobile phone terminal, a tree in surrounding enviroment or street lamp of curbside etc..It, can when acquiring panoramic view data
It is same with (visual angle such as from left to right, visual angle from right to left, the from top to bottom visual angle overlooked) acquisition with different view
A target object (such as vehicle, pedestrian, building, mobile phone terminal, a tree in surrounding enviroment or the street lamp of curbside),
It (visual angle such as from left to right, visual angle from right to left, the visual angle overlooked from top to bottom) can acquire not with different view
Same target object.
Using these panoramic view datas for using above-mentioned acquisition mode to obtain as training sample, training sample is inputted into disaggregated model
In classify, obtain classification results, classification results can be in any image data of panoramic view data and distinguish each target pair
As the image-region at place, which has semantic feature and corresponding coordinate information.For example, semanteme point can be passed through
Class knows which region is vehicle in image data, which region is pedestrian, which region is building, which region is hand
Machine terminal, which region is a tree in surrounding enviroment or which region is street lamp of curbside etc..
Step 103 obtains the positioning map based on semantic feature according to the classification results.
In one example, classification results include in any image data of panoramic view data for distinguishing each target object place
Image-region, which has semantic feature and corresponding coordinate information.It is the image according to corresponding coordinate information
Area assignment observation visual angle can correspond at least two observation visual angles for the same target object.Target object is in three-dimensional space
Between in be that a hexahedron correspondingly, observation visual angle can be divided in three-dimensional space is certainly not limited to this, can also be
Two-dimensional space divides, and can also be mapped to three-dimensional space etc. after two-dimensional space division.The example of observation visual angle, for example, from
Left-to-right visual angle, visual angle from right to left, the visual angle overlooked from top to bottom etc..
At least one image data to be processed of current target object acquisition is inputted the disaggregated model, knot by step 104
The positioning map is closed, positioning obtains the direction of the target object.
In one example, target object may include: vehicle, pedestrian, building, mobile phone terminal, one in surrounding enviroment
Tree or the street lamp of curbside etc..For example, target object is mobile phone terminal, user has clapped a current location using the mobile phone terminal
Under scene image (scene image can be the acquisition image under different perspectives), which can be the image to be processed
Data.Due to 101-104 through the above steps, the training sample constituted by inputting panoramic view data, trained obtain can
With the disaggregated model of classification, and available corresponding classification results.So, in practical applications, still input image data
(such as image data to be processed) in conjunction with obtained positioning map, that is, utilizes and step 101-104 into existing disaggregated model
In the same processing logic, be directly positioned to the direction of mobile phone terminal, naturally it is also possible to using same processing logic positioning
It obtains holding pose of people of the mobile phone terminal etc., it can also be by being positioned to set phase to the laggard line position in the direction of mobile phone terminal
The pose is derived to transformation.
Using the embodiment of the present invention, above-mentioned processing logic can be located at terminal and acquire side, can also be located at the service on backstage
Device side, it may be assumed that the excellent of direction positioning is carried out using the processing logic at front end (target object such as mobile phone terminal or vehicle termination etc.)
Change, and in background server, carries out the optimization of direction positioning in the cluster being such as made of server cluster using the processing logic.
For being difficult to the case where determining direction (or direction) of target object, the embodiment of the present invention passes through using panoramic view data as training
Sample and inputting in disaggregated model is classified, and the positioning map based on semantic feature is obtained, then using the disaggregated model and
Positioning map orients the corresponding direction of current target object (or direction).Due to by disaggregated model and positioning map just
The direction that target object can be oriented haves no need to change current hardware, therefore, can realize standard with existing magnetometer
True direction positioning, while reducing the hardware cost of upgrading magnetometer.
Fig. 2 shows the flow charts of vision positioning method according to an embodiment of the present invention.As shown in Fig. 2, the process includes:
Step 201, acquisition panoramic view data.
Step 202 inputs panoramic view data as training sample in disaggregated model, according to semantic segmentation in disaggregated model
Strategy carries out image preprocessing at least one image data in the panoramic view data, obtains pre-processed results, the pre- place
Managing result is the partial image region at least one described image data.
Step 203 classifies to the pre-processed results, obtain corresponding to the partial image region semantic feature and
The semantic feature and the coordinate information are determined as the classification and tied by the coordinate information of the corresponding partial image region
Fruit.
202-203 through the above steps may be implemented using panoramic view data as point in training sample input disaggregated model
Class, obtained classification results include: the semantic feature of partial image region and corresponding coordinate letter at least one image data
Breath.Partial image region can be using semantic segmentation extract image in static information (such as building, board etc. for a long time
The image-region that will not become), since static information is the image-region that will not become for a long time, have number for sort operation
According to stability and operational reliability, therefore, static information is extracted and is classified, can achieve accurate classification results,
After then obtaining positioning map subsequently through the classification results, available accurate direction locating effect.Wherein, the parts of images
Region can be the semantic chunk region in the semantic chunk region in positioning map, or corresponding positioning map.
In one example, panoramic view data can be the image data acquired at the parting of the ways, the row including crossroad walking
People, vehicle, building, mobile phone terminal, a tree in surrounding enviroment or street lamp of curbside etc..It, can when acquiring panoramic view data
It is same with (visual angle such as from left to right, visual angle from right to left, the from top to bottom visual angle overlooked) acquisition with different view
A target object (such as vehicle, pedestrian, building, mobile phone terminal, a tree in surrounding enviroment or the street lamp of curbside),
It (visual angle such as from left to right, visual angle from right to left, the visual angle overlooked from top to bottom) can acquire not with different view
Same target object.
