CN113392370A - Efficient long-time large-scale SLAM method and system - Google Patents
Efficient long-time large-scale SLAM method and system Download PDFInfo
- Publication number
- CN113392370A CN113392370A CN202110658243.4A CN202110658243A CN113392370A CN 113392370 A CN113392370 A CN 113392370A CN 202110658243 A CN202110658243 A CN 202110658243A CN 113392370 A CN113392370 A CN 113392370A
- Authority
- CN
- China
- Prior art keywords
- information
- key frame
- level
- storage
- level information
- 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
- 238000000034 method Methods 0.000 title claims abstract description 16
- 238000004364 calculation method Methods 0.000 claims description 27
- 239000011159 matrix material Substances 0.000 claims description 16
- 238000000605 extraction Methods 0.000 claims description 14
- 230000007774 longterm Effects 0.000 claims description 7
- 238000001514 detection method Methods 0.000 abstract description 5
- 238000010276 construction Methods 0.000 abstract description 3
- 230000001360 synchronised effect Effects 0.000 abstract description 3
- 239000013598 vector Substances 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000014509 gene expression Effects 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/60—Memory management
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Devices For Executing Special Programs (AREA)
- Instructional Devices (AREA)
Abstract
The invention discloses a high-efficiency long-time large-scale SLAM method and a system; belongs to the field of synchronous positioning and map construction of a movable platform; the technical key points are as follows: abstracting and expressing images received by machine vision into different levels of information; the higher the information single memory space of the higher layer is, the higher the information single memory space of the lower layer is; the higher the information number of the layer, the more; the lower layers have fewer information numbers. By adopting the high-efficiency long-time large-scale SLAM method and the system, the performance and the high efficiency of loop detection under the long-time large-scale SLAM are achieved.
Description
Technical Field
The invention relates to the technical field of automobile synchronous positioning and map construction, in particular to an efficient long-time large-scale SLAM method and system.
Background
SLAM, an acronym for Simultaneous localization and Mapping, which is interpreted as: synchronous positioning and map construction; the technology is a core technology (which belongs to support software in the field of intelligent transportation) for realizing autonomous navigation positioning and building an environment map for an automobile.
The existing Visual SLAM algorithm faces the main problem that a large amount of historical information needs to be stored to carry out Loop Detection (Loop Closure Detection) in a larger scale and a longer time range to reduce accumulated deviation when facing Long-term (Long-term) and large-scale (Large-scale) scenes.
For example, CN111767854A proposes a SLAM loop detection method combining scene text semantic information, which has the core steps of: for any key frame in the key frame set, calculating the cosine similarity of the total information vector and the current frame total information vector, and taking the key frame of which the similarity is greater than a certain threshold and which is not directly adjacent to the current frame as a loop candidate frame. As this approach, there is the aforementioned problem of storing information vectors for each key frame. This is very stressful for hardware storage devices.
Disclosure of Invention
The invention aims to provide an efficient long-time large-scale SLAM method and system, which overcome the defects in the prior art.
An efficient long-term large-scale SLAM system, comprising: the system comprises a current key frame information extraction module, a current key frame information storage importance coefficient calculation module and a key frame set level information storage importance coefficient calculation and storage module; a key frame set level information calculating and storing module;
the current key frame information storage importance coefficient calculation module is connected with the key frame set level information storage importance coefficient calculation and storage module;
the current key frame information extraction module and the key frame set level information storage importance coefficient calculation and storage module are respectively connected with the key frame set level information calculation and storage module;
the current key frame information extraction module is used for extracting information of different levels from a current key frame image;
the current key frame information storage importance coefficient calculation module is used for calculating storage importance coefficients of different levels of information of the current key frame;
the key frame set level information storage importance coefficient calculation and storage module is used for storing the storage importance coefficients of different level information of the current key frame set;
the key frame set level information calculation and storage module is used for determining a storage mode for extracting information of different levels from the current key frame image transmitted by the current key frame information extraction module.
Further, the SLAM system further includes: the system comprises an information level sequence establishing module and an information level memory space establishing module;
the information level sequence establishing module and the information level memory space establishing module are respectively connected with the current key frame information extracting module; the information level sequence establishing module and the information level storage capacity establishing module are respectively connected with the current key frame information storage importance coefficient calculating module.
Further, the information hierarchy sequence establishing module is configured to establish an information hierarchy sequence of a key frame set, and is configured to instruct the current key frame information extracting module to extract information of a current key frame image, and is a carrier operated in step S100; the information level storage quantity establishing module is used for establishing different level information storage spaces of the key frame set, and is a carrier operated in step S200.
