University of Twente Student Theses

Login

Generating IMU data from video using deep neural networks for use in animal activity recognition

Bovenkerk, Jasper (2022) Generating IMU data from video using deep neural networks for use in animal activity recognition.

[img] PDF
3MB
Abstract:Inertial measurement unit (IMU) data has proven to be quite successful in the field of activity recognition, both on human and animal activities. This IMU data can be gathered relatively easily from animals using collars with sensors. However, for training accurate models a large amount of this data needs to be labeled, which is a very expensive and time-consuming process. To overcome the issue, researchers have come up with several ways to generate IMU data from video, as labeled video data is abundantly avail�able. Previous approaches mainly make use of pose estimation and forward kinematics. In this paper, however, the viability of using end-to-end deep learning for generating IMU data is evaluated. This research consists of two parts, the first part will be to use end-to-end deep learning to generate IMU data from video data. The second part will be to train an animal activity recognition(AAR) model to evaluate the effects of adding generated IMU data to the training data of the AAR model. In this research is shown that, albeit with a fairly small dataset with a limited amount of activities, there are indications that IMU data generated from video using neural networks can contribute to the training of an AAR model.
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science BSc (56964)
Link to this item:https://purl.utwente.nl/essays/92024
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page