Canopy Lidar point cloud data K-means clustering Watershed segmentation method
This is a modal window.
The media could not be loaded, either because the server or network failed or because the format is not supported.
Formal Metadata
| Title |
| |
| Title of Series | ||
| Number of Parts | 39 | |
| Author | ||
| License | CC Attribution 3.0 Unported: You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal purpose as long as the work is attributed to the author in the manner specified by the author or licensor. | |
| Identifiers | 10.5446/52931 (DOI) | |
| Publisher | ||
| Release Date | ||
| Language |
Content Metadata
| Subject Area | |
| Genre |
36
00:00
InformationDISMAPoint cloudPoint (geometry)MathematicsGene clusterStudent's t-testGoodness of fitPoint cloudComputer animation
00:20
Message passing
00:33
Physical systemForestRegular graphBasis <Mathematik>TheoryNetwork topologyRegular graphData managementGame controllerImperative programmingMereologySoftware developerNetwork topologyBasis <Mathematik>TheoryInformationPhysical systemForestSingle-precision floating-point formatSound effect
01:22
Digital signalReal numberLocal ringComputer-generated imageryTransformation (genetics)BildsegmentierungMaxima and minimaGradientAlgorithmGamma functionData modelSurfaceWater vaporDirected graphProcess (computing)Similarity (geometry)GradientLine (geometry)Medical imagingDigitizingWater vaporFigurate numberCustomer relationship managementPole (complex analysis)Error messageState of matterBuffer overflowSurfaceAlgorithmMaxima and minimaDiagramNetwork topologyPartition (number theory)Point (geometry)Real numberProcess (computing)Capability Maturity ModelMereologyPulse (signal processing)Endliche ModelltheorieStructural loadDiagram
03:54
WindowMaxima and minimaCapability Maturity ModelVariable (mathematics)Point (geometry)AlgorithmDigital filterPoint cloudWindowEndliche ModelltheorieExecution unitProcess (computing)Maxima and minimaGradientOperator (mathematics)Medical imagingSingle-precision floating-point formatAverageNetwork topologyDiameterGraphical user interfaceNonlinear systemVariable (mathematics)Neighbourhood (graph theory)Figurate numberCloningComputer animation
04:57
Maxima and minimaData modelPixelNetwork topologyBildsegmentierungAlgorithmBasis <Mathematik>Control flowProcess (computing)Block (periodic table)Computer-generated imageryCalculationVarianceVariable (mathematics)Network topologyPoint (geometry)Endliche ModelltheorieMaxima and minimaMathematicsGene clusterPixelStandard deviationZustandsgrößeHazard (2005 film)Medical imagingCone penetration testAreaDirection (geometry)WindowBlock (periodic table)VarianceSocial classSimilarity (geometry)Local ringNeighbourhood (graph theory)Group actionArithmetic meanIterationEnvelope (mathematics)Ring (mathematics)Film editingComputer animation
06:55
Point cloudDiagramPoint (geometry)Point (geometry)CodeNumberCircleIntegrated development environmentMereologyRadiusComputer animation
07:15
Shooting methodArtificial neural networkSound effectWeb pageRadio-frequency identification
07:35
Asynchronous Transfer ModeCapability Maturity ModelData modelWindowVariable (mathematics)Pattern recognitionPoint (geometry)Mixed realityResultantNetwork topologyWell-formed formulaLocal ringEndliche ModelltheorieMaxima and minimaRaw image formatComputer animation
08:07
AlgorithmPairwise comparisonData modelCapability Maturity ModelElectronic visual displayScale (map)Capability Maturity ModelEndliche ModelltheorieScaling (geometry)AlgorithmCustomer relationship managementWater vaporElectronic visual displayGene clusterComputer animation
08:58
Electronic visual displayAreaNetwork topologyAlgorithmResultantSingle-precision floating-point formatNetwork topologyAlgorithmWater vaporElectronic visual displayNumberComputer animation
09:20
AlgorithmMathematical analysisParameter (computer programming)Well-formed formulaAlgorithmWell-formed formulaResultantNetwork topologyNumberComputer animation
Transcript: English(auto-generated)
00:00
Good morning, ladies and gentlemen. I'm a student from Guilin University of Technology. The title of my paper is Canopy Lidar Point Cloud Date, Chemin's Clustering Watershed Segmentation Method.
