object counting in high resolution remote sensing images with otb
DESCRIPTION
Object counting in high resolution remote sensing images with OTB Emmanuel Christophe; CRISP Jordi Inglada; CNESTRANSCRIPT
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion
Object counting in high resolution remotesensing images with OTB
E. Christophe1, J. Inglada2
1CENTRE FOR REMOTE IMAGING, SENSING AND PROCESSING,NATIONAL UNIVERSITY OF SINGAPORE
2CENTRE NATIONAL D’ÉTUDES SPATIALES, TOULOUSE, FRANCE
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion
Outline
Motivation
Original solutionWorkflowPan SharpeningClassificationSegmentation: Mean shiftVector data
Simplified versionsTrade-offDescriptionResults
Conclusion
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion
Motivation
Object counting
I Correspond to a wide range of problems for remotesensing data users
I Often have to be performed on large areaI Time consuming
Examples
I Houses in a city: particularly for country where urbanplanning data is not available
I Tents in refugee campI Tree stands in a field
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion
PRRS 2008 algorithm performance contest
I Time constraints: can’t spend months refining thealgorithm
I Deliver a result: whole processing chain requiredI Each step can be improvedI Goal: illustrate the on the shelf approach
ContestI Count the building from a Quickbird scene over Legaspi,
PhilippinesI XS and Pan images were provided
Aksoy et al., ”Performance evaluation of building detection and digital surface model extraction algorithms:Outcomes of the PRRS 2008 algorithm performance contest,” in 5th IAPR Workshop on Pattern Recognitionin Remote Sensing, Tampa, Florida, Dec. 2008.
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Workflow PanSharp Classif. Segment. Vector data
Outline
Motivation
Original solutionWorkflowPan SharpeningClassificationSegmentation: Mean shiftVector data
Simplified versionsTrade-offDescriptionResults
Conclusion
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Workflow PanSharp Classif. Segment. Vector data
Algorithm description
Pan
Mul
Pan sharpening
Classification
Segmentation Vectorization
Edge dectect.
Refinement Obj
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Workflow PanSharp Classif. Segment. Vector data
Algorithm description
Pan
Mul
Pan sharpening
Classification
Segmentation Vectorization
Edge dectect.
Refinement Obj
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Workflow PanSharp Classif. Segment. Vector data
Algorithm description
Pan
Mul
Pan sharpening
Classification
Segmentation Vectorization
Edge dectect.
Refinement Obj
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Workflow PanSharp Classif. Segment. Vector data
Algorithm description
Pan
Mul
Pan sharpening
Classification
Segmentation
Vectorization
Edge dectect.
Refinement Obj
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Workflow PanSharp Classif. Segment. Vector data
Algorithm description
Pan
Mul
Pan sharpening
Classification
Segmentation Vectorization
Edge dectect.
Refinement Obj
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Workflow PanSharp Classif. Segment. Vector data
Algorithm description
Pan
Mul
Pan sharpening
Classification
Segmentation Vectorization
Edge dectect.
Refinement Obj
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Workflow PanSharp Classif. Segment. Vector data
Algorithm description
Pan
Mul
Pan sharpening
Classification
Segmentation Vectorization
Edge dectect.
Refinement
Obj
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Workflow PanSharp Classif. Segment. Vector data
Algorithm description
Pan
Mul
Pan sharpening
Classification
Segmentation Vectorization
Edge dectect.
Refinement Obj
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Workflow PanSharp Classif. Segment. Vector data
Preprocessing
Pansharpening
I The pansharpening is the first step to perform to takeadvantage of the high resolution of the Panchromatic band(61 cm) with the four spectral bands of the multispectral.
