CN104572639A - Centralized distribution type big traffic data clustering method aiming at behavior model of pedestrians - Google Patents
Centralized distribution type big traffic data clustering method aiming at behavior model of pedestrians Download PDFInfo
- Publication number
- CN104572639A CN104572639A CN201310468804.XA CN201310468804A CN104572639A CN 104572639 A CN104572639 A CN 104572639A CN 201310468804 A CN201310468804 A CN 201310468804A CN 104572639 A CN104572639 A CN 104572639A
- Authority
- CN
- China
- Prior art keywords
- data
- point
- unique
- unique point
- region
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Probability & Statistics with Applications (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a centralized distribution type big traffic data clustering method aiming at the behavior model of pedestrians. The clustering method comprises the following steps: extracting feature points of distributed data of pedestrians running a red light; collecting information of coordinates, weightiness, radiuses referring to ranges and the like of the feature points of multiple locations; adopting a density based clustering algorithm to cluster global data in order to acquire the global data mode; feeding back the global clustering result to a single location. As for distributed data clustering, when the data amount is relatively big, the cost of data copy and transmission can usually be unbearable; the clustering method can use the feature points to represent data of a single zone according to the actual data point transmission capability, and then adopts clustering of the concentrated feature points of each zone to replace concentrated big data clustering, so that the problem caused by big data migration is effectively solved; the phenomenon of pedestrians running the red light is an important factor resulting in traffic accidents, the novel clustering method provided by the invention particularly aims at the centralized distribution type big traffic data of the behavior model of pedestrians, and has high practical significance and practicalness.
Description
Technical field
The invention belongs to large data digging method field, be specifically related to the centralized large data pattern discover method of a kind of distribution
Background technology
Along with the arriving of large data age, under increasing application scenario, people need data scale to be processed to expand to TB even PB rank, and wish therefrom fast and effeciently to excavate reliable, useful hiding Info.Therefore, how rapidly and accurately from large data tap value information current significant.Cluster analysis, as a kind of one of core technology of Data Mining, usually can process as the early stage of other data mining algorithms.But in the face of so huge data scale, traditional clustering method can not meet reality need in data storage, computational complexity etc.
Such as we will excavate the data in Beijing, Shanghai, Guangzhou, and idea data centralization is got up process again intuitively, such as the data in Shanghai and Guangzhou can be copied directly to Beijing.But after data volume reaches certain scale, this copy just seems it is not so reality, and namely under our said large data background, copy or the cost concentrated may be unacceptable, as time, equipment, money etc.So just require that we must change the data digging method of traditional local type with improving or even overturn formula.
The present invention proposes a kind of pattern discovery methods for distributed large data, first respectively feature point extraction is carried out to the data of different location, data scale significantly drops to the magnitude that can transmit, then the data characteristics of different regions point is focused on, excavate the pattern of all data.Such as the data of regional certain industry in Beijing, Shanghai, three, Guangzhou, we first extract some representational unique points to the data in single area, then three regional data characteristics points are aggregated into Beijing, Pekinese's machine carries out the excavation of all data, finally excavates the one-piece pattern of the sector data.
In concrete application aspect, the present invention mainly for be the clustering problem of the large data of the centralized traffic of distribution of pedestrian behavior pattern.Further, instantly traffic hazard takes place frequently, accounting for significant proportion due to pedestrian running red light causes, if the behavior pattern of the pedestrian that energy road pavement pedestrian especially goes across the road is carried out discovery and correspondingly formulates counte-rplan, the generation of pedestrian running red light event just effectively can be controlled thus the appearance of corresponding minimizing traffic hazard.
Pedestrian's data of making a dash across the red light can be found by monitoring video, can also obtain road surface at that time by image processing techniques simultaneously and wait for the information such as pedestrian's quantity of going across the road simultaneously, and real-time average link speed data and the information such as width of roadway and red light duration are also known.Each pedestrian made a dash across the red light correspond to the information of these dimensions, each area also exists the data of a large amount of pedestrians made a dash across the red light, but from the angle in the whole nation, want the behavior pattern excavating national jaywalker but to need in the face of the concentrated problem of large Data Migration, the invention of a large data clustering method of the centralized traffic of effective distribution seems and is extremely necessary.
