CN110060343A

CN110060343A - Map constructing method and system, server, computer-readable medium

Info

Publication number: CN110060343A
Application number: CN201910334825.XA
Authority: CN
Inventors: 蔡育展; 郑超; 高超
Original assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Current assignee: Apollo Intelligent Technology Beijing Co Ltd
Priority date: 2019-04-24
Filing date: 2019-04-24
Publication date: 2019-07-26
Anticipated expiration: 2039-04-24
Also published as: CN110060343B

Abstract

Present disclose provides a kind of map constructing methods, comprising: obtains the crowdsourcing image data of each crowdsourcing source acquisition, the crowdsourcing image data in each crowdsourcing source includes at least two crowdsourcing images；For the crowdsourcing image data of each crowdsourcing source acquisition, local map building is carried out based on the first preset algorithm, obtains the corresponding local map of crowdsourcing image data in the crowdsourcing source；For the crowdsourcing image data of any two crowdsourcing source acquisition, similar frame detection is carried out based on the second preset algorithm, determines the similar crowdsourcing image of at least one set between the crowdsourcing image data in any two crowdsourcing source；For every group of similar crowdsourcing image, the relative pose between the similar crowdsourcing image of the group is calculated；Based on the corresponding local map of crowdsourcing image data of relative pose and each crowdsourcing source between every group of similar crowdsourcing image, map structuring is carried out using third preset algorithm.The disclosure additionally provides map structuring system, server and computer-readable medium.

Description

Map constructing method and system, server, computer-readable medium

Technical field

The embodiment of the present disclosure is related to map structuring technical field, in particular to map constructing method and system, server, meter Calculation machine readable medium.

Background technique

Intelligent driving industry has been built consensus, towards the even more high level autonomous driving vehicle of L3/L4, high-precision map It is important support, and high-precision map need to be just quickly to update, the mode of traditional deployment collecting vehicle is unable to satisfy high-precision The renewal frequency of map.And the construction method of current high-precision map, precision and robustness are lower, and application scenarios also have one Fixed limitation.

Summary of the invention

The embodiment of the present disclosure provides a kind of map constructing method and system, server, computer-readable medium.

In a first aspect, the embodiment of the present disclosure provides a kind of map constructing method, which includes:

The crowdsourcing image data of each crowdsourcing source acquisition is obtained, the crowdsourcing image data in each crowdsourcing source includes at least two A crowdsourcing image；

For the crowdsourcing image data of each crowdsourcing source acquisition, local map building is carried out based on the first preset algorithm, is obtained To the corresponding local map of crowdsourcing image data in the crowdsourcing source；

For the crowdsourcing image data of any two crowdsourcing source acquisition, similar frame detection is carried out based on the second preset algorithm, Determine the similar crowdsourcing image of at least one set between the crowdsourcing image data in any two crowdsourcing source, every group of similar crowd Packet image includes at least two crowdsourcing images；

For every group of similar crowdsourcing image, the relative pose between the similar crowdsourcing image of the group is calculated；

Crowdsourcing image data based on relative pose and each crowdsourcing source between every group of similar crowdsourcing image is corresponding Local map carries out map structuring using third preset algorithm.

In some embodiments, for the crowdsourcing image data of each crowdsourcing source acquisition, in the crowdsourcing image data, arbitrarily The difference of the acquisition time of two adjacent crowdsourcing images is within a predetermined range.

In some embodiments, the crowdsourcing image data for the acquisition of each crowdsourcing source, is based on the first preset algorithm Local map building is carried out, the corresponding local map of crowdsourcing image data for obtaining the crowdsourcing source includes:

For each crowdsourcing source acquisition crowdsourcing image data, arbitrarily selected from the crowdsourcing image data two it is adjacent Crowdsourcing image；

For the crowdsourcing image data of each crowdsourcing source acquisition, according to two adjacent crowdsourcing images of selected taking-up, Construct the corresponding initialization local map of the crowdsourcing image data；

For the crowdsourcing image data of each crowdsourcing source acquisition, locally according to the corresponding initialization of the crowdsourcing image data Figure and remaining crowdsourcing image in addition to two adjacent crowdsourcing image datas, construct the crowdsourcing image data pair in the crowdsourcing source The local map answered.

In some embodiments, it is described for each crowdsourcing source acquisition crowdsourcing image data, according to select this two A adjacent crowdsourcing image, constructing the corresponding initialization local map of the crowdsourcing image data includes:

For the crowdsourcing image data of each crowdsourcing source acquisition, the selected two adjacent crowdsourcing images difference taken out is extracted Corresponding characteristic point；

For the crowdsourcing image data of each crowdsourcing source acquisition, the two adjacent crowdsourcing images selected are respectively corresponded Characteristic point matched, determine the Feature Points Matching pair between the two adjacent crowdsourcing images selected；

For the crowdsourcing image data of each crowdsourcing source acquisition, according between the two adjacent crowdsourcing images selected Feature Points Matching pair calculates the corresponding initialization local map of the crowdsourcing image data.

In some embodiments, the crowdsourcing image data for the acquisition of each crowdsourcing source, according to the crowdsourcing picture number According to corresponding initialization local map and remaining crowdsourcing image in addition to two adjacent crowdsourcing image datas, the crowd is constructed The corresponding local map of crowdsourcing image data of Bao Yuan includes:

For the crowdsourcing image data of each crowdsourcing source acquisition, in addition to two adjacent crowdsourcing image datas The initialization local map is projected on the crowdsourcing image, obtains the crowd by each crowdsourcing image in remaining crowdsourcing image Characteristic matching pair between packet image and initialization local map；

For the crowdsourcing image data of each crowdsourcing source acquisition, in addition to two adjacent crowdsourcing image datas Each crowdsourcing image in remaining crowdsourcing image according to the crowdsourcing image and initializes the characteristic matching pair between local map, Calculate the corresponding relative pose of crowdsourcing image；

For the crowdsourcing image data of each crowdsourcing source acquisition, in addition to two adjacent crowdsourcing image datas Remaining crowdsourcing image, according to initialization local map relative pose corresponding with remaining crowdsourcing image, generating should The corresponding local map of crowdsourcing image data in crowdsourcing source.

In some embodiments, every group of similar crowdsourcing image includes two crowdsourcing images, and two crowdsourcing images are corresponding Crowdsourcing source it is different, the crowdsourcing image data for the acquisition of any two crowdsourcing source carries out phase based on the second preset algorithm It is detected like frame, determines the similar crowdsourcing image packet of at least one set between the crowdsourcing image data in any two crowdsourcing source It includes:

For each crowdsourcing image, the corresponding image vector of crowdsourcing image is obtained；

For any two crowdsourcing image in the different crowdsourcing sources of correspondence, calculates any two crowdsourcing image and respectively correspond The distance between image vector；

For any two crowdsourcing image in the different crowdsourcing sources of correspondence, judge whether the distance is less than or equal to default threshold Value；

For any two crowdsourcing image in the different crowdsourcing sources of correspondence, if judging, the distance is less than or equal to default threshold Value, it is determined that going out any two crowdsourcing image is similar crowdsourcing image.

In some embodiments, the crowdsourcing image data for the acquisition of any two crowdsourcing source, it is default based on second Algorithm carries out similar frame detection, determines the similar crowd of at least one set between the crowdsourcing image data in any two crowdsourcing source Packet image includes:

For each crowdsourcing image, the corresponding image vector of each crowdsourcing image is obtained；

According to the corresponding image vector of each crowdsourcing image and default clustering algorithm, the crowdsourcing in any two crowdsourcing source is determined The similar crowdsourcing image of at least one set between image data.

