A salient region detection model combining background distribution measure for indoor robots

Area of Science:

• Computer Vision
• Robotics
• Artificial Intelligence

Background:

• Saliency detection is crucial for enhancing indoor robot visual perception.
• Existing saliency detection methods perform poorly in complex indoor environments.
• Robust saliency detection is needed for reliable indoor robot navigation.

Purpose of the Study:

• To develop an effective saliency detection method for indoor mobile robots.
• To improve the performance of visual perception systems in complicated indoor settings.
• To address the limitations of current saliency detection techniques in non-natural image environments.

Main Methods:

• A novel method combining graph-based RGB-D segmentation, primary saliency, and background distribution measures.
• Introduction of region roundness for robust background distribution measurement.
• Validation against eleven existing methods on DSD and ECSSD datasets, and real-world mobile robot experiments.

Main Results:

• The proposed method demonstrates superior performance compared to eleven state-of-the-art saliency detection techniques.
• Experimental results on a mobile robot platform confirm the model's effectiveness in varied indoor conditions.
• The method shows significant improvements in identifying salient regions within complex indoor scenes.

Conclusions:

• The developed saliency detection method is highly effective for indoor mobile robot applications.
• The approach offers a robust solution for visual perception challenges in complicated indoor environments.
• This work contributes to advancing autonomous navigation and interaction capabilities of indoor robots.