An efficient long-time large-scale SLAM method is characterized in that:
abstracting and expressing images received by machine vision into different levels of information;
the higher the information individual storage amount, the higher the information individual storage amount of the lower layer (i.e., X)j,βThe storage space occupied by a single is less than Xj+1,βA single occupied storage space);
the higher the information number of the layer, the more; the lower the layer, the less the information number (i.e., Q described below)j>Qj+1)。
An efficient long-time large-scale SLAM method comprises the following steps:
s100, establishing an information hierarchy sequence of the key frames:
the information hierarchy sequence of the key frame is expressed as follows according to the descending order of the hierarchy level by level: level 1 information, level 2 information, level j information … … …, level k information:
s200, determining the storage capacity of each information level of the key frame:
the storage amount of the 1 st level information is Q1The storage amount of the 2 nd level information is Q2… … … … level jThe storage amount of information is Qj… … … … the k-th level information has a storage capacity of Qk:
For Q1~QkFor example, there are: q1>Q2>……Qj-1>Qj>……Qk;
S300, the current key frame is the Yth key frame, and the information of layers 1 to k is extracted from the Yth key frame: x1,Y、X2,Y、……Xj,Y、……Xk,Y(ii) a And simultaneously calculating the storage importance coefficients of the information of the 1 st to k th layers extracted from the Yth key frame: z1,Y、Z2,Y、……Zj,Y、……Zk,Y;
X1,Y、X2,Y、……Xj,Y、……Xk,YRespectively represent: level 1 information of the Y-th key frame, level 2 information of the Y-th key frame, … … level j information of the Y-th key frame, … … level k information of the Y-th key frame;
Z1,Y、Z2,Y、……Zj,Y、……Zk,Yrespectively represent: a storage importance coefficient of level 1 information of a Y-th key frame, a storage importance coefficient of level 2 information of the Y-th key frame, a storage importance coefficient of a j-th level information of … … the Y-th key frame, a storage importance coefficient of a k-th level information of … … the Y-th key frame;
s400, the storage importance coefficients of the information of the 1 st to k th layers extracted from the current Y-th key frame are: z1,Y、Z2,Y、……Zj,Y、……Zk,YAdding the importance parameter into a storage importance coefficient matrix of the current key frame level information (all the storage importance parameters of the extracted 1 st-k level information of the Y-th key frame are stored into the storage importance coefficient matrix of the current key frame level information);
S500,
extracting the information of the 1 st to k th layers from the current Y-th key frame: x1,Y、X2,Y、……Xj,Y、……Xk,YAddingEntering into the current key frame level information matrix:
when Y is more than 0 and less than or equal to QkWhen the current key frame set is selected, all the extracted 1 st-k level information of the Y key frame is stored into the current key frame set;
the above steps are expressed as follows:
Xj,Iexpressing j level information of the I key frame;
when Q isj<Y≤Qj-1When j represents any natural number from 2 to k;
storing all the level information from level 1 to level j-1 of the Y key frame into the current key frame set, wherein the steps are expressed by adopting the following formula:
for j-th to k-th level information of the Y-th key frame, calculating layer by layer, and determining whether to store:
for any of the j-th to k-th layers:
first, from Zβ,1,Zβ,2……Zβ,I……Zβ,Y-1Finding the minimum value, which takes Zβ,mThe information storage importance parameter of the mth key frame beta level is minimum;
secondly, according to Zβ,m、Zβ,YDetermines whether to store the beta level information X of the Y-th key frameβ,Y:
When Z isβ,m>Zβ,Y
Then Zβ,YGiven a new value such that Zβ,YNo longer participate in the ranking;
Xβ,Ygiving a new value of NULL or 0, namely not storing the beta level information of the Yth key frame;
when Z isβ,m≤Zβ,Y
Then Zβ,mGiven a new value such that Zβ,mNo longer participate in the ranking;
Xβ,massigning a new value of NULL or 0, i.e., deleting the beta level information of the mth key frame and adding the beta level information X of the Yth key frameβ,YStoring, namely, the storage space of the beta level information of the mth key frame is reserved to the storage space of the beta level information of the Yth key frame, namely, the storage space of the beta level information is always kept at Qβ;
When Q is1When < Y:
for the level 1-k information of the Y key frame, calculating layer by layer, and determining whether to store:
for any of the 1 st to k th layers,
first, from Zβ,1,Zβ,2……Zβ,I……Zβ,Y-1Finding the minimum value, which takes Zβ,mThe information storage importance parameter of the mth key frame beta level is minimum;
secondly, according to Zβ,m、Zβ,YDetermines whether to store the beta level information X of the Y-th key frameβ,Y:
When Z isβ,m>Zβ,Y
Then Zβ,YGiven a new value such that Zβ,YNo longer participate in the ranking;
Xβ,Ygiving a new value of NULL or 0, namely not storing the beta level information of the Yth key frame;
when Z isβ,m≤Zβ,Y
Then Zβ,mGiven a new value such that Zβ,mNo longer participate in the ranking;
Xβ,massigning a new value of NULL or 0, i.e., deleting the beta level information of the mth key frame and adding the beta level information X of the Yth key frameβ,YStoring, namely, the storage space of the beta level information of the mth key frame is saved to the second of the Y key frameThe storage space of the beta-level information, i.e. the storage amount of the beta-level information is always kept at Qβ。
Further, "when Z" in step S500β,m>Zβ,Y,Zβ,YNew values are assigned ", NULL or + ∞.
Further, in step S100, k is not less than 4.
The invention has the beneficial effects that:
first, the basic idea of the present application is: abstracting and expressing information received by machine vision at different semantic levels, and coding the information from the information level; the higher the semantic feature of the higher layer, the more simplified the information quantity stored, the higher the extraction frequency, the longer the time stored in the SLAM system; conversely, the lower the level of the point cloud feature, the larger the amount of data stored, the lower the frequency of extraction, and the shorter the time for storage in the SLAM system. By setting different storage and retrieval index configurations, the performance and the efficiency of freely allocating storage/retrieval and long-time large-scale SLAM loop detection are achieved.
Secondly, based on the basic concept of the application, a method for storing the SLAM key frame matrix is provided: the storage is determined by means of a double coefficient, namely: and storing the importance coefficient matrix and the current key frame level information matrix to jointly determine how the newly acquired 1 st-k level information of the Y-th key frame is stored into the current key frame level information matrix.
Drawings
The invention will be further described in detail with reference to examples of embodiments shown in the drawings to which, however, the invention is not restricted.
Fig. 1 is a sequence layout diagram of information hierarchy according to the first embodiment.
Fig. 2 is a schematic diagram of "original image-dense point cloud feature" extraction according to the first embodiment.
Detailed Description
An efficient long-time large-scale SLAM method comprises the following steps:
s100, establishing an information hierarchy sequence of the key frames:
the information hierarchy sequence of the key frame is expressed as follows according to the descending order of the hierarchy level by level: level 1 information, level 2 information, level j information … … …, level k information:
s200, determining the storage capacity of each information level of the key frame:
the storage amount of the 1 st level information is Q1The storage amount of the 2 nd level information is Q2… … … … the storage capacity of the j level information is Qj… … … … the k-th level information has a storage capacity of Qk:
For Q1~QkFor example, there are: q1>Q2>……Qj-1>Qj>……Qk;
S300, the current key frame is the Yth key frame, and the information of layers 1 to k is extracted from the Yth key frame: x1,Y、X2,Y、……Xj,Y、……Xk,Y(ii) a And simultaneously calculating the storage importance coefficients of the information of the 1 st to k th layers extracted from the Yth key frame: z1,Y、Z2,Y、……Zj,Y、……Zk,Y;
X1,Y、X2,Y、……Xj,Y、……Xk,YRespectively represent: level 1 information of the Y-th key frame, level 2 information of the Y-th key frame, … … level j information of the Y-th key frame, … … level k information of the Y-th key frame;
Z1,Y、Z2,Y、……Zj,Y、……Zk,Yrespectively represent: a storage importance coefficient of level 1 information of a Y-th key frame, a storage importance coefficient of level 2 information of the Y-th key frame, a storage importance coefficient of a j-th level information of … … the Y-th key frame, a storage importance coefficient of a k-th level information of … … the Y-th key frame;
s400, the storage importance coefficients of the information of the 1 st to k th layers extracted from the current Y-th key frame are: z1,Y、Z2,Y、……Zj,Y、……Zk,YAdding into the current key frame (at this time, the corresponding current key frame has 1 st to Y-1 st key frames) Storage importance parameter set (i.e. storage importance coefficient matrix added to the current key frame level information):
all storage importance parameters of the extracted 1 st-k level information of the Y-th key frame are stored into a storage importance coefficient matrix of the current key frame level information;
the above steps can be expressed as follows:
S500,
extracting the information of the 1 st to k th layers from the current Y-th key frame: x1,Y、X2,Y、……Xj,Y、……Xk,YAdding to the current key frame set (i.e. adding to the current key frame level information matrix):
kwhen Y is more than 0 and less than or equal to QStoring all the extracted 1 st-k level information of the Y-th key frame into the current key frame set;
the above steps can be expressed as follows:
Xj,Iinformation of j level expressing the I key frame (the meaning of other expressions can be known from it, for example, Xk,YK-th level information expressed as the Y-th key frame).
