SLAM-OR: Simultaneous Localization, Mapping and Object Recognition Using Video Sensors Data in Open Environments from...

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Simultaneous Localization and Mapping (SLAM) is crucial for autonomous systems.
• Object recognition in real-time remains a challenge, especially with sparse data.
• Integrating SLAM with object recognition can enhance environmental understanding.

Purpose of the Study:

• To develop a novel approach for simultaneous visual localization, mapping, and object recognition.
• To enable robust performance in open environments using sparse data point clouds.
• To simplify object positioning without generating dense point clouds.

Main Methods:

• Utilizing ORB-SLAM with monocular visual frames for localization and sparse point cloud generation.
• Employing deep neural networks (e.g., YOLO, MobileNet, RetinaNet) for object detection and recognition.
• Integrating sparse point clouds with object detection results using clustering (DBSCAN, OPTICS) and PCA-based heuristics for bounding box estimation.

Main Results:

• Demonstrated a novel image processing pipeline for object positioning using sparse SLAM data.
• Achieved simplified object positioning by avoiding dense point cloud generation.
• Obtained promising results on large benchmark datasets when evaluated with various state-of-the-art object recognition architectures and clustering algorithms.

Conclusions:

• The proposed method effectively combines SLAM and object recognition for enhanced environmental perception.
• The algorithm offers a simplified and efficient solution for real-time object detection and localization.
• The open-source availability of code and datasets facilitates reproducibility and further research.