Summary of the invention
Of the present invention to liking distributed large data, first propose a kind of data characteristics point extracting method for single place, then the characteristic point information of different location is aggregated into a place, then carries out the excavation of overall data, utilize Name-based Routing to carry out cluster.
1. individually point data characteristic point information extracts
For the data in single place, we represent by the weight of several unique points and each Feature point correspondence.
The 1.1 data average densities determining each place
According to the scope of data, by data standard in a region, the data as two dimension are the minimum rectangle that can cover all data points, and three-dimensional data is a minimum cube that can cover all data points, by that analogy.
The average density D of data is defined as the ratio that S is estimated in the number N of data point and normalized area, namely
1.2 radiuses determining unique point overlay area
First, the number that we define the transmission data point that single place can bear is M, so the number of unique point that finally this place is transmitted to data processing centre (DPC) is equally also M, here the content transmitted comprises coordinate and the weight of unique point simultaneously, and weight sum is the data point summation of this area.
Ideally, the region represented by each unique point should be do not occur simultaneously, so we define estimating representated by average each unique point
The conveniently searching of unique point, the region of Based on Feature Points is defined as circle when data for during two dimension by us, is defined as spheroid time three-dimensional, by that analogy, if the dimension of data is n, the radius in region represented by unique point is R, so has according to multidimensional spheroid cubature formula
Distinguishingly, as n=2, S
0=π R
2, during n=3,
thus, radius R can be obtained.
1.3 cyclic search unique points
(1) in all data points, Stochastic choice one is not by the point marked, will with this point for the center of circle, R is the region that the region of radius can represent as this point, if the data point density in this region is greater than the K of average density doubly, then using this o'clock as a unique point, the number of the point in region represented by it is as the weight of this unique point, and the data point in region represented by this unique point is all marked, no longer will consider these points in the process of next search characteristics point.If the number of the unique point searched reaches the maximal value M of the point of the transmission that this area can bear, then stop search, transmit coordinate and the weight of all unique points;
(2) allly still do not found new unique point by the data point stamping label if traveled through, then K value is subtracted 1, enter (1), stop search until K value is less than or equal to 1, transmit coordinate and the weight of all unique points.
2. agglomerative clustering
The characteristic point information (comprising coordinate, weight) in multiple place is focused on a place with acceptable cost by us, is further processed.The present invention uses density clustering method.
The characteristic point information in multiple place is gathered in a unified place, carries out cluster as follows.
(1) all unique points be placed in a coordinate system, the density of each unique point represents by the weight of its correspondence.The radius of scope represented by each unique point is identical with at the radius in its original area.
(2) circulate the process of carrying out below:
Random selecting unique point be not labeled is as the root node of one tree, the generative process set is carried out according to the principle of breadth First, each node wherein set is a unique point, all nodes of one tree belong to one and cluster, and each newly-generated child node meets following two conditions:
A the distance of () child node and father node is no more than the radius in region representated by character pair point and (namely tangent);
B corresponding to () child node, the density of unique point is greater than the density of root node character pair point.
Whether often increase a child node all has mark to make a decision to it, if not mark, then this node characteristic of correspondence point is included into and currently clusters and mark, if there has been mark, then clustering corresponding to the cluster corresponding to the tree of current generation and this node newly increased is merged into one to cluster, and start the search procedure of a new tree.
Until: all unique points be all classified as one cluster in the middle of.
3. the analysis of all data
Pass through said process, obtain the clustering information of all data, we can analyze all data and each feature clustered as required on the one hand, and concrete clustering information is returned to each area by us on the other hand, can provide decision-making by the overall classification of data.
Below for three areas, the process of cluster after the demonstration extraction of unique point and concentrated unique point.Accompanying drawing 2, accompanying drawing 3, accompanying drawing 4 are three regional Data distribution8 schematic diagram respectively.