k k-1When Q is more than Y and less than QAnd storing all the level information from level 1 to level k-1 of the Y-th key frame into the current key frame set, wherein the steps can be expressed as follows:
for the kth layer information of the Y-th key frame:
first, from Zk,1,Zk,2……Zk,I……Zk,Y-1Finding the minimum value, which takes Zk,mThe information storage importance parameter of the kth level of the mth key frame is minimum;
secondly, according to Zk,m、Zk,YDetermines whether to store the kth level information X of the Yth key framek,Y:
When Z isk,m>Zk,Y
Then Zk,YAssigning a new value (e.g., NULL or + ∞) (so that Zk,YNo longer participating in the ranking);
Xk,Ygiving a new value of NULL or 0, namely not storing the kth level information of the Yth key frame;
when Z isk,m≤Zk,Y
Then Zk,mGiven a new value such that Zk,mNo longer participate in the ranking;
Xk,massigning a new value of NULL or 0, i.e. removing the kth level information of the mth key frame and adding the kth level information X of the Yth key framek,YStoring, namely, the storage space of the k-level information of the mth key frame is saved to the storage space of the k-level information of the Yth key frame (namely, the storage amount of the k-level information is always kept at Qk)。
……
j j-1When Q is more than Y and less than QAnd storing all the level information from level 1 to level j-1 of the Y-th key frame into the current key frame set, wherein the steps can be expressed as follows:
for the j-th to k-th level (it should be noted that j in this application represents any natural number between [1, k ] rather than k-1, i.e. the order of j, k in english does not represent the distance between the two) information of the Y-th key frame, the calculation is performed layer by layer to determine whether to store:
for the jth level information of the yth key frame:
first, from Zj,1,Zj,2……Zj,I……Zj,Y-1Finding the minimum value, which takes Zj,mThe j level information storage importance parameter of the mth key frame is minimum;
secondly, according to Zj,m、Zj,YDetermines whether to store the j level information X of the Y key framej,Y:
When Z isj,m>Zj,Y
Then Zj,YAssigning a new value (e.g., NULL or + ∞) such that Zj,YNo longer participate in the ranking;
Xj,Ygiving a new value of NULL or 0, namely not storing the j level information of the Y key frame;
when Z isj,m≤Zj,Y
Then Zj,mGiven a new value such that Zj,mNo longer participate in the ranking;
Xj,massigning a new value of NULL or 0, i.e. deleting the j level information of the mth key frame and adding the j level information X of the Yth key framej,YStoring, namely, the storage space of the j level information of the mth key frame is made to be equal to the storage space of the j level information of the Yth key frame (namely, the storage amount of the j level information is always kept at Qj)。
For the j +1 th level information of the Y-th key frame:
first, from Zj+1,1,Zj+1,2……Zj+1,I……Zj+1,Y-1Finding the minimum value, which takes Zj+1,mThe j +1 th level information storage importance parameter of the mth key frame is minimum;
secondly, according to Zj+1,m、Zj+1,YDetermines whether to store the j +1 th level information X of the Yth key framej+1,Y:
When Z isj+1,m>Zj+1,Y
Then Zj+1,YAssigning a new value (e.g., NULL or + ∞) such that Zj+1,YNo longer participate in the ranking;
Xj+1,Ygiving a new value of NULL or 0, namely not storing the j +1 level information of the Yth key frame;
when Z isj+1,m≤Zj+1,Y
Then Zj+1,mGiven a new value such that Zj+1,mNo longer participate in the ranking;
Xj+1,massigning a new value of NULL or 0, i.e. removing the j +1 level information of the mth key frame and deleting the j +1 level information X of the Yth key framej+1,YStoring, namely, the storage space of the j +1 level information of the mth key frame is made to be equal to the storage space of the j +1 level information of the Yth key frame (namely, the storage amount of the j +1 level information is always kept at Qj+1)。
………
For the kth layer information of the Y-th key frame:
first, from Zk,1,Zk,2……Zk,I……Zk,Y-1Finding the minimum value, which takes Zk,mThe information storage importance parameter of the kth level of the mth key frame is minimum;
secondly, according to Zk,m、Zk,YDetermines whether to store the kth level information X of the Yth key framek,Y:
When Z isk,m>Zk,Y
Then Zk,YAssigning a new value (e.g., NULL or + ∞) (so that Zk,YNo longer participating in the ranking);
Xk,Ygiving a new value of NULL or 0, namely not storing the kth level information of the Yth key frame;
when Z isk,m≤Zk,Y
Then Zk,mGiven a new value such that Zk,mNo longer participate in the ranking;
Xk,massigning a new value of NULL or 0, i.e. the kth level information of the mth key frameDeleting information to delete the kth level information X of the Yth key framek,YStoring, namely, the storage space of the k-level information of the mth key frame is saved to the storage space of the k-level information of the Yth key frame (namely, the storage amount of the k-level information is always kept at Qk)。
……
2 1When Q is more than Y and less than Q,
Storing all the level 1 information of the Y-th key frame into the current key frame set, wherein the steps can be expressed as follows:
|X1,1 X1,2 ...... X1,I ...... X1,Y-1|—>|X1,1 X1,2 ...... X1,I ...... X1,Y-1X1,Y|
for the 2 nd to k th level information of the Y-th key frame, calculating layer by layer, and determining whether to store:
for the level 2 information of the Y-th key frame:
first, from Z2,1,Z2,2……Z2,I……Z2,Y-1Finding the minimum value, which takes Z2,mTo indicate that the storage importance parameter of the mth key frame level 2 information is minimum;
secondly, according to Z2,m、Z2,YDetermines whether to store the 2 nd level information X of the Yth key frame2,Y:
When Z is2,m>Z2,Y
Then Z2,YAssigning a new value (e.g., NULL or + ∞) such that Z2,YNo longer participate in the ranking;
X2,Ygiving a new value of NULL or 0, namely not storing the 2 nd level information of the Yth key frame;
when Z is2,m≤Z2,Y
Then Z2,mGiven a new value such that Z2,mNo longer participate in the ranking;
