This study proposes a human activity forecasting system using transformer-based temporal encoding, enabling assistive robots to anticipate human intent and improve interaction fluidity in healthcare and home settings.

Marcus David Ellison, Zhang Xueqi, Roshni Anusha Kannan, Louis Bernard Lefevre, Ayaka Hoshino, Cristina Beatriz Morales

This paper presents a cross-modal fusion method that integrates sonar signals with visual data using attention encoding, significantly improving object detection accuracy in murky underwater environments.

Victor Isaac Hamilton, Liu Xingmei, Pranita Sandeep Joshi, Baptiste Claude Morel, Haruka Naomi Sugimoto, Eliana Sofia Delgado

This study introduces a visual localization framework incorporating feature aging and scene adaptation to maintain robustness under long-term changes in urban environments such as construction and seasonal variation.

Jonathan Ellis Whitford, Zhang Huiqin, Sneha Lalitha Narayanan, Jean-Noel Olivier Beauchamp, Yui Aiko Fujita, Mariana Isabel Costa

This paper proposes an energy-efficient object detection framework for IoT surveillance using quantized lightweight CNNs and smart wake-up triggers, reducing power consumption without compromising recognition accuracy.

Bradley Owen Armstrong, Zhao Mingliang, Nivedita Pradeep Shah, Mathieu Alain Dubois, Hiroko Kazumi Nakamura, Lucia Sofia Romero

This work introduces a gesture-based control interface for smart vehicles using depth-aware hand tracking and multi-modal sensor fusion, enhancing driver interaction and safety through contactless input recognition.

Colin Patrick Saunders, Zhang Zihan, Aditi Meenal Rathi, Antoine Julien Morel, Emi Haruka Taniguchi, Isadora Camila Freitas

This study proposes a self-supervised pretraining method using contrastive enhancement networks, improving object recognition performance in low-light visual data for autonomous systems and surveillance applications.

Mitchell Ryan Dawson, Zhang Yutong, Kavitha Srinidhi Ramaswamy, Leo François Bernard, Aiko Fumiko Sato, Celina Mariana Torres

This research introduces a scene-adaptive instance segmentation model using spatial memory networks and hierarchical feature aggregation, enabling precise segmentation of overlapping objects in highly crowded environments.

Nathaniel George Prescott, Zhao Wenjuan, Ishita Gaurav Menon, Pierre Antoine Delacroix, Haruna Emi Takeda, Fernanda Isabella Ruiz