The scale at three area count strong points is respectively 10000,10000,10000, and is two-dimemsional number strong point, and two dimensions are respectively average velocity and the red light duration of the on-site vehicle traveling of red light.Need now to do total cluster analysis to three areas, but there is restriction in the quantity of data point that each area can be transmitted, establishes here and be 500, be i.e. M=500.
For convenience of the effect of observing feature point extraction, the present invention's hypothesis is not when transmitting difficulty, and the schematic diagram that all data points (totally 30000) pool together as shown in Figure 5.
First the feature point extraction in single area is carried out.Radius according to the Data distribution8 determination Based on Feature Points region of each department is respectively R1=1.1411, R2=1.0577, R3=1.2030.During regulation first time circulation searching unique point, represent that the density in region is more than or equal to 10 times of average density and just can becomes unique point, even K=10, every circulation primary afterwards, K value subtracts 1 until K=1, namely thinks and represents that the density of data point in region is not representative lower than the point of average density.
According to above-mentioned feature point extraction algorithm, three areas obtain 260,336,269 unique points respectively, using the number of data point in region represented by each unique point as weight.For area one, the red data point in accompanying drawing 6 represents the unique point found, and be that the circle in the center of circle represents scope for it with unique point, the point within the scope of expression is by Based on Feature Points.
Three regional characteristic point informations (comprising the radius in the coordinate of unique point, weight, expression region) are focused on a place, as shown in Figure 7.
The weight of each unique point is regarded as the density of this point, according to above-mentioned density clustering clustering, the unique point gathered is divided into 17 classes, cluster code is that the number of the contained unique point that clusters of 1,2,23 account for great majority, find well out, to have fully demonstrated the validity of this algorithm by several main class.Further, we find that some cluster only comprises areal unique point, as cluster number be 13 unique point only from a regional cluster number be 10 only second-class from area, and what is more important, some clusters cover the unique point in multiple area, and the number that this cluster contains unique point is huge, be the global schema of whole data, as the cluster that cluster code is 1 and 2, and this cluster may not be Main Patterns in single area, such as area one, cluster code be 13 or 23 cluster more representative.Concrete outcome is shown in accompanying drawing 8.
The system flowchart of whole invention is shown in accompanying drawing 1.
In order to specific implementation this method, need to follow following steps:
Step 1: carry out respective independent process to the distributed data in multiple place, extract minutiae information, comprises the coordinate of unique point, weight and represents the radius in region.
Step 1.1: the data average density determining each department according to the number of the dimension of data, distribution and data point.
Step 1.2: the radius in the number determination Based on Feature Points region of the transmission data point can born according to various places.
Step 1.3: determine that (packing density as before constantly in Based on Feature Points region is 10 times of average density for the relation of packing density and average density in Based on Feature Points region, successively decrease successively afterwards until equal average density), and on this basis cyclic search unique point until travel through the number upper limit that number that all data points also can not find qualified point or unique point reaches the transmission data point can born this area.
Step 2: the characteristic point information (coordinate, weight, radius) in each place is copied (transmission) with acceptable cost and is aggregated into a place of specifying.
Step 3: the data in each place are placed in the same coordinate system, the weight of a unique point are considered as the density in region represented by it by process combined data.
Step 4: the data gathered are divided into some clustering according to density-based algorithms.
Step 4.1: the pattern analysis carrying out overall data as required.
Step 4.2: clustering information is returned to each area.
Accompanying drawing explanation
Fig. 1: system flowchart
Fig. 2: the Data distribution8 schematic diagram in area one
Fig. 3: the Data distribution8 schematic diagram in area two
Fig. 4: the Data distribution8 schematic diagram in area three
Fig. 5: all data point schematic diagram (supposing to there is not transmission difficulty)
Fig. 6: a regional feature point extraction schematic diagram
Fig. 7: the unique point schematic diagram after gathering
Fig. 8: 2-D data cluster result schematic diagram
Specific embodiments
Below in conjunction with drawings and Examples, the inventive method is further described.