X2,massigning a new value of NULL or 0, i.e. mthDeleting the 2 nd level information of the key frame, and deleting the 2 nd level information X of the Y th key frame2,YStoring, namely, the storage space of the 2 nd level information of the mth key frame is saved to the storage space of the 2 nd level information of the Yth key frame (namely, the storage amount of the 2 nd level information is always kept at Q2)。
For the level 3 information of the Yth key frame:
first, from Z3,1,Z3,2……Z3,I……Z3,Y-1Finding the minimum value, which takes Z3,mTo indicate that the 3 rd level information storage importance parameter of the mth key frame is minimum;
secondly, according to Z3,m、Z3,YDetermines whether to store the 3 rd level information X of the Yth key frame3,Y:
When Z is3,m>Z3,Y
Then Z3,YAssigning a new value (e.g., NULL or + ∞) such that Z3,YNo longer participate in the ranking;
X3,Ygiving a new value of NULL or 0, namely not storing the 3 rd level information of the Yth key frame;
when Z is3,m≤Z3,Y
Then Z3,mGiven a new value such that Z3,mNo longer participate in the ranking;
X3,massigning a new value of NULL or 0, i.e., removing the 3 rd level information of the mth key frame and adding the 3 rd level information X of the yth key frame3,YStoring, namely, the storage space of the 3 rd level information of the mth key frame is saved to the 3 rd level information storage space of the Yth key frame (namely, the storage space of the 3 rd level information is always kept at Q3)。
………
For the kth layer information of the Y-th key frame:
first, from Zk,1,Zk,2……Zk,I……Zk,Y-1Finding the minimum value, which takes Zk,mI.e. the kth level information of the mth key frameThe storage importance parameter is minimal;
secondly, according to Zk,m、Zk,YDetermines whether to store the kth level information X of the Yth key framek,Y:
When Z isk,m>Zk,Y
Then Zk,YAssigning a new value (e.g., NULL or + ∞) (so that Zk,YNo longer participating in the ranking);
Xk,Ygiving a new value of NULL or 0, namely not storing the kth level information of the Yth key frame;
when Z isk,m≤Zk,Y
Then Zk,mGiven a new value such that Zk,mNo longer participate in the ranking;
Xk,massigning a new value of NULL or 0, i.e. removing the kth level information of the mth key frame and adding the kth level information X of the Yth key framek,YStoring, namely, the storage space of the k-level information of the mth key frame is saved to the storage space of the k-level information of the Yth key frame (namely, the storage amount of the k-level information is always kept at Qk)。
1When Q < Y:
For the level 1-k information of the Y key frame, calculating layer by layer, and determining whether to store:
for level 1 information of the Y-th key frame:
first, from Z1,1,Z1,2……Z1,I……Z1,Y-1Finding the minimum value, which takes Z1,mThe 1 st level information storage importance parameter of the mth key frame is minimum;
secondly, according to Z1,m、Z1,YDetermines whether to store the level 1 information X of the Yth key frame1,Y:
When Z is1,m>Z1,Y
Then Z1,YAssigning a new value (e.g., NULL or + ∞) such that Z1,YNo longer participate in the ranking;
X1,Ygiving a new value of NULL or 0, namely not storing the level 1 information of the Yth key frame;
when Z is1,m≤Z1,Y
Then Z1,mGiven a new value such that Z1,mNo longer participate in the ranking;
X1,massigning a new value of NULL or 0, i.e. deleting the level 1 information of the mth key frame and deleting the level 1 information X of the Yth key frame1,YStoring, namely, the storage space of the 1 st level information of the mth key frame is saved to the 1 st level information storage space of the Yth key frame (namely, the storage amount of the 1 st level information is always kept at Q1)。
For the level 2 information of the Y-th key frame:
first, from Z2,1,Z2,2……Z2,I……Z2,Y-1Finding the minimum value, which takes Z2,mTo indicate that the storage importance parameter of the mth key frame level 2 information is minimum;
secondly, according to Z2,m、Z2,YDetermines whether to store the 2 nd level information X of the Yth key frame2,Y:
When Z is2,m>Z2,Y
Then Z2,YAssigning a new value (e.g., NULL or + ∞) such that Z2,YNo longer participate in the ranking;
X2,Ygiving a new value of NULL or 0, namely not storing the 2 nd level information of the Yth key frame;
when Z is2,m≤Z2,Y
Then Z2,mGiven a new value such that Z2,mNo longer participate in the ranking;
X2,massigning a new value of NULL or 0, i.e., deleting the 2 nd level information of the mth key frame and deleting the 2 nd level information X of the Yth key frame2,YStoring, namely, the storage space of the 2 nd level information of the mth key frame is saved to the Yth key frameThe storage space of the 2 nd level information of the key frame (namely, the storage amount of the 2 nd level information is always kept at Q2)。
For the level 3 information of the Yth key frame:
first, from Z3,1,Z3,2……Z3,I……Z3,Y-1Finding the minimum value, which takes Z3,mTo indicate that the 3 rd level information storage importance parameter of the mth key frame is minimum;
secondly, according to Z3,m、Z3,YDetermines whether to store the 3 rd level information X of the Yth key frame3,Y:
When Z is3,m>Z3,Y
Then Z3,YAssigning a new value (e.g., NULL or + ∞) such that Z3,YNo longer participate in the ranking;
X3,Ygiving a new value of NULL or 0, namely not storing the 3 rd level information of the Yth key frame;
when Z is3,m≤Z3,Y
Then Z3,mGiven a new value such that Z3,mNo longer participate in the ranking;
X3,massigning a new value of NULL or 0, i.e., removing the 3 rd level information of the mth key frame and adding the 3 rd level information X of the yth key frame3,YStoring, namely, the storage space of the 3 rd level information of the mth key frame is saved to the 3 rd level information storage space of the Yth key frame (namely, the storage space of the 3 rd level information is always kept at Q3)。
………
For the kth layer information of the Y-th key frame:
first, from Zk,1,Zk,2……Zk,I……Zk,Y-1Finding the minimum value, which takes Zk,mThe information storage importance parameter of the kth level of the mth key frame is minimum;
