Computer vision & 3D measurements

Cameras are becoming omnipresent and collect a lot of information. To fully benefit from this, the useful information has to be extracted automatically from the images, since it is infeasible to do it manually. This is the area of Computer Vision.

In the Image Analysis and Computer Graphics section at DTU Compute conducts research in computer vision with special emphasis on extraction of 3D information.

Homotopy based 3D reconstruction of water columns from 2D cross section acoustic data


The objective of this research is 3D reconstitution and rendering of the water column from the water surface till the sea floor, and its contents (flora and fauna) in between 2D sections acoustic data. This research will produce the construction of a 3D oceanographic view of the water column and the sea floor by reconstituting the space in between the 2D sections with a focus on preserving topology of all the contents of the water column.

Gain analysis using a Time-Of-Flight camera and a treadmill


We present a system that analyzes human gait using a treadmill and a Time-of-flight camera. The camera provides spatial data with local intensity measures of the scene, and data are collected over several gait cycles. These data are then used to model and analyze the gait. For each frame the spatial data and the intensity image are used to fit an articulated model to the data using a Markov random field. To solve occlusion issues the model movement is smoothened providing the missing data for the occluded parts. The created model is then cut into cycles, which are matched and through Fourier fitting a cyclic model is created. The output data are: Speed, Cadence, Step length and Range of motion. The described output parameters are computed with no user interaction using a setup with no requirements to neither background nor subject clothing. Contact Rasmus Ramsbøl Jensen

Structure from Motion


The topic of this project is structure from motion, i.e. the inference of 3D structure of a usually rigid object from normal images hereof. The primary focus is on:

1. The statistical issues regarding this highly popular problem, i.e. robust statistics and the including of prior knowledge

2. The extension to dealing with non-rigid objects

3. The estimation of how the objects surface, once a crude structure and camera motion have been estimated

Contact Henrik Aanæs

Change Detection


Detecting change over time in bi- or multi-temporal multivariate spatial data is a challenging and important task in many fields of application. Many methods apply. When dealing with bi-temporal optical data we work with methods that are insensitive to scaling (actually invariant to affine transformations) within the two sets of variables. This is a huge advantage since normalization and other corrections that are linear over time are not needed.

Polarimetric radar data presents special problems when compared to optical data. This can be dealt with by applying the complex Wishart distribution to so-called multi-look radar data. Present work in the area deals with regularization methods to avoid problems that occur when the number of observations is small compared to the number of variables. This may be the case for hyperspectral data. Also, regularization may be desirable when the data are highly correlated as with spectral or spatial data. Read more about this here... Contact Allan Aasbjerg Nielsen

Estimation of Velocity Fields in Meteosat Image Sequences


The estimation of flow fields from time sequences of satellite imagery has a number of important applications. For visualization of cloud or sea ice movements in sequences of crude temporal sampling a satisfactory non blurred temporal interpolation can be performed only when the flow field or an estimate there-of is known.

Estimated flow fields in weather satellite imagery might also be used on an operational basis as inputs to short-term weather prediction. Local measurements of motion are obtained by analysis of the local energy distribution, which is sampled using a set of separable 3-D spatio-temporal Gabor filters evenly distributed of all spatio-temporal directions.

The estimated local energy distribution also allows us to compute a certainty measure of the estimated local flow. We use Markovian random fields in order to incorporate smoothness across the field. To obtain smoothness we constrain first as well as second order derivatives of the flow field. Read more about this here... Contact: Rasmus Larsen

Enhancement and Analysis of Ocean Water Variability


Satellite altimetry has successfully been used to monitor the ocean surface and has provided valuable information about the dynamics of the worlds oceans and the marine gravity field. Other Earth observing sensors onboard satellites have provided an enormous amount of information about the sea surface temperature and the ocean colour.

Through interdisciplinary projects methods are developed for integrating multi mission, multi sensor and multi channel satellite data for improved determination and analysis of the sea level. Read more about this here... Contact Allan Aasbjerg Nielsen, Klaus Baggesen Hilger

Numerical Weather Predictions for GPS Positioning


During transmission through the atmosphere the GPS satellite signals are affected by the media. In the neutral atmosphere the refraction causes a signal delay, which is a function of the meteorological conditions along the signal path.

Numerical weather predictions (NWP) are predictions of the three dimensional meteorological conditions for a given area and point in time, and can as such be used for predicting the delay for a satellite signal by integration along the signal path through the NWP. This estimate of the signal delay can be used in the positioning process to account for the error caused by the delay, and the goal of the research is thus to obtain improvements in positioning accuracy and reliability for high accuracy GPS positioning. Read more about this here...

Restoration of Polarimetric EMISAR Data 


A core aspect of this project is investigation and development of statistical methods and models for optimal use of polarimetric EMISAR data. Contextual information in an image is embedded not only in the individual pixels but also in the relative position of neighbouring pixel values.

The potential of utilizing this relative position to explain the structure underlying the EMISAR data is explored. The restorations are carried out using Markov Random Fields in a Bayesian framework. The work is a part of the multidisciplinary project named DANMAC (DANish Multisensor Airborne Campaign).

The purpose of the DANMAC project has been to achieve a better understanding of the physical conditions and processes at or near the surface and their influence on the signals registered by radar and optical, remote sensing sensors.

The use of Polarimetric EMISAR for the Mapping and Characterization of the Semi-natural Environment


In the recent years the use of Earth-observing satellites has become increasingly important in the monitoring of our planet. Synthetic Aperture Radar (SAR) is a technology for collecting image data, based upon active microwave remote sensing. Polarimetric SAR, represents some of the most sophisticated and up-to-date developments in SAR remote sensing, providing wide scope for research and application development work.

The general topic of this project is the application of polarimetric EMISAR data for mapping and characterization of semi-natural ecosystems. The work is a part of the multidisciplinary project named DANMAC (DANish Multisensor Airborne Campaign). The purpose of the DANMAC project has been to achieve a better understanding of the physical conditions and processes at or near the surface and their influence on the signals registered by radar and optical, remote sensing sensors.


Henrik Aanæs
Associate Professor
DTU Compute