The inventive method carries out the examples show of work in every for the traffic data that makes a dash across the red light:
Step one, feature point extraction
Pedestrian running red light is the large inducement that traffic hazard occurs, if can find the behavior pattern of jaywalker, just can suit the remedy to the case, avoid the frequent generation of pedestrian running red light, and then greatly reduce the generation of traffic hazard.Here, we will find jaywalker's behavior pattern in the whole nation, namely will carry out cluster to the jaywalker in the whole nation, and need the unified process to a place of the huge data summarization in each area, the present invention can realize this work easily.
We think whether pedestrian makes a dash across the red light mainly by the impact of four aspects: 1. red light duration; 2. wait for number; 3. road width; 4. the real-time average speed in this section.Wherein, whether pedestrian
The number of making a dash across the red light and wait for is obtained by camera image treatment technology, red light duration, road width, and the real-time average speed in this section is the data that traffic department has grasped.That is the data of a jaywalker are 4 dimension data.
Be located at i-th the regional maximin of data on four dimensions and be respectively max_i_1, min_i_1, max_i_2, min_i_2, max_i_3, min_i_3, max_i_4, min_i_4.Then the estimating for S=(max_i_1-min_i_1) * (max_i_2-min_i_2) * (max_i_3-min_i_3) * (max_i_4-min_i_4) of regional i data area
If the upper limit of regional i uploading data point is M, then represented by each unique point, estimating of space is S0=S/M, then according to multidimensional spheroid cubature formula
The radius in the Based on Feature Points region of n=4 herein, regional i can be asked.
If China's traffic data storage in east, western part, three, middle part area, i.e. i=1,2,3.Each area containing 6000 data points, the upper limit can transmitting four-dimensional data point be set to 500 namely M be equal to 500, afterwards for each area, according to algorithm circulation searching unique point.
Set K initial value as 10 times herein, namely initial cycle thinks that representing the point that region is greater than average density 10 times has extremely strong representativeness.
(1) in all data points, Stochastic choice one is not by the point marked, will with this point for the center of circle, R is the region that the region of radius can represent as this point, if the data point density in this region is greater than the K of average density doubly, then using this o'clock as a unique point, the number of the point in region represented by it is as the weight of this unique point, and the data point in region represented by this unique point is all marked, no longer will consider these points in the process of next search characteristics point.If the number of the unique point searched reaches the maximal value M of the point of the transmission that this area can bear, then stop search, transmit coordinate and the weight of all unique points.
(2) allly still do not found new unique point by the data point stamping label if traveled through, then K value is subtracted 1, enter (1), stop search until K value is less than or equal to 1, transmit coordinate and the weight of all unique points.
Finally, we obtain three regional feature point number and are respectively 324,375,94.
Step 2, summary information/transmission data
The radius information three places being amounted to the region representated by the coordinate information of 793 unique points, the weight information of each Feature point correspondence and each unique point is transferred to a central place with the cost that can bear
Step 3, be placed in the same coordinate system by the unique point that three places are gathered, the weight of unique point is considered as the density of unique point within the scope of its expression, represents that the radius of scope there are differences because of the difference of zone data feature.
Step 4, density clustering
Circulate the process of carrying out below:
Random selecting unique point be not labeled is as the root node of one tree, the generative process set is carried out according to the principle of breadth First, each node wherein set is a unique point, all nodes of one tree belong to one and cluster, and each newly-generated child node meets following two conditions:
A the distance of () child node and father node is no more than the radius in region representated by character pair point and (namely tangent);
B corresponding to () child node, the density of unique point is greater than the density of root node character pair point.
Whether often increase a child node all has mark to make a decision to it, if not mark, then this node characteristic of correspondence point is included into and currently clusters and mark, if there has been mark, then clustering corresponding to the cluster corresponding to the tree of current generation and this node newly increased is merged into one to cluster, and start the search procedure of a new tree.
Until: all unique points be all classified as one cluster in the middle of.
