Walking Human Detection Using Stereo Camera Based on Feature Classification Algorithm of Second Re-projection Error.

Area of Science:

• Computer Vision
• Robotics
• Environmental Sensing

Background:

• Accurate feature classification is crucial for autonomous systems operating in dynamic environments.
• Existing methods struggle with distinguishing static and dynamic targets, especially large moving objects.
• Vision sensors offer a cost-effective solution but require robust algorithms for complex scenarios.

Purpose of the Study:

• To develop a vision-based feature classification method for dynamic environments.
• To improve the accuracy of distinguishing static and dynamic objects under varying conditions.
• To reduce the impact of large-area moving targets on feature classification.

Main Methods:

• A novel double-projection error metric utilizing ORB (Oriented FAST and Rotated BRIEF) and SURF (Speeded Up Robust Features) was proposed.
• Texture and region constraints were integrated to enhance classification accuracy.
• The method was designed to operate solely using visual sensor data, optimizing inter-frame relationships.

Main Results:

• The proposed method achieved accurate feature classification across four distinct environmental conditions.
• It effectively mitigated the influence of large-area moving targets on classification performance.
• The algorithm demonstrated superiority over other methods in both static and dynamic environments.

Conclusions:

• The developed vision-based feature classification method is effective in dynamic environments.
• The integration of double-projection error with texture and region constraints enhances robustness.
• This approach offers a promising solution for real-time object recognition and tracking in complex, moving scenes.