A Comprehensive Hybrid Approach for Indoor Scene Recognition Combining CNNs and Text-Based Features

Area of Science:

• Computer Vision
• Artificial Intelligence
• Robotics

Background:

• Indoor scene recognition is crucial for robotics, security, and assistive technologies.
• Convolutional Neural Networks (CNNs) excel at outdoor scenes but struggle with indoor environments due to reliance on global features.
• Indoor scenes require recognition of local features like furniture and objects.

Purpose of the Study:

• To improve indoor scene recognition accuracy.
• To develop a hybrid model integrating image and text processing for indoor environments.
• To leverage object recognition outputs for enhanced contextual information.

Main Methods:

• Utilized the "MIT 67 Indoor Scene" dataset.
• Extracted and combined features from a Convolutional Neural Network (CNN) and a text-based model.
• Developed a two-channel hybrid model integrating image and natural language processing techniques.

Main Results:

• The hybrid model achieved an 8.3% improvement in test accuracy compared to the image processing model alone.
• Demonstrated a notable success rate in recognizing diverse indoor environments.
• Successfully integrated natural language processing and image processing for enhanced performance.

Conclusions:

• The proposed hybrid model significantly enhances indoor scene recognition.
• Integrating text-based object recognition with image processing offers a robust approach for indoor environments.
• This research contributes to advancements in indoor scene understanding, mapping, and remote sensing applications.