secondly, according to Zk,m、Zk,YDetermines whether to store the kth level information X of the Yth key framek,Y:
When Z isk,m>Zk,Y
Then Zk,YAssigning a new value (e.g., NULL or + ∞) (so that Zk,YNo longer participating in the ranking);
Xk,Ygiving a new value of NULL or 0, namely not storing the kth level information of the Yth key frame;
when Z isk,m≤Zk,Y
Then Zk,mGiven a new value such that Zk,mNo longer participate in the ranking;
Xk,massigning a new value of NULL or 0, i.e. removing the kth level information of the mth key frame and adding the kth level information X of the Yth key framek,YStoring, namely, the storage space of the k-level information of the mth key frame is saved to the storage space of the k-level information of the Yth key frame (namely, the storage amount of the k-level information is always kept at Qk)。
An efficient long-term large-scale SLAM system, comprising: the system comprises an information level sequence establishing module, an information level memory space establishing module, a current key frame information extracting module, a current key frame information storage importance coefficient calculating module and a key frame set level information storage importance coefficient calculating and storing module; a key frame set level information calculating and storing module;
the information level sequence establishing module and the information level memory space establishing module are respectively connected with the current key frame information extracting module;
the information level sequence establishing module and the information level storage capacity establishing module are respectively connected with the current key frame information storage importance coefficient calculating module;
the current key frame information storage importance coefficient calculation module is connected with the key frame set level information storage importance coefficient calculation and storage module;
the current key frame information extraction module and the key frame set level information storage importance coefficient calculation and storage module are respectively connected with the key frame set level information calculation and storage module;
the information level sequence establishing module is used for establishing an information level sequence of a key frame set and guiding the current key frame information extracting module to extract information of a current key frame image, and is a carrier operated in the step S100;
the information level storage quantity establishing module is used for establishing different level information storage spaces of the key frame set and is a carrier operated in the step S200;
the current key frame information extraction module is used for extracting information of different levels from the current key frame image, and is a carrier operated in the step S300;
the current key frame information storage importance coefficient calculation module is used for calculating storage importance coefficients of different levels of information of the current key frame;
the key frame set level information storage importance coefficient calculation and storage module is used for storing the storage importance coefficients of different level information of the current key frame set;
the key frame set level information calculation and storage module is used for determining a storage mode (namely determining whether the key frame set level information is stored) for extracting information of different levels from the current key frame image transmitted by the current key frame information extraction module.
It should be noted that: the information level sequence establishing module and the information level storage capacity establishing module are not necessary technical features, that is, the S100 and the S200 are not necessarily set by the two modules, and a designer can directly set related parameters in the software system.
The first embodiment is as follows: as shown in fig. 1, an efficient long-term large-scale SLAM method performs hierarchical processing on information of a key frame set; the method comprises the following steps:
s100, establishing an information level sequence of the key frame set:
level 1 information: a feature word set (i.e., a BOW model) of a scene;
level 2 information: spatial relationship expression of spatial entities;
level 3 information: point/line/surface features;
level 4 information: dense point cloud features.
S200, establishing the storage priority of the information of the key frame set according to the information level sequence established in the step S100: the storage amount of the information level of the hierarchy is larger than that of the information level of the hierarchy, and the storage time of the information level of the hierarchy is longer than that of the information level of the hierarchy.
Specifically, S200 is to determine the storage capacity of the 1 st to 4 th information levels of the key frame:
the storage amount of the 1 st level information is Q1The amount of storage of the 2 nd level information is Q50000240000, the amount of storage of the 3 rd level information is Q310000, the storage amount of the 4 th hierarchy information is Q4=700:
S300, the current key frame is the Y-th key frame, as shown in fig. 1: extracting information of layers 1-4 from the Yth key frame: x1,Y、X2,Y、X3,Y、Xk,Y;
And simultaneously calculating the storage importance coefficients of the information of the 1 st to k th layers extracted from the Yth key frame: z1,Y、Z2,Y、Z3,Y、Zk,Y(ii) a The calculation principle of the storage importance coefficient can be specifically determined and adjusted by a system designer according to actual requirements and environments. For example: i am giving the machine in an indoor simulated environment a design of: the storage importance coefficient of the crossroad is positioned by 1.0, the storage importance coefficient of the T-shaped crossroad is positioned by 0.8, and the storage importance coefficient of the middle of the road is positioned by 0.5.
