XXII ISPRS Congress 2012: Technical Commission VIII

Shortis, M.; Shimoda, H.; Cho, K.

Bibliographic data

Multivolume work

Persistent identifier:: 1663813779

Title:: XXII ISPRS Congress 2012

Sub title:: Melbourne, Australia, 25 August-1 September 2012

Year of publication:: 2013

Place of publication:: Red Hook, NY

Publisher of the original:: Curran Associates, Inc.

Identifier (digital):: 1663813779

Language:: English

Additional Notes:: Kongress-Thema: Imaging a sustainable future

Corporations:: International Society for Photogrammetry and Remote Sensing, Congress, 22., 2012, Melbourne; International Society for Photogrammetry and Remote Sensing

Adapter:: International Society for Photogrammetry and Remote Sensing, Congress, 22., 2012, Melbourne; International Society for Photogrammetry and Remote Sensing

Founder of work:: International Society for Photogrammetry and Remote Sensing, Congress, 22., 2012, Melbourne; International Society for Photogrammetry and Remote Sensing

Other corporate:: International Society for Photogrammetry and Remote Sensing, Congress, 22., 2012, Melbourne; International Society for Photogrammetry and Remote Sensing

Document type:: Multivolume work

Volume

Persistent identifier:: 1663822514

Title:: Technical Commission VIII

Scope:: 590 Seiten

Year of publication:: 2014

Place of publication:: Red Hook, NY

Publisher of the original:: Curran Associates, Inc.

Identifier (digital):: 1663822514

Illustration:: Illustrationen, Diagramme

Signature of the source:: ZS 312(39,B8)

Language:: English

Additional Notes:: Erscheinungsdatum des Originals ist ermittelt.; Literaturangaben

Usage licence:: Attribution 4.0 International (CC BY 4.0)

Editor:: Shortis, M.; Shimoda, H.; Cho, K.

Corporations:: International Society for Photogrammetry and Remote Sensing, Congress, 22., 2012, Melbourne; International Society for Photogrammetry and Remote Sensing

Adapter:: International Society for Photogrammetry and Remote Sensing, Congress, 22., 2012, Melbourne; International Society for Photogrammetry and Remote Sensing

Founder of work:: International Society for Photogrammetry and Remote Sensing, Congress, 22., 2012, Melbourne; International Society for Photogrammetry and Remote Sensing

Other corporate:: International Society for Photogrammetry and Remote Sensing, Congress, 22., 2012, Melbourne; International Society for Photogrammetry and Remote Sensing

Publisher of the digital copy:: Technische Informationsbibliothek Hannover

Place of publication of the digital copy:: Hannover

Year of publication of the original:: 2019

Document type:: Volume

Collection:: Earth sciences

Chapter

Title:: [VIII/6: Agriculture, Ecosystems and Bio-Diversity]

Document type:: Multivolume work

Structure type:: Chapter

Chapter

Title:: ESTIMATION OF VEGETATION HEIGHT THROUGH SATELLITE IMAGE TEXTURE ANALYSIS Z. I. Petrou, C. Tarantino, M. Adamo, P. Blonda, M. Petrou