Using these panoramic view datas for using above-mentioned acquisition mode to obtain as training sample, training sample is inputted into disaggregated model
In classify, obtain classification results, classification results can be in any image data of panoramic view data and distinguish each target pair
As the image-region at place, which has semantic feature and corresponding coordinate information.For example, semanteme point can be passed through
Class knows which region is vehicle in image data, which region is pedestrian, which region is building, which region is hand
Machine terminal, which region is a tree in surrounding enviroment or which region is street lamp of curbside etc..
Step 204 obtains the positioning map based on semantic feature according to the classification results.
In one example, classification results include in any image data of panoramic view data for distinguishing each target object place
Image-region, which has semantic feature and corresponding coordinate information.It is the image according to corresponding coordinate information
Area assignment observation visual angle can correspond at least two observation visual angles for the same target object.Target object is in three-dimensional space
Between in be that a hexahedron correspondingly, observation visual angle can be divided in three-dimensional space is certainly not limited to this, can also be
Two-dimensional space divides, and can also be mapped to three-dimensional space etc. after two-dimensional space division.The example of observation visual angle, for example, from
Left-to-right visual angle, visual angle from right to left, the visual angle overlooked from top to bottom etc..
At least one image data to be processed of current target object acquisition is inputted the disaggregated model, knot by step 205
The positioning map is closed, positioning obtains the direction of the target object.
In one example, target object may include: vehicle, pedestrian, building, mobile phone terminal, one in surrounding enviroment
Tree or the street lamp of curbside etc..For example, target object is mobile phone terminal, user has clapped a current location using the mobile phone terminal
Under scene image (scene image can be the acquisition image under different perspectives), which can be the image to be processed
Data.Due to 201-205 through the above steps, the training sample constituted by inputting panoramic view data, trained obtain can
With the disaggregated model of classification, and available corresponding classification results.So, in practical applications, still input image data
(such as image data to be processed) in conjunction with obtained positioning map, that is, utilizes and step 201-205 into existing disaggregated model
In the same processing logic, be directly positioned to the direction of mobile phone terminal, naturally it is also possible to using same processing logic positioning
It obtains holding pose of people of the mobile phone terminal etc., it can also be by being positioned to set phase to the laggard line position in the direction of mobile phone terminal
The pose is derived to transformation.
Using the embodiment of the present invention, above-mentioned processing logic can be located at terminal and acquire side, can also be located at the service on backstage
Device side, it may be assumed that the excellent of direction positioning is carried out using the processing logic at front end (target object such as mobile phone terminal or vehicle termination etc.)
Change, and in background server, carries out the optimization of direction positioning in the cluster being such as made of server cluster using the processing logic.
For being difficult to the case where determining direction (or direction) of target object, the embodiment of the present invention passes through using panoramic view data as training
Sample and inputting in disaggregated model is classified, and the positioning map based on semantic feature is obtained, then using the disaggregated model and
Positioning map orients the corresponding direction of current target object (or direction).Due to by disaggregated model and positioning map just
The direction that target object can be oriented haves no need to change current hardware, therefore, can realize standard with existing magnetometer
True direction positioning, while reducing the hardware cost of upgrading magnetometer.
In one embodiment, figure is carried out at least one image data in the panoramic view data according to semantic segmentation strategy
As pretreatment, pre-processed results are obtained, comprising: from least one described image data, identify place at the appointed time section
In the object (such as building, the object that board etc. will not move for a long time) of stationary state, by the corresponding image-region of the object
As static information, using the static information as the pre-processed results.
It is described to obtain the positioning map based on semantic feature according to the classification results in one embodiment, comprising: to obtain
The semantic feature and the coordinate information;Language according to the semantic feature and the coordinate information, in corresponding description map
Adopted block region;According to the coordinate information, observation visual angle of the configuration pin to the semantic chunk region;According to the semantic feature,
The coordinate information and the observation visual angle obtain the positioning map being made of multiple semantic chunk regions.
In one example, the panoramic view data acquired using map, building carry semantic feature and Inertial Measurement Unit (IMU,
Inertial measurement unit) information map, that is to say, that be to be constructed by panoramic view data based on semanteme
The positioning map of feature can solve the direction orientation problem to target object.Including following content, i.e., how to construct this and be based on
The positioning map of semantic feature, and vision positioning is carried out according to the image that target object uploads.
For constructing semantic map, details are as follows:
1, the panoramic view data of General maps acquisition has relatively accurate GPS, IMU, magnetometer etc. and position and side
To related location information, therefore the coordinate of the camera site of every Zhang Quanjing's figure, direction are (for the bat of the variant part of panorama sketch
Take the photograph direction) it is all existing and more accurate.
2, static information (such as the figure that will not become for a long time such as building, board in image is extracted using semantic segmentation
As region) and classify.There are semantic information and coordinate information in each region of this sampled images.
3, database (database) such as in map data base one observation visual angle of each semantic chunk region assignment or
Observation visual angle, is finally put in storage by direction of observation (such as all directions, be also possible to specific view angle), with building based on semanteme
The positioning map (semantics for short map) of feature.It include multiple semantic regions in positioning map based on semantic feature, and corresponding
Semantic feature, coordinate information and the directional information used for positioning of each semantic region.