In some embodiments, every group of similar crowdsourcing image includes two crowdsourcing images, described similar for every group Crowdsourcing image, the relative pose calculated between the similar crowdsourcing image of the group include:

For every group of similar crowdsourcing image, the corresponding feature of each crowdsourcing image in the similar crowdsourcing image of the group is extracted Point；

For every group of similar crowdsourcing image, by the corresponding feature of a crowdsourcing image in the similar crowdsourcing image of the group The corresponding characteristic point of another crowdsourcing image in similar with the group crowdsourcing image of point is matched, and the similar crowd of the group is obtained The Feature Points Matching pair of packet image；

This is calculated according to the Feature Points Matching pair of the similar crowdsourcing image of the group for every group of similar crowdsourcing image Relative pose between the similar crowdsourcing image of group.

In some embodiments, the relative pose based between every group of similar crowdsourcing image and each crowdsourcing source The corresponding local map of crowdsourcing image data, carrying out map structuring using third preset algorithm includes:

With the corresponding office of crowdsourcing image data of relative pose and each crowdsourcing source between every group of similar crowdsourcing image Portion's map is iterated optimization as objective function as optimized variable, to minimize re-projection error, generates point cloud map.

Second aspect, the embodiment of the present disclosure provide a kind of map structuring system, and the map structuring system includes:

Multiple crowdsourcing sources；

The corresponding local map in each crowdsourcing source constructs module, for obtaining the crowdsourcing image data of each crowdsourcing source acquisition, The crowdsourcing image data in each crowdsourcing source includes at least two crowdsourcing images；The crowdsourcing image of the corresponding crowdsourcing source acquisition of needle Data carry out local map building based on the first preset algorithm, and the crowdsourcing image data for obtaining the crowdsourcing source is corresponding locally Figure；

Similar frame detection module, for obtaining the crowdsourcing image data of each crowdsourcing source acquisition；For any two crowdsourcing The crowdsourcing image data of source acquisition carries out similar frame detection based on the second preset algorithm, determines any two crowdsourcing source The similar crowdsourcing image of at least one set between crowdsourcing image data, every group of similar crowdsourcing image include at least two crowdsourcing figures Picture；

Pose computing module calculates between the similar crowdsourcing image of the group for being directed to every group of similar crowdsourcing image Relative pose；

Map generation module, for the crowd based on relative pose and each crowdsourcing source between every group of similar crowdsourcing image The corresponding local map of packet image data carries out map structuring using third preset algorithm.

In some embodiments, the corresponding local map building module in each crowdsourcing source is specifically used for:

For the crowdsourcing image data of corresponding crowdsourcing source acquisition, two phases are arbitrarily selected from the crowdsourcing image data Adjacent crowdsourcing image；

For the crowdsourcing image data of corresponding crowdsourcing source acquisition, according to the two adjacent crowdsourcing images selected, Construct the corresponding initialization local map of the crowdsourcing image data；

For the crowdsourcing image data of corresponding crowdsourcing source acquisition, according to the corresponding initialization part of the crowdsourcing image data Remaining the crowdsourcing image of map in addition to the two crowdsourcing image datas adjacent except this, constructs the crowdsourcing image data in the crowdsourcing source Corresponding local map.

For the crowdsourcing image data of corresponding crowdsourcing source acquisition, the two adjacent crowdsourcing images difference selected is extracted Corresponding characteristic point；

It is for the crowdsourcing image data of corresponding crowdsourcing source acquisition, the two adjacent crowdsourcing images selected are right respectively The characteristic point answered is matched, and determines the Feature Points Matching pair between the two adjacent crowdsourcing images selected；

For the crowdsourcing image data of corresponding crowdsourcing source acquisition, according between the two adjacent crowdsourcing images selected Feature Points Matching pair, calculate the corresponding initialization local map of the crowdsourcing image data.

For the crowdsourcing image data of corresponding crowdsourcing source acquisition, in addition to two adjacent crowdsourcing image datas Remaining crowdsourcing image in each crowdsourcing image, the initialization local map is projected on the crowdsourcing image, is somebody's turn to do Characteristic matching pair between crowdsourcing image and initialization local map；

For the crowdsourcing image data of corresponding crowdsourcing source acquisition, in addition to two adjacent crowdsourcing image datas Remaining crowdsourcing image in each crowdsourcing image, according to the crowdsourcing image and initialization local map between characteristic matching It is right, calculate the corresponding relative pose of crowdsourcing image；

For the crowdsourcing image data of corresponding crowdsourcing source acquisition, in addition to two adjacent crowdsourcing image datas Remaining crowdsourcing image generated according to initialization local map relative pose corresponding with remaining crowdsourcing image The corresponding local map of crowdsourcing image data in the crowdsourcing source.

In some embodiments, every group of similar crowdsourcing image includes two crowdsourcing images, and two crowdsourcing images are corresponding Crowdsourcing source it is different, the similar frame detection module is specifically used for:

In some embodiments, the similar frame detection module is specifically used for:

In some embodiments, every group of similar crowdsourcing image includes two crowdsourcing images, the pose computing module tool Body is used for:

In some embodiments, the map generation module is specifically used for:

The third aspect, the embodiment of the present disclosure provide a kind of server, which includes:

One or more processors；

Storage device is stored thereon with one or more programs；

When one or more of programs are executed by one or more of processors, so that one or more of places It manages device and realizes such as above-mentioned map constructing method.

Fourth aspect, the embodiment of the present disclosure provide a kind of computer-readable medium, are stored thereon with computer program, In, described program is performed realization such as above-mentioned map constructing method.

The map constructing method and system, server, computer-readable medium that the embodiment of the present disclosure provides, using crowdsourcing side Formula obtains the crowdsourcing image data in crowdsourcing source, can satisfy the renewal frequency of map；For the crowdsourcing figure of each crowdsourcing source acquisition As data, it is all made of the first preset algorithm and carries out local map building, so that the application scenarios of map structuring are more extensive；For The crowdsourcing image data in different crowdsourcing sources carries out similar frame detection using the second preset algorithm, to quickly find different crowds Similar frame between the crowdsourcing image data of Bao Yuan realizes the information sharing between the crowdsourcing image data in different crowdsourcing sources；Most Afterwards, the crowdsourcing image data based on relative pose and each crowdsourcing source between every group of similar crowdsourcing image is corresponding locally Figure carries out map structuring, and more efficiently cloud map is more accurately put in building by way of asynchronous refresh.Embodiment of the present disclosure institute The map constructing method of offer, application scenarios are wider, and precision and robustness are preferable.

Detailed description of the invention

Attached drawing is used to provide to further understand the embodiment of the present disclosure, and constitutes part of specification, with this public affairs The embodiment opened is used to explain the disclosure together, does not constitute the limitation to the disclosure.By reference to attached drawing to detailed example reality It applies example to be described, the above and other feature and advantage will become apparent those skilled in the art, in the accompanying drawings:

Fig. 1 is a kind of flow chart for map constructing method that the embodiment of the present disclosure provides；

Fig. 2 is a kind of flow chart of specific embodiment of step 12 in the embodiment of the present disclosure；

Fig. 3 is a kind of flow chart of specific embodiment of step 122 in the embodiment of the present disclosure；

Fig. 4 is a kind of flow chart of specific embodiment of step 123 in the embodiment of the present disclosure；

Fig. 5 is a kind of flow chart of specific embodiment of step 13 in the embodiment of the present disclosure；

Fig. 6 is the flow chart of another specific embodiment of step 13 in the embodiment of the present disclosure；

Fig. 7 is a kind of flow chart of specific embodiment of step 14 in the embodiment of the present disclosure；

Fig. 8 is a kind of structural schematic diagram for map structuring system that the embodiment of the present disclosure provides.

