Reproduce “AI-Driven Relocation Tracking in Dynamic Kitchen Environments” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Arash Nasr Esfahani, Hamed Hosseini, Mehdi Tale Masouleh, Ahmad Kalhor, Hedieh Sajedi
Researchers from the Human and Robot Interaction Lab at the University of Tehran developed an AI-driven system that enables robots to track object relocation in dynamic kitchen environments. The system, leveraging YOLOv5 and a novel frame-scoring algorithm, achieved 97.72% accuracy in detecting relocated objects by comparing 'best-associated frames' across different scene states.
Key insights
- A visibility scoring algorithm combining depth, bounding box size, centrality, and detection confidence allows for robust identification of an object's 'best-associated frame' across different scenes.
- By comparing best-associated frames from pre-change and post-change scenes, the system can accurately determine if an object has relocated, even with temporary occlusions or varying viewpoints.
- Excluding distant objects during object detection training improved recall and mAP50 by focusing the model on more discriminative features of closer objects.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.