University of Twente Student Theses

Login

Does Prior Knowledge Affect CNN and LSTM Models on Classifying Improper Sitting Postures?

Valtas, Georgios (2023) Does Prior Knowledge Affect CNN and LSTM Models on Classifying Improper Sitting Postures?

This is the latest version of this item.

[img] PDF
1MB
Abstract:Posture can be defined as the physiological position held by a body when it is maintaining a stationary stance such as sitting or standing. Holding a correct posture implies that the individual maintains a stance in a position that does not exert excess strain on the body and maintains the straight and natural curve of the spine [1]. Lack of having a proper posture can impose a negative impact over extended periods of time such as in the office space. In this study, 3 types of Neural Network models were evaluated and tested on images of individuals holding 3 different postures: leaning to the left, to the right and sitting straight. The 3 models: A simple CNN model was compared to 2 models with transfer learning, a CNN and an LSTM model, with the intention of measuring the importance of prior knowledge in models. Results showed that with a dataset of 2751 images, prior knowledge did not affect the models 99.7% vs 99.7% (with prior knowledge) yet with a smaller dataset of 540 images the model with prior knowledge performed better by 13.3% better. While determinants such as the background and types of clothing can affect the models, a variation of the cosine pose similarity equation was found to aid the models’ accuracy.
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/96093
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page