Specific embodiment

To make those skilled in the art more fully understand the technical solution of the disclosure, the disclosure is mentioned with reference to the accompanying drawing Map constructing method and system, server, the computer-readable medium of confession are described in detail.

Example embodiment will hereinafter be described more fully hereinafter with reference to the accompanying drawings, but the example embodiment can be with difference Form embodies and should not be construed as being limited to embodiment set forth herein.Conversely, the purpose for providing these embodiments is It is thoroughly and complete to make the disclosure, and those skilled in the art will be made to fully understand the scope of the present disclosure.

As it is used herein, term "and/or" includes any and all combinations of one or more associated listed entries.

Term as used herein is only used for description specific embodiment, and is not intended to limit the disclosure.As used herein , "one" is also intended to "the" including plural form singular, unless in addition context is expressly noted that.It will also be appreciated that Be, when in this specification use term " includes " and/or " by ... be made " when, specify there are the feature, entirety, step, Operation, element and/or component, but do not preclude the presence or addition of other one or more features, entirety, step, operation, element, Component and/or its group.

Embodiment described herein can be by the idealized schematic diagram of the disclosure and reference planes figure and/or sectional view are retouched It states.It therefore, can be according to manufacturing technology and/or tolerance come modified example diagram.Therefore, embodiment is not limited to reality shown in the drawings Apply example, but the modification of the configuration including being formed based on manufacturing process.Therefore, the area illustrated in attached drawing, which has, schematically to be belonged to Property, and the shape in area as shown in the figure instantiates the concrete shape in the area of element, but is not intended to restrictive.

Unless otherwise defined, the otherwise meaning of all terms (including technical and scientific term) used herein and this field The normally understood meaning of those of ordinary skill is identical.It will also be understood that such as those those of limit term in common dictionary and answer When being interpreted as having and its consistent meaning of meaning under the background of the relevant technologies and the disclosure, and will be not interpreted as having There are idealization or excessively formal meaning, unless clear herein so limit.

Fig. 1 is the flow chart of a kind of map constructing method that the embodiment of the present disclosure provides, as shown in Figure 1, this method can be with It is executed by map structuring system, which can be realized by way of software and/or hardware, which, which can integrate, is taking It is engaged in device.The map constructing method includes:

Step 11, the crowdsourcing image data for obtaining each crowdsourcing source acquisition, the crowdsourcing image data in each crowdsourcing source includes extremely Few two crowdsourcing images.

In the embodiments of the present disclosure, it is divided according to acquisition camera, the corresponding acquisition camera in each crowdsourcing source.

In some embodiments, it can also be corresponded in an area of space according to Spacial domain decomposition, each crowdsourcing source At least one acquisition camera.Wherein, area of space can be determine according to actual needs.

It should be noted that the embodiment of the present disclosure for crowdsourcing image data acquisition modes with no restriction, can pass through The mode of active upload obtains the crowdsourcing image data of the acquisition camera after acquisition camera acquisition crowdsourcing image data, can also lead to Cross the crowdsourcing image data that other modes (such as periodically inquiry acquisition camera) obtain acquisition camera acquisition.Wherein, camera is acquired It can be in-vehicle camera, digital camera, video camera, smart phone camera etc..

It is to be understood that crowdsourcing image refers to by public after certain method (as shot using acquisition camera) is obtained A kind of open image data provided by internet to the public or associated mechanisms.Wherein, masses can voluntarily provide crowdsourcing Image provides crowdsourcing image by way of participating in the crowdsourcing task that associated mechanisms issue.

In some embodiments, in a step 11, for the crowdsourcing image data of each crowdsourcing source acquisition, the crowdsourcing image In data, the difference of the acquisition time of two crowdsourcing images of arbitrary neighborhood is within a predetermined range.For example, preset range is 5 seconds, Preset range can be configured according to actual needs, the embodiment of the present disclosure to this with no restriction.

In some embodiments, in a step 11, the crowdsourcing picture number that each crowdsourcing source acquires within a preset period of time is obtained According to the crowdsourcing image data that each crowdsourcing source acquires within a preset period of time includes at least one crowdsourcing image.For example, when default Between section can be configured according to actual needs, the embodiment of the present disclosure to this with no restriction.

In some embodiments, in a step 11, each crowdsourcing source uploads the crowdsourcing image of its acquisition in real time.

In the embodiments of the present disclosure, in the crowdsourcing image data in each crowdsourcing source crowdsourcing image quantity can it is identical can also be with Difference is determined with specific reference to actual conditions.

Step 12, the crowdsourcing image data acquired for each crowdsourcing source carry out local map based on the first preset algorithm Building, obtains the corresponding local map of crowdsourcing image data in the crowdsourcing source.

In some embodiments, the first preset algorithm includes increment type exercise recovery structure (Structure from Motion, referred to as: SfM) algorithm.In the embodiment of the present disclosure, used algorithm is constructed with no restriction for local map, as long as Corresponding local map can be generated according to crowdsourcing image data.

Fig. 2 is a kind of flow chart of specific embodiment of step 12 in the embodiment of the present disclosure, in some embodiments, such as Shown in Fig. 2, step 12 includes:

Step 121, the crowdsourcing image data acquired for each crowdsourcing source, arbitrarily select from the crowdsourcing image data Two adjacent crowdsourcing images.

For example, the crowdsourcing image data of some crowdsourcing source acquisition includes 4 crowdsourcing images, then from 4 crowdsourcing images, Arbitrarily select 2 adjacent crowdsourcing images, it is described it is " adjacent " can be understood as it is adjacent on acquisition time.

Step 122, the crowdsourcing image data acquired for each crowdsourcing source, according to two adjacent crowds of selected taking-up Packet image constructs the corresponding initialization local map of the crowdsourcing image data.

In some embodiments, for each crowdsourcing source acquisition crowdsourcing image data, selected taking-up it is two adjacent Crowdsourcing image is two earliest images of the shooting time of crowdsourcing source shooting.

Fig. 3 is a kind of flow chart of specific embodiment of step 122 in the embodiment of the present disclosure, in some embodiments, As shown in figure 3, step 122 includes:

Step 1221, the crowdsourcing image data acquired for each crowdsourcing source extract the selected two adjacent crowds taken out The corresponding characteristic point of packet image.

In some embodiments, the extraction of the characteristic point of crowdsourcing image can use SIFT, SURF or ORB method.This Open embodiment for the characteristic point of crowdsourcing image extracting method with no restriction, as long as the feature of crowdsourcing image can be extracted Point.Wherein, characteristic point is map elements, for example, characteristic point can be lane line, bar, guideboard, building etc..

Step 1222, the crowdsourcing image data acquired for each crowdsourcing source, the two adjacent crowdsourcing figures that will be selected As corresponding characteristic point is matched, the Feature Points Matching between the two adjacent crowdsourcing images selected is determined It is right.

For example, can determine the characteristic point between the two adjacent crowdsourcing images selected using RANSAC method Matching pair.

Step 1223, the crowdsourcing image data acquired for each crowdsourcing source, according to the two adjacent crowdsourcings selected Feature Points Matching pair between image calculates the corresponding initialization local map of the crowdsourcing image data.