Finally be divided into 25 classes, removing only clusters containing the isolated of a unique point, is 18 classes.
Step 5, analysing and decision
On the one hand, the pedestrian behavior mode type for the whole nation is analyzed, and according to the feature of the behavior pattern of different jaywalkers, makes corresponding solution, as according to condition of road surface adjustment lamp time etc.
On the other hand, the result of global clustering is turned back to each place, the difference of more local pedestrian behavior pattern and national pedestrian behavior pattern.
Claims (5)
1. obtain the information of multiple dimensions of jaywalker, the present invention selects 2 (red light duration, the real-time average speeds in this section) and 4 dimensions (red light duration, wait number, road width, the real-time average speed in this section) respectively, and, as the factor affecting pedestrian running red light behavior, namely the dimension of jaywalker's data point is determined by the dimension selected.
2. the long-pending average density as this area's data that the number defining single regional data point is multiplied divided by the difference that the maximal value of data in each dimension in this area deducts minimum value.
3. the estimating of the region that represents of feature points (two dimension is the area of circle, three-dimensional be the volume of ball by that analogy) maximal value that is multiplied by the number that can bear transfer point be the maximal value of data in each dimension in this area deduct that the difference of minimum value is multiplied long-pending, determine the radius in Based on Feature Points region thus.
4. for the data in single place, extract the unique point in single place according to following algorithm, comprise the radius of unique point coordinate, unique point weight, Based on Feature Points scope:
(1) in all data points, Stochastic choice one is not by the point marked, will with this point for the center of circle, R is the region that the region of radius can represent as this point, if the data point density in this region is greater than the K of average density doubly (the present invention gets 10), then using this o'clock as a unique point, the number of the point in region represented by it is as the weight of this unique point, and the data point in region represented by this unique point is all marked, no longer will consider these points in the process of next search characteristics point.If the number of the unique point searched reaches the maximal value M of the point of the transmission that this area can bear, then stop search, transmit coordinate and the weight of all unique points;
(2) allly still new unique point is not found by the data point stamping label if traveled through, then K (the present invention gets 10) value is subtracted 1, enter (1), stop search until K (the present invention gets 10) value is less than or equal to 1, transmit coordinate and the weight of all unique points.
5. the radius of the unique point coordinate in multiple place, unique point weight, Based on Feature Points scope is aggregated into a place with the cost that can bear, for the data after gathering, according to following algorithm cluster:
Random selecting unique point be not labeled is as the root node of one tree, the generative process set is carried out according to the principle of breadth First, each node wherein set is a unique point, all nodes of one tree belong to one and cluster, and each newly-generated child node meets following two conditions:
A the distance of () child node and father node is no more than the radius in region representated by character pair point and (namely tangent);
B corresponding to () child node, the density of unique point is greater than the density of root node character pair point;
Whether often increase a child node all has mark to make a decision to it, if not mark, then this node characteristic of correspondence point is included into and currently clusters and mark, if there has been mark, then clustering corresponding to the cluster corresponding to the tree of current generation and this node newly increased is merged into one to cluster, and start the search procedure of a new tree;
Until: all unique points be all classified as one cluster in the middle of.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310468804.XA CN104572639A (en) | 2013-10-10 | 2013-10-10 | Centralized distribution type big traffic data clustering method aiming at behavior model of pedestrians |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310468804.XA CN104572639A (en) | 2013-10-10 | 2013-10-10 | Centralized distribution type big traffic data clustering method aiming at behavior model of pedestrians |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN104572639A true CN104572639A (en) | 2015-04-29 |
Family