In one embodiment, observation visual angle of the configuration pin to the semantic chunk region, comprising: according in the panoramic view data
The corresponding different positioning accuracies of different objects direction of observation, configure different observation visual angles.Wherein, the observation visual angle at least wraps
It includes: the visual angle at least two directions in east, south, west, north.For the panoramic view data, the sight is divided in the horizontal direction
Examine direction;Alternatively, being directed to the panoramic view data, the direction of observation is divided in the pitch direction.For example, if only in order to full
The focal need (such as determining all directions four direction) of sufficient lower accuracy, can be with the positioning accuracy of lower granularity for seeing
It examines direction to be divided, such as 360 degree of panoramic view data is divided into the observation visual angle of above-mentioned all directions four, to be each
The corresponding observation visual angle of semantic chunk area assignment corresponding to direction.It is of course also possible in order to realize higher positioning accuracy, it can
Panoramic view data is divided into more observation visual angles.In addition, panoramic view data can also not only divide sight in the horizontal direction
Direction is examined, can also be divided in the pitch direction.
In one embodiment, at least one described point of image data input to be processed by current target object acquisition
Class model, in conjunction with the positioning map, positioning obtains the direction of the target object, comprising: in the disaggregated model, according to
Semantic segmentation strategy carries out image preprocessing at least one image data to be processed, retains at least one described image to be processed
Static information in data;According to the corresponding semantic feature of the static information, coordinate information and observation visual angle, by described fixed
Position Orientation on map obtains the direction of the target object.
In one embodiment, the positioning to target object direction is realized using above-mentioned disaggregated model and above-mentioned positioning map,
In simple terms, the fan-shaped region of overlapping can be to look at.Fan-shaped region is that in semantic region localization region in the embodiment of the present invention
An example, the not concrete shape of limited area.According to the corresponding semantic feature of the static information, coordinate information and observation
Visual angle positions to obtain the direction of the target object by the positioning map, comprising: by the static information and the positioning
Semantic chunk region in map carries out images match, obtains at least one target for having image similarity with the static information
Semantic chunk region, at least one described target semantic chunk region correspond to the same coordinate information;When at least one described target language
Adopted block region obtains positional parameter according to multi-angle of view overlapping region there are when the overlapping of multiple observation visual angles;According to the positioning
Parameter, positioning obtain the direction of the target object.
In one example, the panoramic view data acquired using map, building carry semantic feature and Inertial Measurement Unit (IMU,
Inertial measurement unit) information map, that is to say, that be to be constructed by panoramic view data based on semanteme
The positioning map of feature can solve the direction orientation problem to target object.Including following content, i.e., how to construct this and be based on
The positioning map of semantic feature, and vision positioning is carried out according to the image that target object uploads.
How the positioning map based on semantic feature is constructed, previously herein by the agency of mistake, now with regard to how according to target
The image that object uploads carries out vision positioning, i.e., how to realize using above-mentioned disaggregated model and above-mentioned positioning map to target object
The positioning in direction, details are as follows:
1, semantic segmentation is carried out to the acquisition image to be processed of upload, retains the static information in image.
2, image, semantic region in each semantic region in acquisition image and map data base is matched, thus
The matching of many semantic levels is obtained.Wherein, semantic region can also be known as semantic chunk.
3, since when constructing positioning map, each semantic region has observation scope (such as corresponding one or more observations
Visual angle), matching each in this way will generate one piece of fan-shaped region in 2D plane, and the most intensive region of intersection can be obtained slightly
Direction slightly or pose (pose), and the observation visual angle of acquisition image.It is fan-shaped if observation visual angle further includes pitch angle
Region is rather than the fan-shaped region in 2D plane in the 3 d space.
4, if it is intended to obtaining more accurate direction or towards (pose), then needing will positioningly using SFM technology
Figure is built into point cloud data.Then the matching of 2D to 3D mapping is continued to execute in above-mentioned matching.
For SFM, include at least: feature extraction is (generally using SIFT operator, because it is with scale and invariable rotary
Property) the step of;Matching and the step of establish tracking image (such as track list), such as by Euclidean distance to image to two-by-two
Match;The step of initialisation image pair, to find the opposite acquisition maximum image pair of equipment (such as camera) baseline;Initialisation image pair
Relative orientation the step of;The step of sparse reconstruction SFM, waits.It should be pointed out that in the pose information for obtaining target object
Except, observer (user) position can also be pushed away come counter using the overlapping of multiple observation visual angles in above-mentioned localization process, from
And user can be further obtained the location of when shooting image, rather than just target object (current device)
Pose information.It should be pointed out that the panoramic view data in the embodiment of the present invention can also be a cloud map, so as to obtain more
Accurate pose information and position.Using the embodiment of the present invention, in the case of higher hardware cost magnetometer can not be used, determine
Direction of the current devices such as mobile phone terminal in the scene such as crossroad for being difficult to determine direction, it might even be possible to determine specific position
It sets.