Specifically, step 1223 includes: the crowdsourcing image data for the acquisition of each crowdsourcing source, according to two selected Feature Points Matching pair between adjacent crowdsourcing image calculates opposite between the two adjacent crowdsourcing images selected Pose；For the crowdsourcing image data of each crowdsourcing source acquisition, according between the two adjacent crowdsourcing images selected Relative pose generates the corresponding initialization local map of the crowdsourcing image data.

It wherein, can be first according to the choosing when calculating the relative pose between the two adjacent crowdsourcing images selected The Feature Points Matching pair between two adjacent crowdsourcing images taken out, solves the two adjacent crowdsourcing images selected Between essential matrix, the opposite position between the two adjacent crowdsourcing images selected is then solved according to essential matrix Appearance.

Wherein, initialization local map can be generated according to the method for trigonometric ratio, the two adjacent crowds selected according to this The relative pose of packet image and corresponding Feature Points Matching pair, can calculate the three-dimensional coordinate of characteristic point.For example, the choosing The two adjacent crowdsourcing images taken out are respectively to scheme A and figure B, and p1 is the characteristic point of figure A, and p2 is the characteristic point of figure B, p1 and p2 It is Feature Points Matching pair, the Feature Points Matching of p1 and p2 can be obtained after trigonometric ratio to corresponding three-dimensional coordinate P, to generate just Beginningization local map.

Step 123, the crowdsourcing image data acquired for each crowdsourcing source, it is corresponding initial according to the crowdsourcing image data Change local map and remaining crowdsourcing image in addition to two adjacent crowdsourcing image datas, constructs the crowdsourcing figure in the crowdsourcing source As the corresponding local map of data.

For example, the crowdsourcing image data of some crowdsourcing source acquisition includes 4 crowdsourcing images, respectively crowdsourcing image 1, crowdsourcing Image 2, crowdsourcing image 3 and crowdsourcing image 4, for the crowdsourcing image data of crowdsourcing source acquisition, in step 122, according to crowd Packet image 1 and crowdsourcing image 2 generate initialization local map, then in step 123, according to crowdsourcing image 3 and crowdsourcing image 4 And initialization local map, construct the corresponding local map of crowdsourcing image data of crowdsourcing source acquisition.

Fig. 4 is a kind of flow chart of specific embodiment of step 123 in the embodiment of the present disclosure, in some embodiments, As shown in figure 4, step 123 includes:

Step 1231, the crowdsourcing image data acquired for each crowdsourcing source, for removing two adjacent crowdsourcing images Initialization local map is projected on the crowdsourcing image, obtains by each crowdsourcing image in remaining crowdsourcing image other than data Characteristic matching pair between the crowdsourcing image and initialization local map.

For example, the crowdsourcing image data of some crowdsourcing source acquisition includes 4 crowdsourcing images, respectively crowdsourcing image 1, crowdsourcing Image 2, crowdsourcing image 3 and crowdsourcing image 4, for the crowdsourcing image data of crowdsourcing source acquisition, in step 122, according to crowd Packet image 1 and crowdsourcing image 2 generate initialization local map.So in step 1231, for the crowd in remaining crowdsourcing image Packet image 3 projects to initialization local map on crowdsourcing image 3, so that it is determined that crowdsourcing image 3 and initialization local map out Between characteristic matching pair；For the crowdsourcing image 4 in remaining crowdsourcing image, initialization local map is projected into crowdsourcing image On 4, so that it is determined that the characteristic matching pair between crowdsourcing image 4 and initialization local map out.

Step 1232, the crowdsourcing image data acquired for each crowdsourcing source, for removing two adjacent crowdsourcing images Each crowdsourcing image in remaining crowdsourcing image other than data, according to the spy between the crowdsourcing image and initialization local map Sign matching pair, calculates the corresponding relative pose of crowdsourcing image.

For example, the crowdsourcing image data of some crowdsourcing source acquisition includes 4 crowdsourcing images, respectively crowdsourcing image 1, crowdsourcing Image 2, crowdsourcing image 3 and crowdsourcing image 4, for the crowdsourcing image data of crowdsourcing source acquisition, in step 122, according to crowd Packet image 1 and crowdsourcing image 2 generate initialization local map, in step 1231, determine crowdsourcing image 3 and initialization part Characteristic matching pair between characteristic matching pair and crowdsourcing image 4 between map and initialization local map；So in step In 1232, according to the characteristic matching pair between crowdsourcing image 3 and initialization local map, it is corresponding to calculate the crowdsourcing image 3 It is corresponding to calculate the crowdsourcing image 4 according to the characteristic matching pair between crowdsourcing image 4 and initialization local map for relative pose Relative pose.

In some embodiments, in step 1232, according to the feature between the crowdsourcing image and initialization local map Matching pair can use PNP algorithm, calculate the corresponding relative pose of crowdsourcing image.

Step 1233, the crowdsourcing image data acquired for each crowdsourcing source, for removing two adjacent crowdsourcing images Remaining crowdsourcing image other than data, according to the initialization local map opposite position corresponding with remaining crowdsourcing image Appearance generates the corresponding local map of crowdsourcing image data in the crowdsourcing source.

Step 13, the crowdsourcing image data acquired for any two crowdsourcing source are carried out similar based on the second preset algorithm Frame detection, determines the similar crowdsourcing image of at least one set between the crowdsourcing image data in any two crowdsourcing source, every group Similar crowdsourcing image includes at least two crowdsourcing images.

In the embodiments of the present disclosure, the first preset algorithm that similar frame detection uses is based on depth convolutional neural networks The algorithm of habit, wherein the method based on deep learning can suitably be able to carry out the detection method of similar frame, this public affairs for all Embodiment is opened for the first preset algorithm with no restriction.

Fig. 5 is a kind of flow chart of specific embodiment of step 13 in the embodiment of the present disclosure, in some embodiments, often The similar crowdsourcing image of group includes two crowdsourcing images, and the corresponding crowdsourcing source of two crowdsourcing images is different, as shown in figure 5, step Rapid 13 include:

Step 1311 is directed to each crowdsourcing image, obtains the corresponding image vector of crowdsourcing image.

In some embodiments, for each crowdsourcing image, using NetVLAD network, it is corresponding to obtain the crowdsourcing image The dimension of image vector, the image vector can be determines according to actual conditions.

Step 1312, for any two crowdsourcing image in the different crowdsourcing sources of correspondence, calculate any two crowdsourcing figure As respectively corresponding the distance between image vector.

Step 1313, for any two crowdsourcing image in the different crowdsourcing sources of correspondence, judge the distance whether be less than or Equal to preset threshold.

Wherein, pre-determined distance can determines according to actual conditions, the embodiment of the present disclosure to this with no restriction.

Step 1314, for any two crowdsourcing image in the different crowdsourcing sources of correspondence, if judge the distance be less than or Equal to preset threshold, it is determined that going out any two crowdsourcing image is similar crowdsourcing image.

For any two crowdsourcing image in the different crowdsourcing sources of correspondence, if judging, the distance is greater than preset threshold, table The bright any two crowdsourcing image is dissimilar crowdsourcing image, therefore without any processing to any two crowdsourcing image.