Pan Mul Pan Shapening
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Workflow PanSharp Classif. Segment. Vector data
Classification
Obvious sources of errorsI There is some obvious sources of error: boat in the middle of the water which
look like houses, cars in the middle of the streetI Classification (used as a mask) can help remove these sources of error
ClassificationI Here we used a simple classification by SVMI Non classification specialists just provided a few samples per class (water,
vegetation, road, shadows, 4 colors of buildings)I Only a pixel classification: no use of texture here (that was before all the textures
were introduces in OTB)
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Workflow PanSharp Classif. Segment. Vector data
QB scene Land cover classification
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Workflow PanSharp Classif. Segment. Vector data
Segmentation: Mean shift
Too much detailsI Higher resolution is better but. . . sometimes, you would like
less details (roof superstructures, cars)I What details to remove?I Mean shift algorithm
D. Commaniciu,“Mean shift: A robust approach toward feature space analysis,”IEEE Transactions on PatternAnalysis and Machine Intelligence, vol. 24, no. 5, pp. 603–619, May 2002.
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Workflow PanSharp Classif. Segment. Vector data
Pan Sharpened Mean shift clustering
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Workflow PanSharp Classif. Segment. Vector data
Refining the boundaries
Simplification
I Easier to handle vector data than raster: vectorizationI The vectorization led to too many contour point:
simplification of the points which are roughly aligned
Fine adjustment
I Using an image contour as an input, an energy iscomputed along the polygon contour
I Introducing a random perturbation in the position of eachpoint, the energy is maximized
I Only a very basic optimization used here (ground forimprovement)
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Workflow PanSharp Classif. Segment. Vector data
Filtering
Filtering on compacity
I Buildings are usually compactI C = 4π A
L2
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Workflow PanSharp Classif. Segment. Vector data
Performances on the contest
ResultsI Two results were submitted with a difference mainly in the classificationI One result was very close with 3600 building detected for 3065 in the ground
truthI Interesting to see that most other algorithms tend to underdetection while the
proposed algorithm tends to overdetect.
Evaluation criteriaI {Correct, over, under, missed} detection and false alarm rates based on
Overlapping Area MatrixI Maximum-weight bipartite graph matchingI Normalized Hamming distanceI Clustering indices (Fowlkes-Mallows and Jaccard index)
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Workflow PanSharp Classif. Segment. Vector data
Performances on the contest
Conclusion on these resultsI Particularly hard to conclude given the wide variety of
criteria: organizer of the contest have been careful not todeclare an overall winner
I However, the proposed methods provided goodperformances (particularly on the clusterings indicescriteria) with a bias towards over segmentation.
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Trade-off Description Results
Outline
Motivation
Original solutionWorkflowPan SharpeningClassificationSegmentation: Mean shiftVector data
Simplified versionsTrade-offDescriptionResults
Conclusion
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Trade-off Description Results
Trade-off of the previous method
DrawbacksThe previous method relies on the classification of the image
I require a good understanding of the algorithm that followI influence significantly the output
Simplified version
I different trade-offs on complexity-performanceI remove the classification stepI just require the operator to click on several (2 to 5)
examples of objects
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Trade-off Description Results
Simplified version
Algorithm
I Produce a likelihood map of the region containing theobjects of interest
I Followed by the same step as the previous algorithm:segmentation, vectorization,. . .
Likelihood map: 2 choices
I Spectral angleI One class SVM
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Trade-off Description Results
Algorithm description
Pan
Mul
Pan sharpening
Spectral angle
One class SVM
Segmentation Vectorization
Edge dectect.
Refinement Obj
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion Trade-off Description Results
Results of the simplified version
Poorer preformances
I Not as good as the original one (expected)⇒ required tounderstand the algorithm
AdviceI Spectral angle: spectral characteristics of the objects are
stableI SVM: better when object have radiometric differences but
more samples are required
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion
Outline
Motivation
Original solutionWorkflowPan SharpeningClassificationSegmentation: Mean shiftVector data
Simplified versionsTrade-offDescriptionResults
Conclusion
IGARSS 2009, Cape Town
Motivation Original Simplified Conclusion
Conclusion
Modular processing chain
I An application with GUI isavailable
I It can be used for processingremote sensing images (noconstraint on the size)
I Can be easily modified andimproved as the steps aremodular and follow the pipelinephilosophy.