Using example:
Fig. 3 shows vision positioning schematic diagram of a scenario according to an embodiment of the present invention, as shown in figure 3, acquisition panoramic view data (is adopted
Collect image), the acquisition image is inputted into disaggregated model, image can be subjected to digitized processing before the input.This is adopted
During collection image in training sample input disaggregated model as being classified, first the acquisition image is pre-processed, with
Retain the static information (building, board etc.) in image, is classified in disaggregated model based on semantic feature, obtain static state
The semantic feature and coordinate information of image-region where information, using the corresponding semantic feature of the image-region and coordinate information as
Classification results output.Acquisition image can be the image data acquired at the parting of the ways, the pedestrian including crossroad walking, vehicle
, a tree in building, mobile phone terminal, surrounding enviroment or the street lamp of curbside etc. can be with for the acquisition image
(visual angle such as from left to right, visual angle from right to left, the visual angle overlooked from top to bottom) acquisition is same with different view
Target object (such as vehicle, pedestrian, building, mobile phone terminal, a tree in surrounding enviroment or the street lamp of curbside), can also
It is different with (visual angle such as from left to right, visual angle from right to left, the from top to bottom visual angle overlooked) acquisition with different view
Target object.By semantic classification, can know which region is vehicle in image data, which region is pedestrian, which area
Domain is building, which region is mobile phone terminal, which region is a tree in surrounding enviroment or which region is the road of curbside
Lamp etc..It is obtained based on semantic feature positioningly according at least one observation visual angle of semantic feature, coordinate information and assignment
Figure.Wherein it is possible to be that the image-region assignment corresponds to observation visual angle according to the coordinate information, it, can for the same target object
With corresponding at least two observation visual angles.Target object is a hexahedron in three dimensions, correspondingly, observation visual angle can be
Three-dimensional space is divided, this is certainly not limited to, and can also be divided in two-dimensional space, can also be mapped after two-dimensional space division
Arrive three-dimensional space etc..The example of observation visual angle, for example, visual angle from left to right, visual angle from right to left, overlooking from top to bottom
Visual angle etc..Finally, positioning map and disaggregated model based on semantic feature carry out direction positioning to target object.
Fig. 4 shows the structural block diagram of vision positioning device of the embodiment of the present invention, and described device includes: acquisition unit 31, uses
In acquisition panoramic view data;Taxon 32 is divided for inputting in disaggregated model the panoramic view data as training sample
Class obtains classification results;Map generation unit 33, for being obtained based on semantic feature positioningly according to the classification results
Figure;Positioning unit 34, at least one image data to be processed for acquiring current target object input the disaggregated model,
In conjunction with the positioning map, positioning obtains the direction of the target object.
In one embodiment, the taxon further comprises: pretreatment subelement, in the disaggregated model
In, image preprocessing is carried out at least one image data in the panoramic view data according to semantic segmentation strategy, obtains pre- place
Reason is as a result, the pre-processed results are the partial image region at least one described image data;
Classification subelement obtains the language for corresponding to the partial image region for classifying to the pre-processed results
The coordinate information of adopted feature and the corresponding partial image region;The semantic feature and the coordinate information are determined as described
Classification results.
In one embodiment, the pretreatment subelement is further used for: from least one described image data, knowing
The object that Chu do not remain static in section at the appointed time;Using the corresponding image-region of the object as static information;It will
The static information is as the pre-processed results.
In one embodiment, the map generation unit further comprises: acquisition of information subelement, described for obtaining
Semantic feature and the coordinate information;Region description subelement, for corresponding to according to the semantic feature and the coordinate information
Semantic chunk region in map is described;Visual angle configures subelement, for according to the coordinate information, configuration pin to be to the semantic chunk
The observation visual angle in region;Map generates subelement, for being regarded according to the semantic feature, the coordinate information and the observation
Angle obtains the positioning map being made of multiple semantic chunk regions.
In one embodiment, the visual angle configures subelement, is further used for: according to different objects in the panoramic view data
The corresponding different positioning accuracies of direction of observation, configure different observation visual angles;The observation visual angle includes at least: east, south, west,
The visual angle at least two directions in north.
In one embodiment, described device further includes direction division unit, is used for: the panoramic view data is directed to, in level
The direction of observation is divided on direction;Alternatively, being directed to the panoramic view data, the direction of observation is divided in the pitch direction.
In one embodiment, the positioning unit further comprises: image preprocessing subelement, in the classification
In model, image preprocessing is carried out at least one image data to be processed according to semantic segmentation strategy, retains described at least one
Static information in a image data to be processed;Object locator unit, for corresponding semantic special according to the static information
Sign, coordinate information and observation visual angle, position to obtain the direction of the target object by the positioning map.
In one embodiment, the object locator unit is used for: will be in the static information and the positioning map
Semantic chunk region carries out images match, obtains at least one the target semantic chunk area for having image similarity with the static information
Domain, at least one described target semantic chunk region correspond to the same coordinate information;When at least one described target semantic chunk region
There are when the overlapping of multiple observation visual angles, positional parameter is obtained according to multi-angle of view overlapping region;According to the positional parameter, positioning
Obtain the direction of the target object.
The function of each module in each device of the embodiment of the present invention may refer to the corresponding description in the above method, herein not
It repeats again.
Fig. 5 shows the structural block diagram of information processing unit according to an embodiment of the present invention.As shown in figure 5, the device includes:
Memory 910 and processor 920 are stored with the computer program that can be run on processor 920 in memory 910.Processor
The automatic Pilot method in above-described embodiment is realized when 920 execution computer program.The quantity of memory 910 and processor 920
It can be one or more.
The device further include: communication interface 930 carries out data interaction for being communicated with external device.
Memory 910 may include high speed RAM memory, it is also possible to further include nonvolatile memory (non-
Volatile memory), a for example, at least magnetic disk storage.