In some embodiments, every group of similar crowdsourcing image includes multiple crowdsourcing images, such as 3, multiple crowdsourcing Image at least corresponds to two crowdsourcing sources (multiple crowdsourcing images are at least from two crowdsourcing sources), for example, every group of similar crowdsourcing Image includes 3 crowdsourcing images, wherein assuming that one of crowdsourcing image corresponds to crowdsourcing source 1, then remaining two crowdsourcing images Crowdsourcing source 2 can be both corresponded to, crowdsourcing source 2 and crowdsourcing source 3 can also be corresponded respectively to.

In some embodiments, for the crowdsourcing image data of any two crowdsourcing source acquisition, for example, any two crowdsourcing Source is respectively crowdsourcing source A and crowdsourcing source B, it is assumed that the crowdsourcing image data in the A of crowdsourcing source includes crowdsourcing image A1, crowdsourcing source B's Crowdsourcing image data includes crowdsourcing image B1 and crowdsourcing image B2, the crowd of crowdsourcing image data and crowdsourcing source B for crowdsourcing source A Packet image data carries out similar frame detection based on the second preset algorithm, when detecting crowdsourcing image A1 and crowdsourcing image B1 for phase As crowdsourcing image, when crowdsourcing image A1 and crowdsourcing image B2 are also similar crowdsourcing image, then crowdsourcing image A1, crowdsourcing image B1 and crowdsourcing image B2 constitutes one group of similar crowdsourcing image, and in such cases, the similar crowdsourcing image of the group includes three crowds Packet image, and so on, so that it is determined that the similar crowd of at least one set between the crowdsourcing image data in any two crowdsourcing source out Packet image.

Fig. 6 is the flow chart of another specific embodiment of step 13 in the embodiment of the present disclosure, in some embodiments, As shown in fig. 6, step 13 includes:

Step 1321 is directed to each crowdsourcing image, obtains the corresponding image vector of each crowdsourcing image.

Step 1322, according to the corresponding image vector of each crowdsourcing image and default clustering algorithm, determine any two crowd The similar crowdsourcing image of at least one set between the crowdsourcing image data of Bao Yuan.

In step 1322, it presets clustering algorithm and uses K-Means clustering algorithm.Specifically, setting cluster number of clusters K=is many The quantity of Bao Yuan；It is randomly provided K current cluster centres (current mean vector)；Cluster division is initialized as Indicate empty set, i=1,2 ..., K；For the crowdsourcing image data in any two crowdsourcing source, any two crowdsourcing source is calculated The corresponding image vector of each crowdsourcing image in crowdsourcing image data is at a distance from each current cluster centre；It is any for this The corresponding image vector of each crowdsourcing image in the crowdsourcing image data in two crowdsourcing sources, determination are corresponding apart from crowdsourcing image The nearest current cluster centre of image vector；For each crowdsourcing image in the crowdsourcing image data in any two crowdsourcing source The corresponding image vector of the crowdsourcing image is divided into the nearest corresponding cluster of current cluster centre of distance by corresponding image vector, It updates cluster and divides C={ Ci }；For each current cluster Ci, its corresponding cluster centre is updated；Using new cluster centre as working as Preceding cluster centre to the corresponding image vector of each crowdsourcing image in the crowdsourcing image data in any two crowdsourcing source again into Row cluster, and so on, until exporting final cluster when new cluster centre is no longer changed and dividing C={ Ci }.Wherein, Each cluster is one group of similar crowdsourcing image.In the embodiments of the present disclosure, it in order to improve the efficiency that similar frame detects, needs to move State adjusts each current cluster centre.

It in the embodiment of the present disclosure, is detected for the similar frame between the crowdsourcing image data in different crowdsourcing sources, is based on depth The algorithm of convolutional neural networks study, directly using the corresponding image vector of crowdsourcing image as input；And traditional method, first It needs to extract image local feature, is then inputted according to the form that visual dictionary is organized into vector.The embodiment of the present disclosure is used The algorithm of depth convolutional neural networks study can excavate deeper characteristics of image, accuracy and robustness are better than passing The method of the view-based access control model dictionary of system.

It should be noted that in the embodiments of the present disclosure, the quantity that difference organizes crowdsourcing image in similar crowdsourcing image can It can also be different with identical, for example, having two groups between the crowdsourcing image data in crowdsourcing source 1 and the crowdsourcing image data in crowdsourcing source 2 Similar crowdsourcing image, wherein one group of similar crowdsourcing image includes 2 crowdsourcing images, another group of similar crowdsourcing image includes 3 crowdsourcing images.In the embodiment of the present disclosure, the quantity of crowdsourcing image specifically can basis in the similar crowdsourcing image of difference group Depending on the detection of similar frame.

It in the embodiments of the present disclosure, can be using the method for above-mentioned steps 1311 to step 1314 by every group of similar crowdsourcing The quantity of crowdsourcing image is fixed on two in image, i.e., only compares two crowdsourcing images, one group is classified as if similar.It can also be with It is determined between the crowdsourcing image data in any two crowdsourcing source at least using the method for above-mentioned steps 1321 to step 1322 One group of similar crowdsourcing image.

Step 14 is directed to every group of similar crowdsourcing image, calculates the relative pose between the similar crowdsourcing image of the group.

Fig. 7 is a kind of flow chart of specific embodiment of step 14 in the embodiment of the present disclosure, in some embodiments, often The similar crowdsourcing image of group includes two crowdsourcing images, as shown in fig. 7, step 14 includes:

Step 1411 is directed to every group of similar crowdsourcing image, extracts each crowdsourcing image in the similar crowdsourcing image of the group Corresponding characteristic point.

Step 1412 is directed to every group of similar crowdsourcing image, by a crowdsourcing image in the similar crowdsourcing image of the group The corresponding characteristic point of another crowdsourcing image in corresponding characteristic point crowdsourcing image similar with the group is matched, and is somebody's turn to do The Feature Points Matching pair of the similar crowdsourcing image of group.

For example, can determine a crowdsourcing image pair in the similar crowdsourcing image of the group using the method such as RANSAC The Feature Points Matching pair between another crowdsourcing image in the characteristic point answered and the similar crowdsourcing image of the group.

Step 1413 is directed to every group of similar crowdsourcing image, according to the Feature Points Matching pair of the similar crowdsourcing image of the group, Calculate the relative pose between the similar crowdsourcing image of the group.

In some embodiments, step 14 includes:

Step 1421 is directed to every group of similar crowdsourcing image, extracts each crowdsourcing image in the similar crowdsourcing image of the group Corresponding characteristic point.

Step 1422, for any two crowdsourcing image in every group of similar crowdsourcing image, by one of crowdsourcing figure As carrying out Feature Points Matching with another crowdsourcing image, the Feature Points Matching pair between any two crowdsourcing image is obtained.

Step 1423, for any two crowdsourcing image in every group of similar crowdsourcing image, according to any two crowd Feature Points Matching pair between packet image calculates the relative pose between any two crowdsourcing image.

Step 1424 is directed to every group of similar crowdsourcing image, raw according to the relative pose between any two crowdsourcing image At the relative pose between the similar crowdsourcing image of the group.

Step 15, the crowdsourcing image data based on relative pose and each crowdsourcing source between every group of similar crowdsourcing image Corresponding local map carries out map structuring using third preset algorithm.

In the embodiments of the present disclosure, the crowd based on relative pose and each crowdsourcing source between every group of similar crowdsourcing image The corresponding local map of packet image data carries out map structuring using third preset algorithm, by the crowdsourcing image data in crowdsourcing source Corresponding local map is merged, and point cloud map is generated.

