Ana İçerik

Lisans Öğrencilerimizin Konferans Bildirisi Başarısı

Abstract:

While autonomous, cashier-less retail environments significantly reduce checkout queues and improve customer experience, existing commercial solutions like Amazon Go rely on expensive sensor networks that remain cost-prohibitive for small and medium-sized businesses. To address this financial barrier, this paper presents an end-to-end, low-cost smart market prototype integrating computer vision, embedded edge sensing, and an asynchronous multi-modal sensor fusion framework. The proposed architecture leverages Keypoint R-CNN for precise wrist-level human pose estimation to filter out volatile body movements, alongside StrongSORT with a specialized position-only gating mechanism to maintain persistent customer identities in crowded scenarios. On the hardware layer, decentralized ESP32based wireless nodes monitor shelf weight changes. To overcome the temporal mismatch between varying visual frame rates and network-delayed weight telemetry, a history-based matching algorithm is introduced to asynchronously associate stable mass displacements with recent visual hand-shelf interactions. Supported by a complete software ecosystem comprising an SQLite database, Flask backend, and Streamlit dashboard, experimental results demonstrate that the proposed system operates reliably under multi-user interactions and severe occlusions using only commercial off-the-shelf hardware.

 

Bildirinin tamamına buradan bu linkten ulaşabilirsiniz: 10.1109/ASYU71790.2026.11772188

 

 

Öğrencimizi tebrik eder, başarılarının devamını dileriz.

HIZLI ERİŞİM