This study proposes an edge computing framework optimized for real-time visual inference in autonomous delivery robots, minimizing latency and bandwidth constraints while maintaining high accuracy in dynamic urban landscapes.

Julian Ernest McAllister, Cheng Huiyuan, Veena Priyadarshini Rao, Matteo Rinaldo Esposito, Omar Ghassan Mahmoud, Xavier Louis Fontaine

We introduce a hierarchical vision transformer architecture that fuses multi-camera poses to enhance large-scale object tracking accuracy in outdoor environments, improving spatial consistency and reducing occlusion errors.

Lena Margot Keller, Wei Guozhen, Rajan Vikas Narayanan, Sophie Juliette Lemoine, Tanaka Riku, Jakob Henrik Petersen

This paper addresses occlusion challenges in video-based perception systems by proposing a temporal consistency-aware video synthesis method that restores missing views using generative adversarial networks and sequential context analysis.

Henrik Gustav Lang, Bao Jiahao, Arun Mohan Prasad, Ingrid Helena Torres, Natalia Beatrice Caruso, Farid Abdul Wahid

This study presents a fine-grained action recognition system using two-stream temporal convolution and semantic keypoint tracking to classify nuanced sports actions from professional gameplay footage.

Mason Oliver Thurston, Gao Linfeng, Priyanka Santosh Joshi, Esteban Lucas Molina, Nathaniel Bruce Holloway, Zhang Ruiwen

This work proposes a federated learning framework for vision-based anomaly detection across smart campuses, preserving privacy while enabling collaborative training using distributed institutional surveillance data.

Juliette Marie Langdon, Li Weihao, Abdul Rashid Mir, Jean-Claude François Moreau, Samantha Rose Nichols, Zhang Yuanlin