In some embodiments, such as distributed light-stream adjustment (Bundle Adjustment, referred to as: BA), according to every The corresponding local map of crowdsourcing image data of relative pose and each crowdsourcing source between the similar crowdsourcing image of group, generates complete The point cloud map of office's consistency.

In some embodiments, third preset algorithm can be ADMM algorithm, or DGS algorithm.

In some embodiments, step 15 include: between every group of similar crowdsourcing image relative pose and each crowd The corresponding local map of crowdsourcing image data of Bao Yuan is carried out using minimizing re-projection error as objective function as optimized variable Iteration optimization generates point cloud map.

In step 15, similar with every group to minimize re-projection error as objective function using ADMM Optimization Framework The corresponding local map of crowdsourcing image data of relative pose and each crowdsourcing source between crowdsourcing image is constraint condition, is carried out Distributed optimization.So-called distributed optimization refers to, is unknown, only authorities in different moments changed local map When portion's map changes, distributed optimization process can be just triggered.

In some embodiments, for each crowdsourcing source, when the crowdsourcing image data in the crowdsourcing source updates (the crowdsourcing source Shooting uploads new crowdsourcing image, and new crowdsourcing image is added in crowdsourcing image data) when, it can be by the crowdsourcing figure in the crowdsourcing source As the corresponding current local map of data projects on new crowdsourcing image, to find new characteristic matching pair, generate new Relative pose, so that the corresponding local map in crowdsourcing source is updated, final updated point cloud map.

Map constructing method provided by the embodiment of the present disclosure obtains the crowdsourcing picture number in crowdsourcing source using crowdsourcing mode According to can satisfy the renewal frequency of map；For the crowdsourcing image data of each crowdsourcing source acquisition, it is all made of the first preset algorithm Local map building is carried out, so that the application scenarios of map structuring are more extensive；For the crowdsourcing picture number in different crowdsourcing sources According to using the similar frame detection of the second preset algorithm progress, to quickly find between the crowdsourcing image data in different crowdsourcing sources Similar frame realizes the information sharing between the crowdsourcing image data in different crowdsourcing sources；Finally, being based on every group of similar crowdsourcing image Between relative pose and the corresponding local map of crowdsourcing image data in each crowdsourcing source carry out map structuring, by it is asynchronous more New mode more efficiently constructs more accurately point cloud map.Map constructing method provided by the embodiment of the present disclosure, applied field Scape is wider, and precision and robustness are preferable.

Fig. 8 is a kind of structural schematic diagram for map structuring system that the embodiment of the present disclosure provides, as shown in figure 8, the map Building system includes: for realizing above-mentioned map constructing method, the map structuring system

Multiple crowdsourcing sources 21, for acquiring crowdsourcing image data.

The corresponding local map in each crowdsourcing source 21 constructs module 22, for obtaining the crowdsourcing picture number of each crowdsourcing source acquisition According to the crowdsourcing image data in each crowdsourcing source 21 includes at least two crowdsourcing images；The crowdsourcing figure of the corresponding crowdsourcing source acquisition of needle As data, local map building is carried out based on the first preset algorithm, obtains the corresponding office of crowdsourcing image data in the crowdsourcing source 21 Portion's map.

Similar frame detection module 23, for obtaining the crowdsourcing image data of each crowdsourcing source acquisition；For any two crowd The crowdsourcing image data of packet source acquisition carries out similar frame detection based on the second preset algorithm, determines any two crowdsourcing source Crowdsourcing image data between the similar crowdsourcing image of at least one set, every group of similar crowdsourcing image includes at least two crowdsourcings Image.

Pose computing module 24, for be directed to every group of similar crowdsourcing image, calculate the similar crowdsourcing image of the group it Between relative pose.

Map generation module 25, for based between every group of similar crowdsourcing image relative pose and each crowdsourcing source The corresponding local map of crowdsourcing image data carries out map structuring using third preset algorithm.

In some embodiments, for the crowdsourcing image data of each crowdsourcing source 21 acquisition, in the crowdsourcing image data, appoint Anticipate two adjacent crowdsourcing images acquisition time difference within a predetermined range.

In some embodiments, the corresponding local map in each crowdsourcing source 21 building module 22 is specifically used for: for correspondence Crowdsourcing source acquisition crowdsourcing image data, two adjacent crowdsourcing images are arbitrarily selected from the crowdsourcing image data；Needle The crowdsourcing is constructed according to the two adjacent crowdsourcing images selected to the crowdsourcing image data of corresponding crowdsourcing source acquisition The corresponding initialization local map of image data；For the crowdsourcing image data of corresponding crowdsourcing source acquisition, according to the crowdsourcing figure As remaining the crowdsourcing image of the corresponding initialization local map of data in addition to the two crowdsourcing image datas adjacent except this, building The corresponding local map of crowdsourcing image data in the crowdsourcing source.

In some embodiments, the corresponding local map in each crowdsourcing source 21 building module 22 is specifically used for: for correspondence Crowdsourcing source acquisition crowdsourcing image data, extract the corresponding characteristic point of two adjacent crowdsourcing images selected；Needle To the crowdsourcing image data of corresponding crowdsourcing source acquisition, the corresponding characteristic point of two adjacent crowdsourcing images that will be selected It is matched, determines the Feature Points Matching pair between the two adjacent crowdsourcing images selected；For corresponding crowdsourcing The crowdsourcing image data of source acquisition is calculated according to the Feature Points Matching pair between the two adjacent crowdsourcing images selected The corresponding initialization local map of the crowdsourcing image data.

In some embodiments, the corresponding local map in each crowdsourcing source 21 building module 22 is specifically used for: for correspondence Crowdsourcing source acquisition crowdsourcing image data, in remaining crowdsourcing image in addition to two adjacent crowdsourcing image datas Each crowdsourcing image, the initialization local map is projected on the crowdsourcing image, obtain the crowdsourcing image and initialization Characteristic matching pair between local map；For the crowdsourcing image data of corresponding crowdsourcing source acquisition, for except this adjacent two Each crowdsourcing image in remaining crowdsourcing image other than a crowdsourcing image data according to the crowdsourcing image and initializes locally Characteristic matching pair between figure calculates the corresponding relative pose of crowdsourcing image；For the crowdsourcing of corresponding crowdsourcing source acquisition Image data, for remaining crowdsourcing image in addition to two adjacent crowdsourcing image datas, according to the initialization part Map relative pose corresponding with remaining crowdsourcing image, the crowdsourcing image data for generating the crowdsourcing source are corresponding locally Figure.

In some embodiments, every group of similar crowdsourcing image includes two crowdsourcing images, and two crowdsourcing images are corresponding Crowdsourcing source it is different, the similar frame detection module 23 is specifically used for: for each crowdsourcing image, it is corresponding to obtain the crowdsourcing image Image vector；For any two crowdsourcing image in the different crowdsourcing sources of correspondence, any two crowdsourcing image difference is calculated The distance between correspondence image vector；For any two crowdsourcing image in the different crowdsourcing sources of correspondence, whether the distance is judged Less than or equal to preset threshold；For any two crowdsourcing image in the different crowdsourcing sources of correspondence, if judging, the distance is less than Or it is equal to preset threshold, it is determined that going out any two crowdsourcing image is similar crowdsourcing image.

In some embodiments, the similar frame detection module 23 is specifically used for: being directed to each crowdsourcing image, obtains each The corresponding image vector of crowdsourcing image；According to the corresponding image vector of each crowdsourcing image and default clustering algorithm, determine any The similar crowdsourcing image of at least one set between the crowdsourcing image data in two crowdsourcing sources.