Document type:: Multivolume work

Structure type:: Chapter

220 2 Theory Let us consider two subsets of IR 3 , Siaser and the DSM, repre senting both of them the same topographic landscape, but ex pressed into two different frames. Registering Si ase r and the DSM consists in retrieving the unknown n parameter transform Mth that maps one geometry into the other. The registration problem is dual in case of a global deformation: First finding correspondences (tying features) followed by the estimation of a global transform M (a model of M t h) minimizing a cost func tion F{M) = Y.d(M{X i ),Y i ) (1) i where Xi £ <5n aser and Yi £ DSM are the i th homologous fea ture, d is a distance function. Tying features may be of different nature. As mentioned in the introducing part, point correspon dences between fidar and DSM are difficult to calculate. We will therefore search for surface patch correspondences without any limitation on plane or linear feature extraction. We will search for correspondences by regularly paving laser strips. 2.1 Determination of patch correspondences Considering the registration problem of a laser strip with regard to a DSM, we will suppose that M t h is regular enough to be ap proximated with piecewise shifts which represent the local offset between both point clouds. The calculation of these local shifts provides homologous patches of points which can be represented as homologous centroids for convenience during the global esti mation process. Let us consider adjacent square regions R that pave a laser strip, and the set Lr of laser points included in R. R= [zi, x 2 \ x [yi, y 2 ] £ M 2 Lr {lk {Xk i Vk j ) /jg [0,/f] ^ ^laser/ {XkiDk) C R} Vi k (equation 2) is a neighborhood of DSM points centered onto the planimetrie coordinates of a laser point l k . Note that the neighborhood’s shape does not have any influence on the process ing. ={P: = (*i>yi,*i) i6N € DSM/ max(\xk - Xj\,\y k - yj\) < C} (2) where C is a constant. Let Tl r be the unknown approximation of Mth (local shift) onto a surface patch Lr that we want to retrieve. Each point lk € Lr will have a nearest homologous point (n.h.p) in Vi k (provided that Vi k be wide enough) through a translation tk (tk is reached when pjlk has the nearest orientation of Tl r ) satisfy ing: tk = arg min \\pjlk A Tl r || VZ fc £ Lr, pj £ V lk (3) where A denotes the vector product of both vectors, pjlk (a po tential shift candidate) is the vector between the extracted DSM nodes Vi k and the laser point l k , t k and Tl r are unknown. Since Tl r is supposed to be unique over the surface patch Lr, Tl r is the translation for which vectors tk are similar for all laser points lk in Lr. Equation 3 cannot be solved directly. Each point belonging to Vi k is a potential n.h.p of l k . The most represented potential n.h.p. may be seen as the maximum of the distribution d v ofV = {PiMvi fc €£ K ,v Pj -€Vi • Tl r is therefore defined as: Tl r = arg max dr (2f) xer 2.2 Estimation process The point matching part of the algorithm provides piecewise shift approximations of Mth■ In order to estimate its analytical rep resentation, we will consider both the initial point cloud and the piecewise corrected one by the above calculated Tl r ■ The idea is to apply a continuous transform to the whole strip so that the final point cloud should be continuously corrected. We applied a 12 parameter affinity (A\T a ) where A is any 3x3 matrix and T a a 3D translation. (A\T a ) is estimated using a least power es timation process. This estimator belongs to the family of robust M-estimators. Unlike the standard least-square method that tries to minimize ]T\ r\ where n is the difference between the i th ob servation di and its fitted value H A / n mi, the M-estimators try to reduce the effect of outliers by replacing the squared residu als rf by another function of the residuals yielding to minimize p(ri) where p is a symmetric, positive function with a unique maximum at zero, and is chosen to be less increasing than square. Following Xu and Zhang (Xu and Zhang, 1996), for regression problems, the best choice is the L p function which consists of minimizing IIdi — H^/rmi\\ p i with p = 1.2. This optimization is implemented as an iterative re-weighted least power algorithm. In a robust cost model, noth ing special needs to be done with outliers. They are just normal measurements that happen to be down-weighted owing to their large deviation. 2.3 Global Correction The hypothetical time dependency is modelled with the estima tion of a set of transforms (here, affinities) along the flight track through a sliding window of constant width w (strip width) and of tunable length L (see figure 1). L is defined to be linearly pro portional to w (L = aw, a £ R + *). This window evolves with a defined moving step k > 0. We prefer to define an overlapping ratio C = 1 — j; with C > 0.5 so that a majority of laser point should be processed at least twice. Depending on k, a laser point will be processed E[j] times where E is the integer part func tion. The final corrected point (&#[ A] M) will be their weighted mean value, which is motivated by the almost zero standard de viation of the independently processed 3D points. The weighting function W is defined as a Gaussian function depending on both C and on the distance d £ [0, |-] between the laser point M and a fine defined by a normal vector of the flight track direction and the barycenter of the sliding window. We have W{d) = e fc =e 3 Theoretical experiments 3.1 Description Before applying the algorithm onto raw laser data, several sim ulations have been performed. This simulation aims to decide whether or not the algorithm is able to retrieve a global motion through the detection of local 3D offsets as well as to evaluate its Figur dow times W(d] abili appi resp late alent gles then that The lies The High roll 3.2 We algc Her rouj in rate ian

Access restriction

Copyright

fullscreen: From sensor to imagery (Part B)

Multivolume work

Volume

Chapter

Chapter

Table of contents

Cite and reuse

Volume

Chapter

Image

Image fragment

Citation links

Citation recommendation