If memory 910, processor 920 and the independent realization of communication interface 930, memory 910,920 and of processor
Communication interface 930 can be connected with each other by bus and complete mutual communication.Bus can be industry standard architecture
(ISA, Industry Standard Architecture) bus, external equipment interconnection (PCI, Peripheral
Component) bus or extended industry-standard architecture (EISA, Extended Industry Standard
Component) bus etc..Bus can be divided into address bus, data/address bus, control bus etc..For convenient for indicating, in Fig. 5 only
It is indicated with a thick line, it is not intended that an only bus or a type of bus.
Optionally, in specific implementation, if memory 910, processor 920 and communication interface 930 are integrated in one piece of core
On piece, then memory 910, processor 920 and communication interface 930 can complete mutual communication by internal interface.
The embodiment of the invention provides a kind of computer readable storage mediums, are stored with computer program, the program quilt
Processor realizes any method in above-described embodiment when executing.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " example ", " specifically show
The description of example " or " some examples " etc. means specific features, structure, material or spy described in conjunction with this embodiment or example
Point is included at least one embodiment or example of the invention.Moreover, particular features, structures, materials, or characteristics described
It may be combined in any suitable manner in any one or more of the embodiments or examples.In addition, without conflicting with each other, this
The technical staff in field can be by the spy of different embodiments or examples described in this specification and different embodiments or examples
Sign is combined.
In addition, term " first ", " second " are used for descriptive purposes only and cannot be understood as indicating or suggesting relative importance
Or implicitly indicate the quantity of indicated technical characteristic." first " is defined as a result, the feature of " second " can be expressed or hidden
It include at least one this feature containing ground.In the description of the present invention, the meaning of " plurality " is two or more, unless otherwise
Clear specific restriction.
Any process described otherwise above or method description are construed as in flow chart or herein, and expression includes
It is one or more for realizing specific logical function or process the step of executable instruction code module, segment or portion
Point, and the range of the preferred embodiment of the present invention includes other realization, wherein can not press shown or discussed suitable
Sequence, including according to related function by it is basic simultaneously in the way of or in the opposite order, Lai Zhihang function, this should be of the invention
Embodiment person of ordinary skill in the field understood.
Expression or logic and/or step described otherwise above herein in flow charts, for example, being considered use
In the order list for the executable instruction for realizing logic function, may be embodied in any computer-readable medium, for
Instruction execution system, device or equipment (such as computer based system, including the system of processor or other can be held from instruction
The instruction fetch of row system, device or equipment and the system executed instruction) it uses, or combine these instruction execution systems, device or set
It is standby and use.For the purpose of this specification, " computer-readable medium ", which can be, any may include, stores, communicates, propagates or pass
Defeated program is for instruction execution system, device or equipment or the dress used in conjunction with these instruction execution systems, device or equipment
It sets.The more specific example (non-exhaustive list) of computer-readable medium include the following: there is the electricity of one or more wirings
Interconnecting piece (electronic device), portable computer diskette box (magnetic device), random access memory (RAM), read-only memory
(ROM), erasable edit read-only storage (EPROM or flash memory), fiber device and portable read-only memory
(CDROM).In addition, computer-readable medium can even is that the paper that can print described program on it or other suitable Jie
Matter, because can then be edited, be interpreted or when necessary with other for example by carrying out optical scanner to paper or other media
Suitable method is handled electronically to obtain described program, is then stored in computer storage.
It should be appreciated that each section of the invention can be realized with hardware, software, firmware or their combination.Above-mentioned
In embodiment, software that multiple steps or method can be executed in memory and by suitable instruction execution system with storage
Or firmware is realized.It, and in another embodiment, can be under well known in the art for example, if realized with hardware
Any one of column technology or their combination are realized: having a logic gates for realizing logic function to data-signal
Discrete logic, with suitable combinational logic gate circuit specific integrated circuit, programmable gate array (PGA), scene
Programmable gate array (FPGA) etc..
Those skilled in the art are understood that realize all or part of step that above-described embodiment method carries
It suddenly is that relevant hardware can be instructed to complete by program, the program can store in a kind of computer-readable storage medium
In matter, which when being executed, includes the steps that one or a combination set of embodiment of the method.
It, can also be in addition, each functional unit in each embodiment of the present invention can integrate in a processing module
It is that each unit physically exists alone, can also be integrated in two or more units in a module.Above-mentioned integrated mould
Block both can take the form of hardware realization, can also be realized in the form of software function module.The integrated module is such as
Fruit is realized and when sold or used as an independent product in the form of software function module, also can store in a computer
In readable storage medium storing program for executing.The storage medium can be read-only memory, disk or CD etc..
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can readily occur in its various change or replacement,
These should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with the guarantor of the claim
It protects subject to range.
Claims (18)
1. a kind of vision positioning method, which is characterized in that the described method includes:
Acquire panoramic view data;
Using the panoramic view data as classifying in training sample input disaggregated model, classification results are obtained;
The positioning map based on semantic feature is obtained according to the classification results;
At least one image data to be processed of current target object acquisition is inputted into described disaggregated model, in conjunction with it is described positioningly
Figure, positioning obtain the direction of the target object.
2. the method according to claim 1, wherein described using the panoramic view data as training sample input point
Classify in class model, obtain classification results, comprising:
In the disaggregated model, figure is carried out at least one image data in the panoramic view data according to semantic segmentation strategy
As pretreatment, pre-processed results are obtained, the pre-processed results are the partial image region at least one described image data;
Classify to the pre-processed results, obtains the semantic feature for corresponding to the partial image region and the corresponding part
The coordinate information of image-region;
The semantic feature and the coordinate information are determined as the classification results.
