Date of Award


Publication Type


Degree Name



Computer Science


dataset;Human activity recognition;LRCN;musculoskeletal disorders;RULA


Boubakeur Boufama



Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.


Human activity has been closely related to the development of musculoskeletal disorders. Workers in industries such as manufacturing, assembly, and construction often engage in repetitive and various motions, such as lifting heavy objects or per- forming the same task for a long period. Because such activities can increase the risk of various muscular disorders, it is important to help workers choose the best postures when performing their activities. Human activity recognition plays an important role in recognizing the type of task workers are performing and Ergonomics is an important field that aims at designing tools and work environments that promote workers’ safety and health. Video-based Human Activity Recognition and Rapid Upper Limb Assessment (RULA) method can be used to assess ergonomic risks for workers, by evaluating the physical stresses imposed on the body during work activities. In this thesis, we propose a deep learning approach for task recognition and RULA score calculation. Our approach uses a revised version of the Long Term Recurrent Convolutional Neural network-based model to classify work activities based on video input and then applies a separate neural network to estimate the RULA score for each input activity. We trained and evaluated our approach using a dataset of annotated work activities. Our results show that our approach achieves competitive accuracy for activity recognition and RULA score estimation, demonstrating the potential of deep learning for improving ergonomic assessments in the workplace.

Available for download on Wednesday, February 12, 2025