In some embodiments, every group of similar crowdsourcing image includes two crowdsourcing images, the pose computing module 24 It is specifically used for: for every group of similar crowdsourcing image, extracts the corresponding spy of each crowdsourcing image in the similar crowdsourcing image of the group Sign point；For every group of similar crowdsourcing image, by the corresponding characteristic point of a crowdsourcing image in the similar crowdsourcing image of the group The corresponding characteristic point of another crowdsourcing image in crowdsourcing image similar with the group is matched, and the similar crowdsourcing of the group is obtained The Feature Points Matching pair of image；For every group of similar crowdsourcing image, according to the Feature Points Matching of the similar crowdsourcing image of the group It is right, calculate the relative pose between the similar crowdsourcing image of the group.

In some embodiments, the map generation module 25 is specifically used for: between every group of similar crowdsourcing image The corresponding local map of crowdsourcing image data in relative pose and each crowdsourcing source is as optimized variable, to minimize re-projection mistake Difference is that objective function is iterated optimization, generates point cloud map.

In addition, map structuring system provided by the embodiment of the present disclosure is specifically used for realizing aforementioned map construction method, tool Body can be found in the description of aforementioned map construction method, and details are not described herein again.

The embodiment of the present disclosure additionally provides a kind of server, which includes: one or more processors and storage Device；Wherein, one or more programs are stored on storage device, when said one or multiple programs are by said one or multiple When processor executes, so that said one or multiple processors realize map constructing method above-mentioned.

The embodiment of the present disclosure additionally provides a computer readable storage medium, is stored thereon with computer program, wherein should Computer program, which is performed, realizes map constructing method above-mentioned.

It will appreciated by the skilled person that whole or certain steps, system, dress in method disclosed hereinabove Functional module/unit in setting may be implemented as software, firmware, hardware and its combination appropriate.In hardware embodiment, Division between the functional module/unit referred in the above description not necessarily corresponds to the division of physical assemblies；For example, one Physical assemblies can have multiple functions or a function or step and can be executed by several physical assemblies cooperations.Certain objects Reason component or all physical assemblies may be implemented as by processor, such as central processing unit, digital signal processor or micro process The software that device executes, is perhaps implemented as hardware or is implemented as integrated circuit, such as specific integrated circuit.Such software Can be distributed on a computer-readable medium, computer-readable medium may include computer storage medium (or non-transitory be situated between Matter) and communication media (or fugitive medium).As known to a person of ordinary skill in the art, term computer storage medium includes In any method or skill for storing information (such as computer readable instructions, data structure, program module or other data) The volatile and non-volatile implemented in art, removable and nonremovable medium.Computer storage medium includes but is not limited to RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storages, magnetic Box, tape, disk storage or other magnetic memory apparatus or it can be used for storing desired information and can be visited by computer Any other medium asked.In addition, known to a person of ordinary skill in the art be, communication media generally comprises computer-readable Other numbers in the modulated data signal of instruction, data structure, program module or such as carrier wave or other transmission mechanisms etc According to, and may include any information delivery media.

Example embodiment has been disclosed herein, although and use concrete term, they are only used for simultaneously only should It is interpreted general remark meaning, and is not used in the purpose of limitation.In some instances, aobvious to those skilled in the art and Be clear to, unless otherwise expressly stated, the feature that description is combined with specific embodiment that otherwise can be used alone, characteristic and/ Or element, or the feature, characteristic and/or element of description can be combined with other embodiments and be applied in combination.Therefore, art technology Personnel will be understood that, in the case where not departing from the scope of the present disclosure illustrated by the attached claims, can carry out various forms With the change in details.

Claims

1. a kind of map constructing method, comprising:

The crowdsourcing image data of each crowdsourcing source acquisition is obtained, the crowdsourcing image data in each crowdsourcing source includes at least two many Packet image；

For the crowdsourcing image data of each crowdsourcing source acquisition, local map building is carried out based on the first preset algorithm, is somebody's turn to do The corresponding local map of crowdsourcing image data in crowdsourcing source；

For the crowdsourcing image data of any two crowdsourcing source acquisition, similar frame detection is carried out based on the second preset algorithm, is determined The similar crowdsourcing image of at least one set between the crowdsourcing image data in any two crowdsourcing source out, every group of similar crowdsourcing figure As including at least two crowdsourcing images；

The corresponding part of crowdsourcing image data based on relative pose and each crowdsourcing source between every group of similar crowdsourcing image Map carries out map structuring using third preset algorithm.

2. map constructing method according to claim 1, wherein for each crowdsourcing source acquisition crowdsourcing image data, In the crowdsourcing image data, the difference of the acquisition time of two crowdsourcing images of arbitrary neighborhood is within a predetermined range.

3. map constructing method according to claim 1, wherein the crowdsourcing picture number for the acquisition of each crowdsourcing source According to, based on the first preset algorithm carry out local map building, obtain the corresponding local map of crowdsourcing image data in the crowdsourcing source Include:

For the crowdsourcing image data of each crowdsourcing source acquisition, two adjacent crowds are arbitrarily selected from the crowdsourcing image data Packet image；

For the crowdsourcing image data of each crowdsourcing source acquisition, according to two adjacent crowdsourcing images of selected taking-up, building The corresponding initialization local map of the crowdsourcing image data；

For each crowdsourcing source acquisition crowdsourcing image data, according to the corresponding initialization local map of the crowdsourcing image data with Remaining crowdsourcing image in addition to two adjacent crowdsourcing image datas, the crowdsourcing image data for constructing the crowdsourcing source are corresponding Local map.

4. map constructing method according to claim 3, wherein the crowdsourcing picture number for the acquisition of each crowdsourcing source According to constructing the corresponding initialization local map packet of the crowdsourcing image data according to the two adjacent crowdsourcing images selected It includes:

For the crowdsourcing image data of each crowdsourcing source acquisition, extracts the selected two adjacent crowdsourcing images taken out and respectively correspond Characteristic point；

For the crowdsourcing image data of each crowdsourcing source acquisition, the corresponding spy of two adjacent crowdsourcing images that will be selected Sign point is matched, and determines the Feature Points Matching pair between the two adjacent crowdsourcing images selected；

For the crowdsourcing image data of each crowdsourcing source acquisition, according to the feature between the two adjacent crowdsourcing images selected Point matching pair, calculates the corresponding initialization local map of the crowdsourcing image data.

5. map constructing method according to claim 4, wherein the crowdsourcing picture number for the acquisition of each crowdsourcing source According to according to the corresponding initialization local map of the crowdsourcing image data and its in addition to two adjacent crowdsourcing image datas Remaining crowdsourcing image, the corresponding local map of crowdsourcing image data for constructing the crowdsourcing source include:

For the crowdsourcing image data of each crowdsourcing source acquisition, for remaining in addition to two adjacent crowdsourcing image datas The initialization local map is projected on the crowdsourcing image, obtains the crowdsourcing figure by each crowdsourcing image in crowdsourcing image As the characteristic matching pair between initialization local map；

For the crowdsourcing image data of each crowdsourcing source acquisition, for remaining in addition to two adjacent crowdsourcing image datas Each crowdsourcing image in crowdsourcing image is calculated according to the characteristic matching pair between the crowdsourcing image and initialization local map The corresponding relative pose of crowdsourcing image out；

For the crowdsourcing image data of each crowdsourcing source acquisition, for remaining in addition to two adjacent crowdsourcing image datas Crowdsourcing image generates the crowdsourcing according to initialization local map relative pose corresponding with remaining crowdsourcing image The corresponding local map of crowdsourcing image data in source.