3. according to the method described in claim 2, it is characterized in that, it is described according to semantic segmentation strategy in the panoramic view data
At least one image data carry out image preprocessing, obtain pre-processed results, comprising:
From at least one described image data, the object to remain static at the appointed time section is identified;
Using the corresponding image-region of the object as static information;
Using the static information as the pre-processed results.
4. according to the method described in claim 2, it is characterized in that, described obtain according to the classification results based on semantic feature
Positioning map, comprising:
Obtain the semantic feature and the coordinate information;
Semantic chunk region according to the semantic feature and the coordinate information, in corresponding description map;
According to the coordinate information, observation visual angle of the configuration pin to the semantic chunk region;
According to the semantic feature, the coordinate information and the observation visual angle, the institute being made of multiple semantic chunk regions is obtained
State positioning map.
5. according to the method described in claim 4, it is characterized in that, the configuration pin regards the observation in the semantic chunk region
Angle, comprising:
According to the corresponding different positioning accuracies of different objects direction of observation in the panoramic view data, different observation visual angles is configured;
The observation visual angle includes at least: the visual angle at least two directions in east, south, west, north.
6. direction according to claim 5, which is characterized in that the method also includes: it is directed to the panoramic view data, in water
Square the direction of observation is divided upwards;Alternatively,
For the panoramic view data, the direction of observation is divided in the pitch direction.
7. method according to claim 1 to 6, which is characterized in that described to acquire current target object extremely
A few image data to be processed inputs the disaggregated model, and in conjunction with the positioning map, positioning obtains the target object
Direction, comprising:
In the disaggregated model, image preprocessing is carried out at least one image data to be processed according to semantic segmentation strategy,
Retain the static information at least one described image data to be processed;
According to the corresponding semantic feature of the static information, coordinate information and observation visual angle, it is positioned to by the positioning map
To the direction of the target object.
8. the method according to the description of claim 7 is characterized in that it is described according to the corresponding semantic feature of the static information,
Coordinate information and observation visual angle position to obtain the direction of the target object by the positioning map, comprising:
Semantic chunk region in the static information and the positioning map is subjected to images match, is obtained and the static information
Has at least one target semantic chunk region of image similarity, at least one described target semantic chunk region corresponds to the same seat
Mark information;
When at least one described target semantic chunk region is there are when the overlapping of multiple observation visual angles, obtained according to multi-angle of view overlapping region
To positional parameter;
According to the positional parameter, positioning obtains the direction of the target object.
9. a kind of vision positioning device, which is characterized in that described device includes:
Acquisition unit, for acquiring panoramic view data;
Taxon, for obtaining classification knot using the panoramic view data as classifying in training sample input disaggregated model
Fruit;
Map generation unit, for obtaining the positioning map based on semantic feature according to the classification results;
Positioning unit, at least one image data to be processed for acquiring current target object input the disaggregated model,
In conjunction with the positioning map, positioning obtains the direction of the target object.
10. device according to claim 9, which is characterized in that the taxon further comprises:
Pre-process subelement, in the disaggregated model, according to semantic segmentation strategy in the panoramic view data at least
One image data carries out image preprocessing, obtains pre-processed results, and the pre-processed results are at least one described picture number
Partial image region in;
Subelement of classifying obtains corresponding to the semantic special of the partial image region for classifying to the pre-processed results
The coordinate information for the corresponding partial image region of seeking peace;
The semantic feature and the coordinate information are determined as the classification results.
11. device according to claim 10, which is characterized in that the pretreatment subelement is further used for:
From at least one described image data, the object to remain static at the appointed time section is identified;
Using the corresponding image-region of the object as static information;
Using the static information as the pre-processed results.
12. device according to claim 10, which is characterized in that the map generation unit further comprises:
Acquisition of information subelement, for obtaining the semantic feature and the coordinate information;
Region description subelement, for according to the semantic feature and the coordinate information, the corresponding semantic chunk described in map
Region;
Visual angle configures subelement, for according to the coordinate information, observation visual angle of the configuration pin to the semantic chunk region;
Map generates subelement, for obtaining by multiple according to the semantic feature, the coordinate information and the observation visual angle
The positioning map that semantic chunk region is constituted.
13. device according to claim 12, which is characterized in that the visual angle configures subelement, is further used for:
According to the corresponding different positioning accuracies of different objects direction of observation in the panoramic view data, different observation visual angles is configured;
The observation visual angle includes at least: the visual angle at least two directions in east, south, west, north.
14. device according to claim 13, which is characterized in that described device further includes direction division unit, is used for:
For the panoramic view data, the direction of observation is divided in the horizontal direction;Alternatively,
For the panoramic view data, the direction of observation is divided in the pitch direction.
15. the device according to any one of claim 9-14, which is characterized in that the positioning unit further comprises:
Image preprocessing subelement is used in the disaggregated model, according to semantic segmentation strategy at least one figure to be processed
As data progress image preprocessing, retain the static information at least one described image data to be processed;
Object locator unit, for passing through according to the corresponding semantic feature of the static information, coordinate information and observation visual angle
The positioning map positions to obtain the direction of the target object.
