Background

Ecologists exploit remotely activated camera equipped with a motion sensor for detecting wildlife  in a non-invasive fashion. This technique is known as camera trapping. This technique is particularly useful for detecting nocturnal or rare animals. However, processing the huge amount of camera trapping data is a significant challenge as the ecologists currently process the data manually. This involves work ranging from weeks to months after the data acquisition for proper analysis.

Thesis Objective

The vision behind this thesis is to design a deep learning framework that can significantly speed-up the overall pipeline. Two major goals of this particular thesis is to come up with individual species identification and duplicate individual identification. Identifying species in images is an extremely important pre-processing step for the ecologists because this enables them to concentrate on the species of interest for further processing. Identifying duplicate individual is a more challenging yet important task. Challenging because ecologists rely on qualitative understanding of the species for this task and important because this gives an indication of the abundance of the species within a certain geographic region.

Pre-requisites

1. Experience in Python programming
                - Experience with GPU or Deep learning would be a huge bonus

2. Sincerity to the research problem and systematic approach towards problem solving

We Offer

 Insight into the state-of-the-art of Deep Learning based Computer Vision in a stimulating research environment. We offer co-supervision for the thesis and close collaboration within the research group over the entire duration of the project. 

Design and Implementation of a Crowdsourced Study on Computer-aided Diagnosis
MA 01/17