Aduviso, Arabella Mae M.

Video highlight generator utilizing a convolutional neural network (cnn) combined with a long short-term memory (lstm) network for mobile legends: bang bang/ Arabella Mae M. Aduviso, Luis Pocholo Caducio, Juliana Anne M. Capoquian, Simone Arabella B. Caturla, and Jieson R. Orain.-- - Manila: Technological University of the Philippines, 2025. - x, 124pages: 29cm.

Bachelor's thesis

College of Science.--

Includes bibliographic references and index.

To respond to the increase in demand for efficient content creation in Esports, this
study presents a Video Highlight Generator for Mobile Legends: Bang Bang using two
deep learning architectures. The system merges a Convolutional Neural Network (CNN)
and a Recurrent Neural Network (RNN) specifically a Long Short-Term Memory Network
(LSTM) into a modified Long-term Recurrent Convolutional Network (LRCN) to detect
temporal gameplay patterns such as team fights, and YOLOv8 for real-time in-game banner
detection with an 80% confidence threshold. Detected banners are assigned excitement
scores to prioritize highlight-worthy moments, considering only events lasting at least three
seconds to ensure relevance. The system compiles these into a reviewable highlight reel,
allowing manual review and arrangement based on priority or chronology. Developed and
trained in Google Colaboratory (Colab) and deployed in a Python-based frontend, the
system was evaluated across multiple hardware setups and demonstrated high acceptability
in the product quality aspect of the ISO 25010. The study validated the effectiveness of the
hybrid deep learning model’s integration in generating automated and customizable
gameplay highlights, providing a significant tool for Esports players, teams, and content
creators.


Video highlight generation
Deep learning architectures
Convolutional neural networks (cnn)

BTH T 58.5 / A38 2025