Indoor Scene Recognition via Object Detection and TF-IDF.

Area of Science:

• Computer Vision
• Robotics
• Artificial Intelligence

Background:

• Indoor scene recognition is crucial for social robots and understanding environments.
• Recent advances incorporate object-level information for improved performance.
• Current methods often rely on holistic scene features.

Purpose of the Study:

• To demonstrate that indoor scene recognition can be achieved solely using object-level information.
• To integrate computer vision (object detection) and natural language processing (TF-IDF) techniques.
• To explore applications in embodied AI and dynamic scene classification.

Main Methods:

• A state-of-the-art object detection model (YOLO) was trained on indoor objects.
• The trained model detected objects within scene data.
• Detected objects were utilized as features for room category prediction using TF-IDF.

Main Results:

• Successful indoor scene recognition using only detected objects.
• Demonstrated the efficacy of combining YOLO for object detection and TF-IDF for feature representation.
• Achieved robust room category prediction based on object composition.

Conclusions:

• Object-level information is sufficient for accurate indoor scene recognition.
• The hybrid approach offers a novel method for scene understanding.
• This research has implications for embodied AI and real-time scene analysis.