Optimized Violence Detection in Video Sequences Using Inception-ResNetV2 and Transfer Learning with Redundant Frame Reduction
KIET Journal of Computing and Information Sciences · 01.02.2026 · 8 (2)
DOI: 10.51153/kjcis.v82.266 ↗Kurz gefasst
Fachartikel in der Zeitschrift KIET Journal of Computing and Information Sciences, online erschienen am 01.02.2026. Verfasst von Sajid Ullah Khan. Die Arbeit nennt 59 Literaturangaben.
Abstract (Originalsprache)
Over recent years, violence detection has emerged as a crucial task in computer vision, driving advancements in innovative methodologies. This research employs transfer learning to identify violent scenes, remove redundant frames, and classify video frames as violent or non-violent. Using the Hockey Fight Dataset and Real-Life Violence Situation Dataset, the proposed framework integrates Inception v3, pre-trained on ImageNet, for feature extraction. A heuristic method reduces redundant frames, while Euclidean similarity enhances frame selection. The dataset is divided into 70% training and 30% testing using hold-out cross-validation. The model's performance is evaluated against five deep learning architectures: 3D-CNN, LSTM, I3D, ResNet50, and DenseNet121. Inception-ResNetV2 outperforms all models, achieving 96% accuracy on the Hockey Fight Dataset and 93% on the Real-Life Violence Dataset with Adam, and 97% and 91% with SGDM, respectively, with corresponding F1-scores of 97% and 91%. The 3D-CNN, LSTM, I3D, and DenseNet121 models show lower performance, with accuracies ranging from 64% to 91%. These findings underscore the superior capability of Inception-ResNetV2 for violence detection in video sequences.
Zitieren
APA
Khan, S. U. (2026). Optimized Violence Detection in Video Sequences Using Inception-ResNetV2 and Transfer Learning with Redundant Frame Reduction. KIET Journal of Computing and Information Sciences, 8(2). https://doi.org/10.51153/kjcis.v82.266
Vancouver
Khan SU. Optimized Violence Detection in Video Sequences Using Inception-ResNetV2 and Transfer Learning with Redundant Frame Reduction. KIET Journal of Computing and Information Sciences. 2026;8(2). doi:10.51153/kjcis.v82.266
BibTeX
@article{Khan2026,
author = {Khan, Sajid Ullah},
title = {Optimized Violence Detection in Video Sequences Using Inception-ResNetV2 and Transfer Learning with Redundant Frame Reduction},
journal = {KIET Journal of Computing and Information Sciences},
year = {2026},
volume = {8},
number = {2},
doi = {10.51153/kjcis.v82.266}
}
EUVIMED