X1,Y、X2,Y、X3,Y、X4,YRespectively represent: level 1 information of a Y-th key frame, level 2 information of the Y-th key frame, level 3 information of the Y-th key frame, and level 4 information of the Y-th key frame;
Z1,Y、Z2,Y、Z3,Y、Z4,Yrespectively represent: storage importance coefficient of level 1 information of the Y-th key frame, storage importance coefficient of level 2 information of the Y-th key frame, storage importance coefficient of the Y-th key frameThe storage importance coefficient of the 3 rd level information and the storage importance coefficient of the 4 th level information of the Y th key frame;
s400, the storage importance coefficient of the information of the 1 st to 4 th layers extracted from the current Y-th key frame is: z1,Y、Z2,Y、Z3,Y、Z4,YAdding the importance coefficient into a storage importance coefficient matrix of the current key frame level information:
all storage importance parameters of the extracted 1 st-4 th-level information of the Y-th key frame are stored into a storage importance coefficient matrix of the current key frame level information;
the above steps can be expressed as follows:
s500, extracting the information of the 1 st to 4 th layers from the current Y-th key frame: x1,Y、X2,Y、X3,Y、X4,YAnd adding the key frame into a current key frame level information matrix:
extracting the information of the 1 st to 4 th layers of the current Y-th key frame: x1,Y、X2,Y、X3,Y、X4,YAnd adding the key frame into a current key frame level information matrix:
4when Y is more than 0 and less than or equal to QStoring all the extracted 1 st-4 th-level information of the Y-th key frame into a current key frame set;
the above steps can be expressed as follows:
j j-1when Q is more than Y and less than QAnd storing all the level information from level 1 to level j-1 of the Y-th key frame into the current key frame set, wherein the steps can be expressed as follows:
for the j-th to 4-th level information of the Y-th key frame, calculating layer by layer, and determining whether to store:
for any of the jth to 4 th layers,
first, from Zβ,1,Zβ,2……Zβ,I……Zβ,Y-1Finding the minimum value, which takes Zβ,mThe information storage importance parameter of the mth key frame beta level is minimum;
secondly, according to Zβ,m、Zβ,YDetermines whether to store the beta level information X of the Y-th key frameβ,Y:
When Z isβ,m>Zβ,Y
Then Zβ,YAssigning a new value (e.g., NULL or + ∞) such that Zβ,YNo longer participate in the ranking;
Xβ,Ygiving a new value of NULL or 0, namely not storing the beta level information of the Yth key frame;
when Z isβ,m≤Zβ,Y
Then Zβ,mGiven a new value such that Zβ,mNo longer participate in the ranking;
Xβ,massigning a new value of NULL or 0, i.e., deleting the beta level information of the mth key frame and adding the beta level information X of the Yth key frameβ,YStoring, namely, the storage space of the beta level information of the mth key frame is saved to the storage space of the beta level information of the Yth key frame (namely, the storage space of the beta level information is always kept at Qβ)。
1When Q < Y:
For the level 1-4 information of the Y key frame, calculating layer by layer, and determining whether to store:
for any beta-th layer of the 1 st to 4 th layers,
first, from Zβ,1,Zβ,2……Zβ,I……Zβ,Y-1Finding the minimum value, which takes Zβ,mThe information storage importance parameter of the mth key frame beta level is minimum;
secondly, according to Zβ,m、Zβ,YDetermines whether to store the beta level information X of the Y-th key frameβ,Y:
When Z isβ,m>Zβ,Y
Then Zβ,YAssigning a new value (e.g., NULL or + ∞) such that Zβ,YNo longer participate in the ranking;
Xβ,Ygiving a new value of NULL or 0, namely not storing the beta level information of the Yth key frame;
when Z isβ,m≤Zβ,Y
Then Zβ,mGiven a new value such that Zβ,mNo longer participate in the ranking;
Xβ,massigning a new value of NULL or 0, i.e., deleting the beta level information of the mth key frame and adding the beta level information X of the Yth key frameβ,YStoring, namely, the storage space of the beta level information of the mth key frame is saved to the storage space of the beta level information of the Yth key frame (namely, the storage space of the beta level information is always kept at Qβ)。
It should be noted that, according to the first embodiment: for any I key frame j layer information Xj,IThe value is not limited to be a numerical value, that is, it may be a character set, for example, as in the first embodiment: x1,IIs the BOW model data, and may also be a data set, for example: the first embodiment is as follows: x4,IIs a dense point cloud data set.
The above-mentioned embodiments are only for convenience of description, and are not intended to limit the present invention in any way, and those skilled in the art will understand that the technical features of the present invention can be modified or changed by other equivalent embodiments without departing from the scope of the present invention.
Claims (7)
1. An efficient long-term large-scale SLAM system, comprising: the system comprises a current key frame information extraction module, a current key frame information storage importance coefficient calculation module and a key frame set level information storage importance coefficient calculation and storage module; a key frame set level information calculating and storing module;
the current key frame information storage importance coefficient calculation module is connected with the key frame set level information storage importance coefficient calculation and storage module;
the current key frame information extraction module and the key frame set level information storage importance coefficient calculation and storage module are respectively connected with the key frame set level information calculation and storage module;
the current key frame information extraction module is used for extracting information of different levels from a current key frame image;
the current key frame information storage importance coefficient calculation module is used for calculating storage importance coefficients of different levels of information of the current key frame;
the key frame set level information storage importance coefficient calculation and storage module is used for storing the storage importance coefficients of different level information of the current key frame set;
the key frame set level information calculation and storage module is used for determining a storage mode for extracting information of different levels from the current key frame image transmitted by the current key frame information extraction module.
2. An efficient long-term large-scale SLAM system as recited in claim 1, further comprising: the system comprises an information level sequence establishing module and an information level memory space establishing module;
the information level sequence establishing module and the information level memory space establishing module are respectively connected with the current key frame information extracting module; the information level sequence establishing module and the information level storage capacity establishing module are respectively connected with the current key frame information storage importance coefficient calculating module.
3. The system according to claim 2, wherein the information hierarchy sequence creating module is configured to create an information hierarchy sequence of the key frame set, and is configured to instruct the current key frame information extracting module to extract information of the current key frame image, and is a carrier for the operation of step S100; the information level storage quantity establishing module is used for establishing different level information storage spaces of the key frame set, and is a carrier operated in step S200.
4. An efficient long-time large-scale SLAM method is characterized in that:
abstracting and expressing images received by machine vision into different levels of information;
the higher the information single memory space of the higher layer is, the higher the information single memory space of the lower layer is;
the higher the information number of the layer, the more; the lower layers have fewer information numbers.
