This paper presents a synthetic data generation pipeline using probabilistic scene composition and transfer learning to address class imbalance in rare action recognition within visual datasets for autonomous systems.

Matthew Alan Donnelly, Chen Yiran, Kavitha Suresh Chandra, Otto Karl Reinhardt, Hideo Masaki Fujimoto, Emily Jade Monroe

This research introduces a predictive maintenance framework that analyzes high-frequency image streams using anomaly detection models, identifying early visual signs of mechanical faults in industrial equipment to reduce downtime.

Jonathan Blake Hammond, Liang Wenxuan, Deepa Narayanan Iyer, Felix Noah Berger, Zainab Fatima Sadiq, Han Jialin

This study proposes a collaborative SLAM solution where multiple robots share visual observations via edge fusion techniques, enhancing localization accuracy and resilience in highly dynamic indoor environments.

Christian Paul Donovan, Mei Xueying, Anuradha Sree Raghavan, Francois Xavier Marchal, Carla Denise Gomez, Zhao Tiancheng

This paper introduces a transformer-based semantic segmentation approach tailored for satellite imagery, utilizing multi-scale contextual decoding to accurately delineate urban and agricultural features across high-resolution remote sensing data.

Gregory Michael Huxley, Tang Weihao, Sneha Priya Venkataraman, Alice Caroline Foster, Peng Xiaoru, Jamal Ibrahim Al-Fayez

This paper explores a few-shot learning framework incorporating meta-learning and hierarchical feature generalization to enable accurate urban object detection with minimal annotated data in complex cityscapes.

Arthur Nathaniel Granger, Xu Mingyu, Leena Sudha Prabhakaran, Stefan Rudolf Mertz, Natalia Veronika Kuznetsova, Kim Hyeonwoo

This study introduces a hybrid CNN-RNN attention-based framework for real-time emotion recognition in driver monitoring systems, enhancing safety by detecting stress, fatigue, or distraction through facial expression analysis.

Harrison George MacLeod, Zhao Liancheng, Meenal Kalyani Deshmukh, Ethan Robert Holloway, Aurora Celeste Franco, Wang Jinhai

This research presents a zero-shot VQA framework for surveillance videos using cross-modal conceptual graphs, enabling intelligent systems to answer questions about unseen actions, objects, and contexts without additional training data.

Logan Peter Whitaker, Lin Chunyan, Radhika Anantha Krishnan, Julien Bernard Lefevre, Fatima Noor Al-Farsi, Zhang Mingcheng