16. device according to claim 15, which is characterized in that the object locator unit is used for:
Semantic chunk region in the static information and the positioning map is subjected to images match, is obtained and the static information
Has at least one target semantic chunk region of image similarity, at least one described target semantic chunk region corresponds to the same seat
Mark information;
When at least one described target semantic chunk region is there are when the overlapping of multiple observation visual angles, obtained according to multi-angle of view overlapping region
To positional parameter;
According to the positional parameter, positioning obtains the direction of the target object.
17. a kind of vision positioning device, which is characterized in that described device includes:
One or more processors;
Storage device, for storing one or more programs;
When one or more of programs are executed by one or more of processors, so that one or more of processors
Realize such as method described in any item of the claim 1 to 8.
18. a kind of computer readable storage medium, is stored with computer program, which is characterized in that the program is held by processor
Such as method described in any item of the claim 1 to 8 is realized when row.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910586511.9A CN110298320B (en) | 2019-07-01 | 2019-07-01 | Visual positioning method, device and storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910586511.9A CN110298320B (en) | 2019-07-01 | 2019-07-01 | Visual positioning method, device and storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110298320A true CN110298320A (en) | 2019-10-01 |
CN110298320B CN110298320B (en) | 2021-06-22 |
Family
ID=68029799
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910586511.9A Active CN110298320B (en) | 2019-07-01 | 2019-07-01 | Visual positioning method, device and storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110298320B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111758118A (en) * | 2020-05-26 | 2020-10-09 | 蜂图科技有限公司 | Visual positioning method, device and equipment and readable storage medium |
CN113744352A (en) * | 2021-09-14 | 2021-12-03 | 北京观海科技发展有限责任公司 | Visual space calibration method, device and storage medium |
Citations (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106997466A (en) * | 2017-04-12 | 2017-08-01 | 百度在线网络技术(北京)有限公司 | Method and apparatus for detecting road |
CN206460481U (en) * | 2017-01-12 | 2017-09-01 | 刘曼 | One kind recognizes abnormal behaviour video system based on multi-angle |
CN107833236A (en) * | 2017-10-31 | 2018-03-23 | 中国科学院电子学研究所 | Semantic vision positioning system and method are combined under a kind of dynamic environment |
CN108596974A (en) * | 2018-04-04 | 2018-09-28 | 清华大学 | Dynamic scene robot localization builds drawing system and method |
WO2018213739A1 (en) * | 2017-05-18 | 2018-11-22 | TuSimple | System and method for image localization based on semantic segmentation |
CN109061703A (en) * | 2018-06-11 | 2018-12-21 | 百度在线网络技术(北京)有限公司 | Method, apparatus, equipment and computer readable storage medium used for positioning |
CN109117718A (en) * | 2018-07-02 | 2019-01-01 | 东南大学 | A kind of semantic map structuring of three-dimensional towards road scene and storage method |
CN109186586A (en) * | 2018-08-23 | 2019-01-11 | 北京理工大学 | One kind towards dynamically park environment while position and mixing map constructing method |
CN109272554A (en) * | 2018-09-18 | 2019-01-25 | 北京云迹科技有限公司 | A kind of method and system of the coordinate system positioning for identifying target and semantic map structuring |
US20190043203A1 (en) * | 2018-01-12 | 2019-02-07 | Intel Corporation | Method and system of recurrent semantic segmentation for image processing |
CN109357679A (en) * | 2018-11-16 | 2019-02-19 | 济南浪潮高新科技投资发展有限公司 | A kind of indoor orientation method based on significant characteristics identification |
CN109461211A (en) * | 2018-11-12 | 2019-03-12 | 南京人工智能高等研究院有限公司 | Semantic vector map constructing method, device and the electronic equipment of view-based access control model point cloud |
CN109584302A (en) * | 2018-11-27 | 2019-04-05 | 北京旷视科技有限公司 | Camera pose optimization method, device, electronic equipment and computer-readable medium |
CN109637177A (en) * | 2018-12-19 | 2019-04-16 | 斑马网络技术有限公司 | Vehicle positioning method, device, equipment and storage medium |
US20190130573A1 (en) * | 2017-10-30 | 2019-05-02 | Rakuten, Inc. | Skip architecture neural network machine and method for improved semantic segmentation |
CN109724603A (en) * | 2019-01-08 | 2019-05-07 | 北京航空航天大学 | A kind of Indoor Robot air navigation aid based on environmental characteristic detection |
CN109920055A (en) * | 2019-03-08 | 2019-06-21 | 视辰信息科技(上海)有限公司 | Construction method, device and the electronic equipment of 3D vision map |
-
2019
- 2019-07-01 CN CN201910586511.9A patent/CN110298320B/en active Active
Patent Citations (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN206460481U (en) * | 2017-01-12 | 2017-09-01 | 刘曼 | One kind recognizes abnormal behaviour video system based on multi-angle |
CN106997466A (en) * | 2017-04-12 | 2017-08-01 | 百度在线网络技术(北京)有限公司 | Method and apparatus for detecting road |
WO2018213739A1 (en) * | 2017-05-18 | 2018-11-22 | TuSimple | System and method for image localization based on semantic segmentation |
US20190130573A1 (en) * | 2017-10-30 | 2019-05-02 | Rakuten, Inc. | Skip architecture neural network machine and method for improved semantic segmentation |
