Description »

The LyF RescueDrone project, led by Paul Campredon and Valentin Ibars, aims to create a device able to precisely and autonomously localize victims buried in avalanches, in a faster and more efficient manner than current commercialized solutions. The main objective of this semester project is to create a classification algorithm to segment the region depicting an avalanche, and specifically detect the avalanche boundaries, in an image acquired by a drone.

More specifically, a team of two students will work on the task of discriminating the different textures composing an avalanche path using machine learning based classification algorithms. In particular, they will create a dictionary of different classes based on a data set of training images and compare patches of test images to this dictionary to identify the avalanche boundaries.

The different steps of this project will involve:

• Segment the example images into the “true” classes manually, i.e., snow with sun, snow in shadow, mountain/rock, sky, clouds, avalanche (close) and avalanche (distant)
• Make a training set for each class recording important image features for each patch (e.g., mean color and variance)
• Learn one dictionary for each class
• Segment a complete test image by making test patches centered at each pixel of the test image and comparing them to the dictionaries

More advances machine learning techniques will also be considered.

Back to the project list.