ID=53088742
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310468804.XA Pending CN104572639A (en) | 2013-10-10 | 2013-10-10 | Centralized distribution type big traffic data clustering method aiming at behavior model of pedestrians |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104572639A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1912950A (en) * | 2006-08-25 | 2007-02-14 | 浙江工业大学 | Device for monitoring vehicle breaking regulation based on all-position visual sensor |
| CN1966858A (en) * | 2006-03-13 | 2007-05-23 | 孙世钧 | Convenient and efficient mass transportation system composed mostly of zoned type circulation routes |
| US20100088600A1 (en) * | 2008-10-07 | 2010-04-08 | Hamilton Ii Rick A | Redirection of an avatar |
| US20130075464A1 (en) * | 2011-09-26 | 2013-03-28 | Erik Van Horn | Method of and apparatus for managing and redeeming bar-coded coupons displayed from the light emitting display surfaces of information display devices |
| CN103268707A (en) * | 2013-04-26 | 2013-08-28 | 东南大学 | Signal regulating method for pedestrian crossing road section of bus prior passage |
-
2013
- 2013-10-10 CN CN201310468804.XA patent/CN104572639A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1966858A (en) * | 2006-03-13 | 2007-05-23 | 孙世钧 | Convenient and efficient mass transportation system composed mostly of zoned type circulation routes |
| CN1912950A (en) * | 2006-08-25 | 2007-02-14 | 浙江工业大学 | Device for monitoring vehicle breaking regulation based on all-position visual sensor |
| US20100088600A1 (en) * | 2008-10-07 | 2010-04-08 | Hamilton Ii Rick A | Redirection of an avatar |
| US20130075464A1 (en) * | 2011-09-26 | 2013-03-28 | Erik Van Horn | Method of and apparatus for managing and redeeming bar-coded coupons displayed from the light emitting display surfaces of information display devices |
| CN103268707A (en) * | 2013-04-26 | 2013-08-28 | 东南大学 | Signal regulating method for pedestrian crossing road section of bus prior passage |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107134142B (en) | A kind of urban road method for predicting based on multisource data fusion | |
| CN109410577B (en) | Self-adaptive traffic control subarea division method based on space data mining | |
| CN106504233B (en) | Unmanned plane inspection image electric power widget recognition methods and system based on Faster R-CNN | |
| CN108920481B (en) | Road network reconstruction method and system based on mobile phone positioning data | |
| CN110111574B (en) | Urban traffic imbalance evaluation method based on flow tree analysis | |
| CN107766808A (en) | The method and system that Vehicle Object motion track clusters in road network space | |
| Wang et al. | Exploring traffic congestion correlation from multiple data sources | |
| CN106251625A (en) | Three-dimensional urban road network global state Forecasting Methodology under big data environment | |
| CN111027743B (en) | OD optimal path searching method and device based on hierarchical road network | |
| CN107330422A (en) | A kind of method for carrying out mima type microrelief classification to semiarid zone based on high accuracy number elevation model | |
| CN103824450B (en) | Based on the large-scale activity Special running layout of roads method of traffic behavior rule | |
| CN107977734A (en) | A kind of Forecasting Methodology based on mobile Markov model under space-time big data | |
| CN108961758A (en) | A kind of crossing broadening lane detection method promoting decision tree based on gradient | |
| CN108597227A (en) | Road traffic flow forecasting method under freeway toll station | |
| CN109191849A (en) | A kind of traffic congestion Duration Prediction method based on multi-source data feature extraction | |
| CN109948690A (en) | A kind of high-speed rail scene perception method based on deep learning and structural information | |
| CN104200387A (en) | Land use zoning method giving consideration to geographic entity semantic similarity | |
| CN106373384B (en) | Outlying district regular bus circuit Real-time Generation | |
| CN103971312B (en) | Rural network node radiation domain-oriented rural residential area renovation zoning method | |
| CN104615881A (en) | User normal track analysis method based on movable position application | |
| CN109859480A (en) | Congested link modeling and appraisal procedure based on complex network | |
| CN107194942B (en) | Method for determining image classification segmentation scale threshold | |
| CN104331424B (en) | The purification method of sensitive trajectory model in a kind of user's motion track | |
| CN109785656B (en) | Traffic planning and navigation method based on local approximate rank | |
| Li et al. | Airport detection in remote sensing real-open world using deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20150429 |
|
| WD01 | Invention patent application deemed withdrawn after publication |