Improved Point-Line Feature Based Visual SLAM Method for Indoor Scenes

Area of Science:

• Robotics
• Computer Vision
• Simultaneous Localization and Mapping (SLAM)

Background:

• Traditional feature-based SLAM systems struggle with robustness during sharp camera movements.
• Point and line segment features are commonly used for camera pose estimation in indoor SLAM.

Purpose of the Study:

• To propose an improved indoor visual SLAM method utilizing both point and line segment features.
• To enhance robustness and accuracy in challenging indoor environments with rapid camera motion.

Main Methods:

• Automatic extraction and matching of point and line segment features to construct projection models.
• Incorporating minimization of angle observation alongside endpoint re-projection error for line segment optimization.
• Developing an adaptive motion estimation model with a combinational Hessian matrix and gradient vector for iterated pose estimation.

Main Results:

• The proposed method demonstrates improved localization accuracy compared to traditional methods.
• Experimental validation on EuRoC MAV datasets and custom stereo camera sequences confirmed effectiveness.
• The system shows enhanced performance in scenarios with rapid rotation and violent camera fluctuations.

Conclusions:

• The improved point-line feature-based visual SLAM method offers superior robustness and accuracy.
• This approach effectively addresses limitations of existing SLAM systems in dynamic indoor environments.
• The adaptive motion estimation and enhanced line feature optimization contribute to reliable camera pose tracking.