6. map constructing method according to claim 1, wherein every group of similar crowdsourcing image includes two crowdsourcing figures Picture, the corresponding crowdsourcing source of two crowdsourcing images is different, the crowdsourcing image data for the acquisition of any two crowdsourcing source, base Similar frame detection is carried out in the second preset algorithm, determines at least one between the crowdsourcing image data in any two crowdsourcing source Organizing similar crowdsourcing image includes:

For any two crowdsourcing image in the different crowdsourcing sources of correspondence, calculates any two crowdsourcing image and respectively correspond image The distance between vector；

For any two crowdsourcing image in the different crowdsourcing sources of correspondence, judge whether the distance is less than or equal to preset threshold；

For any two crowdsourcing image in the different crowdsourcing sources of correspondence, if judging, the distance is less than or equal to preset threshold, Then determine that any two crowdsourcing image is similar crowdsourcing image.

7. map constructing method according to claim 1, wherein the crowdsourcing figure for the acquisition of any two crowdsourcing source As data, similar frame detection is carried out based on the second preset algorithm, determine any two crowdsourcing source crowdsourcing image data it Between the similar crowdsourcing image of at least one set include:

According to the corresponding image vector of each crowdsourcing image and default clustering algorithm, the crowdsourcing image in any two crowdsourcing source is determined The similar crowdsourcing image of at least one set between data.

8. map constructing method according to claim 1, wherein every group of similar crowdsourcing image includes two crowdsourcing figures Picture, described to be directed to every group of similar crowdsourcing image, the relative pose calculated between the similar crowdsourcing image of the group includes:

For every group of similar crowdsourcing image, the corresponding characteristic point of each crowdsourcing image in the similar crowdsourcing image of the group is extracted；

For every group of similar crowdsourcing image, by the similar crowdsourcing image of the group the corresponding characteristic point of a crowdsourcing image with Another corresponding characteristic point of crowdsourcing image in the similar crowdsourcing image of the group is matched, and the similar crowdsourcing figure of the group is obtained The Feature Points Matching pair of picture；

This group of phase is calculated according to the Feature Points Matching pair of the similar crowdsourcing image of the group for every group of similar crowdsourcing image As relative pose between crowdsourcing image.

9. map constructing method according to claim 1, wherein the phase based between every group of similar crowdsourcing image To the corresponding local map of crowdsourcing image data of pose and each crowdsourcing source, map structuring packet is carried out using third preset algorithm It includes:

It is corresponding locally with the crowdsourcing image data of relative pose and each crowdsourcing source between every group of similar crowdsourcing image Figure is used as optimized variable, is iterated optimization as objective function to minimize re-projection error, generates point cloud map.

10. a kind of map structuring system, comprising:

Multiple crowdsourcing sources；

The corresponding local map in each crowdsourcing source constructs module, for obtaining the crowdsourcing image data of each crowdsourcing source acquisition, each The crowdsourcing image data in crowdsourcing source includes at least two crowdsourcing images；The crowdsourcing picture number of the corresponding crowdsourcing source acquisition of needle According to, based on the first preset algorithm carry out local map building, obtain the corresponding local map of crowdsourcing image data in the crowdsourcing source；

Similar frame detection module, for obtaining the crowdsourcing image data of each crowdsourcing source acquisition；It is adopted for any two crowdsourcing source The crowdsourcing image data of collection carries out similar frame detection based on the second preset algorithm, determines the crowdsourcing in any two crowdsourcing source The similar crowdsourcing image of at least one set between image data, every group of similar crowdsourcing image include at least two crowdsourcing images；

Pose computing module calculates the phase between the similar crowdsourcing image of the group for being directed to every group of similar crowdsourcing image To pose；

Map generation module, for the crowdsourcing figure based on relative pose and each crowdsourcing source between every group of similar crowdsourcing image As the corresponding local map of data, map structuring is carried out using third preset algorithm.

11. map structuring system according to claim 10, wherein for the crowdsourcing picture number of each crowdsourcing source acquisition According in the crowdsourcing image data, the difference of the acquisition time of two crowdsourcing images of arbitrary neighborhood is within a predetermined range.

12. map structuring system according to claim 10, wherein the corresponding local map in each crowdsourcing source constructs module It is specifically used for:

For the crowdsourcing image data of corresponding crowdsourcing source acquisition, arbitrarily selected from the crowdsourcing image data two adjacent Crowdsourcing image；

For the crowdsourcing image data of corresponding crowdsourcing source acquisition, according to the two adjacent crowdsourcing images selected, building The corresponding initialization local map of the crowdsourcing image data；

For the crowdsourcing image data of corresponding crowdsourcing source acquisition, according to the corresponding initialization local map of the crowdsourcing image data With remaining crowdsourcing image in addition to two crowdsourcing image datas adjacent except this, the crowdsourcing image data for constructing the crowdsourcing source is corresponding Local map.

13. map structuring system according to claim 12, wherein the corresponding local map in each crowdsourcing source constructs module It is specifically used for:

For the crowdsourcing image data of corresponding crowdsourcing source acquisition, extracts the two adjacent crowdsourcing images selected and respectively correspond Characteristic point；

It is for the crowdsourcing image data of corresponding crowdsourcing source acquisition, the two adjacent crowdsourcing images selected are corresponding Characteristic point is matched, and determines the Feature Points Matching pair between the two adjacent crowdsourcing images selected；

For the crowdsourcing image data of corresponding crowdsourcing source acquisition, according to the spy between the two adjacent crowdsourcing images selected Sign point matching pair, calculates the corresponding initialization local map of the crowdsourcing image data.

14. map structuring system according to claim 13, wherein the corresponding local map in each crowdsourcing source constructs module It is specifically used for:

For the crowdsourcing image data of corresponding crowdsourcing source acquisition, for its in addition to two adjacent crowdsourcing image datas The initialization local map is projected on the crowdsourcing image, obtains the crowdsourcing by each crowdsourcing image in remaining crowdsourcing image Characteristic matching pair between image and initialization local map；

For the crowdsourcing image data of corresponding crowdsourcing source acquisition, for its in addition to two adjacent crowdsourcing image datas Each crowdsourcing image in remaining crowdsourcing image, according to the characteristic matching pair between the crowdsourcing image and initialization local map, meter Calculate the corresponding relative pose of crowdsourcing image；

For the crowdsourcing image data of corresponding crowdsourcing source acquisition, for its in addition to two adjacent crowdsourcing image datas Remaining crowdsourcing image generates the crowd according to initialization local map relative pose corresponding with remaining crowdsourcing image The corresponding local map of crowdsourcing image data of Bao Yuan.

15. map structuring system according to claim 10, wherein every group of similar crowdsourcing image includes two crowdsourcing figures Picture, the corresponding crowdsourcing source of two crowdsourcing images is different, and the similar frame detection module is specifically used for:

16. map structuring system according to claim 10, wherein the similar frame detection module is specifically used for:

17. map structuring system according to claim 10, wherein every group of similar crowdsourcing image includes two crowdsourcing figures Picture, the pose computing module are specifically used for:

18. map structuring system according to claim 10, wherein the map generation module is specifically used for:

19. a kind of server, comprising:

One or more processors；

Storage device is stored thereon with 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 the map constructing method as described in any in claim 1-9.

20. a kind of computer-readable medium, is stored thereon with computer program, wherein described program is performed realization as weighed Benefit requires any map constructing method in 1-9.