5. An efficient long-time large-scale SLAM method is characterized by comprising the following steps:
s100, establishing an information hierarchy sequence of the key frames:
the information hierarchy sequence of the key frame is expressed as follows according to the descending order of the hierarchy level by level: level 1 information, level 2 information, level j information … … …, level k information:
s200, determining the storage capacity of each information level of the key frame:
the storage amount of the 1 st level information is Q1The storage amount of the 2 nd level information is Q2… … … … the storage capacity of the j level information is Qj… … … … the k-th level information has a storage capacity of Qk:
For Q1~QkFor example, there are: q1>Q2>……Qj-1>Qj>……Qk;
S300, the current key frame is the Yth key frame, and the information of layers 1 to k is extracted from the Yth key frame: x1,Y、X2,Y、……Xj,Y、……Xk,Y(ii) a And simultaneously calculating the storage importance coefficients of the information of the 1 st to k th layers extracted from the Yth key frame: z1,Y、Z2,Y、……Zj,Y、……Zk,Y;
X1,Y、X2,Y、……Xj,Y、……Xk,YRespectively represent: level 1 information of the Y-th key frame, level 2 information of the Y-th key frame, … … level j information of the Y-th key frame, … … level k information of the Y-th key frame;
Z1,Y、Z2,Y、……Zj,Y、……Zk,Yrespectively represent: a storage importance coefficient of level 1 information of a Y-th key frame, a storage importance coefficient of level 2 information of the Y-th key frame, a storage importance coefficient of a j-th level information of … … the Y-th key frame, a storage importance coefficient of a k-th level information of … … the Y-th key frame;
s400, the storage importance coefficients of the information of the 1 st to k th layers extracted from the current Y-th key frame are: z1,Y、Z2,Y、……Zj,Y、……Zk,YAdding the importance coefficient into a storage importance coefficient matrix of the current key frame level information:
S500,
extracting the information of the 1 st to k th layers from the current Y-th key frame: x1,Y、X2,Y、……Xj,Y、……Xk,Y,Adding the key frame level information into a current key frame level information matrix:
when 0 is present<Y≤QkWhen the current key frame set is selected, all the extracted 1 st-k level information of the Y key frame is stored into the current key frame set;
the above steps are expressed as follows:
Xj,Iexpressing j level information of the I key frame;
when Q isj<Y≤Qj-1When j represents any natural number from 2 to k;
storing all the level information from level 1 to level j-1 of the Y key frame into the current key frame set, wherein the steps are expressed by adopting the following formula:
for j-th to k-th level information of the Y-th key frame, calculating layer by layer, and determining whether to store:
for any of the j-th to k-th layers:
first, from Zβ,1,Zβ,2……Zβ,I……Zβ,Y-1Finding the minimum value, which takes Zβ,mThe information storage importance parameter of the mth key frame beta level is minimum;
secondly, according to Zβ,m、Zβ,YDetermines whether to store the beta level information X of the Y-th key frameβ,Y:
When Z isβ,m>Zβ,Y
Then Zβ,YGiven a new value such that Zβ,YNo longer participate in the ranking;
Xβ,Ygiving a new value of NULL or 0, namely not storing the beta level information of the Yth key frame;
when Z isβ,m≤Zβ,Y
Then Zβ,mGiven a new value such that Zβ,mNo longer participate in the ranking;
Xβ,massigning a new value of NULL or 0, i.e., deleting the beta level information of the mth key frame and adding the beta level information X of the Yth key frameβ,YStoring, i.e. the mth key frameThe storage space of the beta level information of the Y key frame is allowed to be the same as the storage space of the beta level information of the Y key frame, namely the storage space of the beta level information is always kept at Qβ;
When Q is1<When Y is in the range of:
for the level 1-k information of the Y key frame, calculating layer by layer, and determining whether to store:
for any of the 1 st to k th layers,
first, from Zβ,1,Zβ,2……Zβ,I……Zβ,Y-1Finding the minimum value, which takes Zβ,mThe information storage importance parameter of the mth key frame beta level is minimum;
secondly, according to Zβ,m、Zβ,YDetermines whether to store the beta level information X of the Y-th key frameβ,Y:
When Z isβ,m>Zβ,Y
Then Zβ,YGiven a new value such that Zβ,YNo longer participate in the ranking;
Xβ,Ygiving a new value of NULL or 0, namely not storing the beta level information of the Yth key frame;
when Z isβ,m≤Zβ,Y
Then Zβ,mGiven a new value such that Zβ,mNo longer participate in the ranking;
Xβ,massigning a new value of NULL or 0, i.e., deleting the beta level information of the mth key frame and adding the beta level information X of the Yth key frameβ,YStoring, namely, the storage space of the beta level information of the mth key frame is reserved to the storage space of the beta level information of the Yth key frame, namely, the storage space of the beta level information is always kept at Qβ。
6. The method of claim 5, wherein step S500 is executed as Zβ,m>Zβ,Y,Zβ,YNew values are assigned ", NULL or + ∞.
7. The method of claim 5 or 6, wherein k is 4 ≦ k in step S100.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110658243.4A CN113392370B (en) | 2021-06-15 | 2021-06-15 | SLAM system |
CN202111532176.8A CN114154117B (en) | 2021-06-15 | 2021-06-15 | SLAM method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110658243.4A CN113392370B (en) | 2021-06-15 | 2021-06-15 | SLAM system |
Related Child Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111532176.8A Division CN114154117B (en) | 2021-06-15 | 2021-06-15 | SLAM method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113392370A true CN113392370A (en) | 2021-09-14 |
CN113392370B CN113392370B (en) | 2022-01-04 |
Family
ID=77620882
Family Applications (2)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111532176.8A Active CN114154117B (en) | 2021-06-15 | 2021-06-15 | SLAM method |
CN202110658243.4A Active CN113392370B (en) | 2021-06-15 | 2021-06-15 | SLAM system |
Family Applications Before (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111532176.8A Active CN114154117B (en) | 2021-06-15 | 2021-06-15 | SLAM method |
Country Status (1)
Country | Link |
---|---|
CN (2) | CN114154117B (en) |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20150023602A1 (en) * | 2013-07-19 | 2015-01-22 | Kamil Wnuk | Fast recognition algorithm processing, systems and methods |
US20150234477A1 (en) * | 2013-07-12 | 2015-08-20 | Magic Leap, Inc. | Method and system for determining user input based on gesture |
US20180089832A1 (en) * | 2016-09-29 | 2018-03-29 | Intel Corporation | Place recognition algorithm |
CN108830220A (en) * | 2018-06-15 | 2018-11-16 | 山东大学 | The building of vision semantic base and global localization method based on deep learning |
