Direct monocular vision algorithm based on deep constraints of point and line features fusion

Area of Science:

• Computer Vision
• Robotics
• Artificial Intelligence

Background:

• Simultaneous Localization and Mapping (SLAM) is widely used in smart devices due to technological advancements.
• Direct method-based visual odometry struggles with accurate pose estimation in structured environments.
• Existing methods often overlook crucial line segment information and feature constraints.

Purpose of the Study:

• To enhance pose estimation accuracy in structured environments for direct method-based visual odometry.
• To propose a novel direct monocular vision algorithm incorporating deep constraints of point and line features (DMVA-PLF).

Main Methods:

• Integrated environmental line features using colinear and deep constraints.
• Combined line features with historical pose information for optimized pose estimation.
• Developed a direct monocular vision algorithm (DMVA-PLF).

Main Results:

• The DMVA-PLF algorithm significantly improved pose estimation accuracy.
• Experimental results showed superior performance compared to traditional methods in structured environments.
• Efficiently leveraged image features for enhanced accuracy.

Conclusions:

• The proposed DMVA-PLF algorithm effectively addresses limitations of traditional direct methods in structured environments.
• This approach offers a more robust solution for visual odometry and SLAM applications.
• Deep constraints of point and line features are crucial for accurate pose estimation.