A New Method for Classifying Scenes for Simultaneous Localization and Mapping Using the Boundary Object Function...

Area of Science:

• Computer Vision
• Robotics
• Artificial Intelligence

Background:

• Autonomous navigation systems face challenges in scene classification due to environmental variations and computational demands on small devices.
• Traditional methods like SIFT and ORB are computationally intensive, limiting their application on resource-constrained platforms.
• Accurate scene understanding is crucial for effective simultaneous localization and mapping (SLAM).

Purpose of the Study:

• To develop a computationally efficient scene classification method for SLAM systems.
• To reduce the complexity of scene classification algorithms for small-factor computers.
• To maintain high accuracy while improving processing speed in autonomous navigation.

Main Methods:

• A novel scene classification method utilizing the boundary object function (BOF) descriptor on RGB-D point clouds.
• Combining BOF-based descriptors from scene objects to define the scene class.
• Extracting object boundaries from layered RGB-D data and classifying features using a support vector machine (SVM) with a bag-of-visual-words model.

Main Results:

• The proposed BOF-based method achieves comparable accuracy to SIFT-based algorithms for scene classification.
• The method demonstrates a significant speed improvement, being 2.38 times faster than SIFT-based approaches.
• Experimental validation on the 7-Scenes and SUNRGBD datasets confirms the method's effectiveness, accuracy, and robustness.

Conclusions:

• The BOF descriptor offers an efficient alternative for scene classification in SLAM, suitable for resource-limited autonomous systems.
• The proposed method effectively balances accuracy and computational cost, addressing key challenges in autonomous navigation.
• This approach provides a robust solution for real-time scene understanding in robotic applications.