00:23
I will introduce my research from the following four parts, background, method, experiment, conclusion. The first part is the background. On the one hand, due to Chinese regular forest
00:41
resource inventory system, there is an urgent need for accurate and effective control and management of forest information. Airborne Lidar technology has unique advantages over passive remote sensing technology,
01:03
providing theoretical basis and technical support for the accurate extraction of single tree canopy and the development of precision forestry.
01:24
On the other hand, real digital images often shows a lot of local minima in their gradient images, which brings a lot of water pulse. This filter becomes a watershed over segmentation.
01:47
The second part is method. This is a schematic diagram of the watershed algorithm regarding the gray image as a landform surface.
02:01
A hole is punched at each local minimum on the image. In this picture, we can see, water will slowly emirates into the surface from this hole, starting from the loins point.
02:23
And finally, the overflow model will reach a state where only the top of the depth is exposed. The tops of these depths are water dividers and the water accumulation errors
02:41
that raise from different local minima are called sinks or pores. We can see figure A shows a CMM to simulate the processing of loading. We first calculated the complement of the CMM
03:01
as figure B, which resemble two catchment basins. Assume that each basins has a hole punched at its minimum. Then we immerse it gradually into water. The catchment basins will be flooded.
03:23
This algorithm can be solved to automatically build the depth along the divide line to prevent water into neighboring catchment basins from merging as figure C. The constructed depths are called water lines,
03:43
water shined lines, and will be used partition trees. The two pictures below are made from experimental data. The experiment process of this article is shown in the figure.
04:02
First, predict the canopy size window. This experiment will find all the 9-liter relationship between tree height and the tree clone and get the variable window size. TC is the clone diameter of each individual tree.
04:25
TH is the average tree height of individual single tree. Then the CMM model is generated. In this paper, the CHM image is reconstructed based on the own operation.
04:42
And the value of each grid unit is replaced with the maximum value of the lesser point cloud in the neighborhood to obtain the maximum canopy model. The highest point of the local pixel maximum direction
05:03
combined with the cone maximum model, it takes as the highest point of the tree. And then each local highest point is used as the six point of K-means clustering for clustering. Step one, the first clustering center
05:23
represents the pixels as a treetop located by the variable window. Step two, for each point, find the closest clustering center according to a given grid ring and assign it to the area of this tab.
05:42
Step three, the new clustering center obtained in step two. Step four, when the cluster center no longer changes, the iteration is terminated. Watershine segmentation based on K-means clustering.
06:03
Step one, suppose the image is divided into envelopes after K-means segmentation. Step two, define the grid variance of two adjacent blocks.
06:21
Step three, define the mean value of standard deviation of grid value of four neighborhood pixels on the edge of the segmentation block. Step four, define the similarity between two adjacent cut blocks.
06:40
Step five, set a zero-shoot value t initially to maximize the variance between classes. Step five, use t to drive through the adjacent divided blocks. The third part is experiment.
07:00
In this paper, the radius filter is used to denote the point cloud. A circle is drawn around a point to calculate the number of points for all size of the circle. This paper adopts a semi-automatic pit removal method
07:22
designed to artificially set the experiment's dry-shoot and rarely judges the best pit removal effect. Experimental results show that when using CMM model
07:43
to monitor the tree top based on the local mix mark value, the pseudo-local mix mark value can be significantly eliminated. And the formula of drawn and missing points
08:01
in the tree top recognition can be reduced. The K-means watershed algorithm proposed in this article is compared with the traditional watershed algorithm.
08:20
We can see the picture. Picture A is CMM model. The picture B is grayscale displaced CMM model. So picture C is watershed algorithm. The picture D is K-means clustering based on CMM.
08:41
So picture E is grayscale displaced CMM, K-means clustering. The picture F is improved watershed algorithm.
09:01
Finally, a single-wood segmentation result display based on K-means clustering watershed segmentation algorithm is formed. The number at the top of the tree represents the tree ID.
09:24
So far is the conclusion. It can be analyzed by the formula that the number of unsplathed trees and the wrong splathed trees are reduced. Therefore, the segmentation result
09:41
of the improved watershed algorithm is more accurate. Thank you for much.