CN107833236A (en) * | 2017-10-31 | 2018-03-23 | 中国科学院电子学研究所 | Semantic vision positioning system and method are combined under a kind of dynamic environment |
US20190043203A1 (en) * | 2018-01-12 | 2019-02-07 | Intel Corporation | Method and system of recurrent semantic segmentation for image processing |
CN108596974A (en) * | 2018-04-04 | 2018-09-28 | 清华大学 | Dynamic scene robot localization builds drawing system and method |
CN109061703A (en) * | 2018-06-11 | 2018-12-21 | 百度在线网络技术(北京)有限公司 | Method, apparatus, equipment and computer readable storage medium used for positioning |
CN109117718A (en) * | 2018-07-02 | 2019-01-01 | 东南大学 | A kind of semantic map structuring of three-dimensional towards road scene and storage method |
CN109186586A (en) * | 2018-08-23 | 2019-01-11 | 北京理工大学 | One kind towards dynamically park environment while position and mixing map constructing method |
CN109272554A (en) * | 2018-09-18 | 2019-01-25 | 北京云迹科技有限公司 | A kind of method and system of the coordinate system positioning for identifying target and semantic map structuring |
CN109461211A (en) * | 2018-11-12 | 2019-03-12 | 南京人工智能高等研究院有限公司 | Semantic vector map constructing method, device and the electronic equipment of view-based access control model point cloud |
CN109357679A (en) * | 2018-11-16 | 2019-02-19 | 济南浪潮高新科技投资发展有限公司 | A kind of indoor orientation method based on significant characteristics identification |
CN109584302A (en) * | 2018-11-27 | 2019-04-05 | 北京旷视科技有限公司 | Camera pose optimization method, device, electronic equipment and computer-readable medium |
CN109637177A (en) * | 2018-12-19 | 2019-04-16 | 斑马网络技术有限公司 | Vehicle positioning method, device, equipment and storage medium |
CN109724603A (en) * | 2019-01-08 | 2019-05-07 | 北京航空航天大学 | A kind of Indoor Robot air navigation aid based on environmental characteristic detection |
CN109920055A (en) * | 2019-03-08 | 2019-06-21 | 视辰信息科技(上海)有限公司 | Construction method, device and the electronic equipment of 3D vision map |
Non-Patent Citations (2)
Title |
---|
刘智杰 等: "基于卷积神经网络的语义同时定位以及地图构建方法", 《科学技术与工程》 * |
宋为刚: "基于街景与航拍图像配准的视觉定位技术", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111758118A (en) * | 2020-05-26 | 2020-10-09 | 蜂图科技有限公司 | Visual positioning method, device and equipment and readable storage medium |
WO2021237443A1 (en) * | 2020-05-26 | 2021-12-02 | 蜂图志科技控股有限公司 | Visual positioning method and apparatus, device and readable storage medium |
JP7446643B2 (en) | 2020-05-26 | 2024-03-11 | マプサス テクノロジー ホールディング リミテッド | Visual positioning methods, devices, equipment and readable storage media |
CN111758118B (en) * | 2020-05-26 | 2024-04-16 | 蜂图志科技控股有限公司 | Visual positioning method, device, equipment and readable storage medium |
CN113744352A (en) * | 2021-09-14 | 2021-12-03 | 北京观海科技发展有限责任公司 | Visual space calibration method, device and storage medium |
CN113744352B (en) * | 2021-09-14 | 2022-07-29 | 北京观海科技发展有限责任公司 | Visual space calibration method, device and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN110298320B (en) | 2021-06-22 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Merkle et al. | Exploring the potential of conditional adversarial networks for optical and SAR image matching | |
EP2491529B1 (en) | Providing a descriptor for at least one feature of an image | |
US20200401617A1 (en) | Visual positioning system | |
CN102567449B (en) | Vision system and method of analyzing an image | |
CN109978755A (en) | Panoramic image synthesis method, device, equipment and storage medium | |
Li et al. | Cross-layer attention network for small object detection in remote sensing imagery | |
Šegvić et al. | A computer vision assisted geoinformation inventory for traffic infrastructure | |
Brejcha et al. | State-of-the-art in visual geo-localization | |
CN109636854A (en) | A kind of augmented reality three-dimensional Tracing Registration method based on LINE-MOD template matching | |
Houshiar et al. | A study of projections for key point based registration of panoramic terrestrial 3D laser scan | |
Tran et al. | On-device scalable image-based localization via prioritized cascade search and fast one-many ransac | |
CN106503248A (en) | Ground drawing generating method and map creation device | |
Vishal et al. | Accurate localization by fusing images and GPS signals | |
CN109596121A (en) | A kind of motor-driven station Automatic Targets and space-location method | |
CN110298320A (en) | A kind of vision positioning method, device and storage medium | |
You et al. | Joint 2-D–3-D traffic sign landmark data set for geo-localization using mobile laser scanning data | |
Gupta et al. | Augmented reality system using lidar point cloud data for displaying dimensional information of objects on mobile phones | |
CN112215964A (en) | Scene navigation method and device based on AR | |
Tschopp et al. | Superquadric object representation for optimization-based semantic SLAM | |
Wang et al. | iNavigation: an image based indoor navigation system | |
CN109034214A (en) | Method and apparatus for generating label | |
CN111758118B (en) | Visual positioning method, device, equipment and readable storage medium | |
Arth et al. | Full 6dof pose estimation from geo-located images | |
Máttyus et al. | Aerial image sequence geolocalization with road traffic as invariant feature | |
Pham et al. | 3D motion matching algorithm using signature feature descriptor |
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 |