CN111461140A (en) * | 2020-03-30 | 2020-07-28 | 北京航空航天大学 | Linear descriptor construction and matching method suitable for S L AM system |
CN112240768A (en) * | 2020-09-10 | 2021-01-19 | 西安电子科技大学 | Visual inertial navigation fusion SLAM method based on Runge-Kutta4 improved pre-integration |
CN112904901A (en) * | 2021-01-14 | 2021-06-04 | 吉林大学 | Path planning method based on binocular vision slam and fusion algorithm |
Family Cites Families (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR100371513B1 (en) * | 1999-12-06 | 2003-02-07 | 주식회사 팬택앤큐리텔 | Method and apparatus of summerizing and browsing video sequences using fidelity values of key frame stored in its edge of key frame hierarchy |
US9674507B2 (en) * | 2013-04-30 | 2017-06-06 | Qualcomm Incorporated | Monocular visual SLAM with general and panorama camera movements |
US11187536B2 (en) * | 2018-01-12 | 2021-11-30 | The Trustees Of The University Of Pennsylvania | Probabilistic data association for simultaneous localization and mapping |
US11307049B2 (en) * | 2018-07-19 | 2022-04-19 | Uisee Technologies (Beijing) Co., Ltd | Methods, apparatuses, systems, and storage media for storing and loading visual localization maps |
CN109671120A (en) * | 2018-11-08 | 2019-04-23 | 南京华捷艾米软件科技有限公司 | A kind of monocular SLAM initial method and system based on wheel type encoder |
CN109711365A (en) * | 2018-12-29 | 2019-05-03 | 佛山科学技术学院 | A kind of vision SLAM winding detection method and device merging semantic information |
CN109903314A (en) * | 2019-03-13 | 2019-06-18 | 腾讯科技(深圳)有限公司 | A kind of method, the method for model training and the relevant apparatus of image-region positioning |
CN110533587B (en) * | 2019-07-03 | 2023-06-13 | 浙江工业大学 | SLAM method based on visual priori information and map restoration |
CN110598713B (en) * | 2019-08-06 | 2022-05-06 | 厦门大学 | Intelligent image automatic description method based on deep neural network |
CN112258600A (en) * | 2020-10-19 | 2021-01-22 | 浙江大学 | Simultaneous positioning and map construction method based on vision and laser radar |
-
2021
- 2021-06-15 CN CN202111532176.8A patent/CN114154117B/en active Active
- 2021-06-15 CN CN202110658243.4A patent/CN113392370B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20150234477A1 (en) * | 2013-07-12 | 2015-08-20 | Magic Leap, Inc. | Method and system for determining user input based on gesture |
US20150023602A1 (en) * | 2013-07-19 | 2015-01-22 | Kamil Wnuk | Fast recognition algorithm processing, systems and methods |
US20180089832A1 (en) * | 2016-09-29 | 2018-03-29 | Intel Corporation | Place recognition algorithm |
CN108830220A (en) * | 2018-06-15 | 2018-11-16 | 山东大学 | The building of vision semantic base and global localization method based on deep learning |
CN111461140A (en) * | 2020-03-30 | 2020-07-28 | 北京航空航天大学 | Linear descriptor construction and matching method suitable for S L AM system |
CN112240768A (en) * | 2020-09-10 | 2021-01-19 | 西安电子科技大学 | Visual inertial navigation fusion SLAM method based on Runge-Kutta4 improved pre-integration |
CN112904901A (en) * | 2021-01-14 | 2021-06-04 | 吉林大学 | Path planning method based on binocular vision slam and fusion algorithm |
Non-Patent Citations (2)
Title |
---|
CASTRO GASTÓN 等: ""Efficient on-board Stereo SLAM through constrained-covisibility strategies"", 《ROBOTICS AND AUTONOMOUS SYSTEMS》 * |
徐岩 等: ""基于BA优化和KL散度的RGB-D SLAM***"", 《东北大学学报(自然科学版)》 * |
Also Published As
Publication number | Publication date |
---|---|
CN113392370B (en) | 2022-01-04 |
CN114154117B (en) | 2022-08-23 |
CN114154117A (en) | 2022-03-08 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111062951B (en) | Knowledge distillation method based on semantic segmentation intra-class feature difference | |
CN109992636A (en) | Space-time code method, temporal index and querying method and device | |
CN104200523B (en) | A kind of large scene three-dimensional rebuilding method for merging additional information | |
CN114283164B (en) | Breast cancer pathological section image segmentation prediction system based on UNet3+ | |
JPH0452856A (en) | Method and device for allocating processing of document | |
US10387753B1 (en) | Learning method and learning device for convolutional neural network using 1×1 convolution for image recognition to be used for hardware optimization, and testing method and testing device using the same | |
CN110180186A (en) | A kind of topographic map conversion method and system | |
WO2016095068A1 (en) | Pedestrian detection apparatus and method | |
CN113850136A (en) | Yolov5 and BCNN-based vehicle orientation identification method and system | |
CN115100652A (en) | Electronic map automatic generation method based on high-resolution remote sensing image | |
CN113392370B (en) | SLAM system | |
CN114595302A (en) | Method, device, medium, and apparatus for constructing multi-level spatial relationship of spatial elements | |
CN115577106B (en) | Text classification method, device, equipment and medium based on artificial intelligence | |
CN111027551A (en) | Image processing method, apparatus and medium | |
CN112395462A (en) | Method, device, equipment and storage medium for searching matching subgraph in graph data stream | |
CN114912516A (en) | Cross-domain target detection method and system for coordinating feature consistency and specificity | |
CN115131362A (en) | Large-scale point cloud local area feature coding method | |
CN114972956A (en) | Target detection model training method, device, equipment and storage medium | |
CN115187614A (en) | Real-time simultaneous positioning and mapping method based on STDC semantic segmentation network | |
CN113010642A (en) | Semantic relation recognition method and device, electronic equipment and readable storage medium | |
CN117407727B (en) | Vector similarity determining method and vector searching method | |
CN117034942B (en) | Named entity recognition method, device, equipment and readable storage medium | |
CN116187419B (en) | Automatic hierarchical system construction method based on text chunks | |
CN117634486B (en) | Directional 3D instance segmentation method based on text information | |
CN106777131A (en) | A kind of querying method of High dimensional space data, device and computer-readable medium |
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 | ||
TR01 | Transfer of patent right |
Effective date of registration: 20240117 Address after: 1208-1211, 12th Floor, No. 51 Xueyuan Road, Haidian District, Beijing, 100080 Patentee after: METOAK TECHNOLOGY (BEIJING) CO.,LTD. Patentee after: Yuanxiang Technology (Suzhou) Co.,Ltd. Address before: 215163 floor 8, building 10, Su hi tech software park, No. 78, Keling Road, science and Technology City, high tech Zone, Suzhou, Jiangsu Patentee before: Yuanxiang Technology (Suzhou) Co.,Ltd. |
|
TR01 | Transfer of patent right |