IMU based animal activity recognition has been widely used in farm. When it comes to wildlife monitoring, there are new challenges in terms of data annotations. It’s costly to annotate the daily activities of the animals in wildlife, to make the labelled dataset large enough for training an effective deep learning model. If we can synthesis more labelled IMU data from more available data source such as videos, it will solve the problem of lacking labelled data and improve the activity recognition performances.
How to design an IMU generator with end-to-end deep neural networks to generate IMU data from videos.
The project will be divided into 2 main tasks:
- Developing a end-to-end regression model that can predict IMU series given videos,
- Train a classification model with generated IMU data for animal activity classification
30% Theory, 50% Implementation, 20%Writing
Le Viet Duc, firstname.lastname@example.org, room ZI 5013