Exjobbsförslag från företag

Detta är ett uppsatsförslag hämtat från Nationella Exjobb-poolen. Klicka här för att komma tillbaka till samtliga exjobbsförslag.

Förslaget inkom 2011-09-20

Automatic Seal Identification in Aerial Images from the Arctic

Arctic Sea has been diminishing in extent over the past twenty-five years. The quality and type of sea ice has been changing as well. Currently, there is a great deal of uncertainty about how these changes are effecting biological activity in the Arctic.

One important question is how are these changes are affecting seal populations in the Arctic. Some species of seals require free floating ice for breeding. Flights with manned and unmanned aircraft are currently taking a large number of images (over a hundred thousand images are currently a part of this project) of the ice which are being analyzed for seal populations. These images, if properly analyzed will be able to help address questions such as whether the seal population is stable, increasing or decreasing. Currently, there is enough concern about seal populations to have two species being considered for Endangered Species status. Human analysis of images is highly inaccurate and inefficient. There is a strong need for software development to answer these critical and important questions about how our planet is changing.

To investigate the possibility of developing automated methods for detecting and identifying seals in such aerial images, we are announcing a MSc thesis project on this topic. The project will consist of

• developing suitable feature detection methods for detecting candidate regions in the aerial images where seals may be present,
• applying suitable feature description/recognition methods to judge if the detected feature corresponds to a seal or should be regarded as false positive,
• evaluating the performance of the solution on benchmark data sets with aerial images from the Arctic.

There may also be a possibility of developing a semi-automatic solution, where automatic image analysis methods are used for pre-screening the image data sets to generate smaller subsets that can be inspected by humans.

A suitable background for addressing this MSc project is knowledge in computer vision and image processing, preferably complemented by knowledge in machine learning accompanied by good skills in computer programming.

The project will be carried out at KTH in close collaboration with environmental researchers at University of Colorado.

Contact: Tony Lindeberg ([email protected])


Informationen om uppsatsförslag är hämtad från Nationella Exjobb-poolen.