An Improved Human-Body-Segmentation Algorithm with Attention-Based Feature Fusion and a Refined Stereo-Matching...

Area of Science:

• Computer Vision
• Biometrics
• Machine Learning

Background:

• Accurate human body segmentation and anthropometric measurements are crucial for various applications.
• Existing systems often face challenges in achieving high precision and efficiency.

Purpose of the Study:

• To propose an improved human-body-segmentation algorithm and a refined stereo-matching scheme for anthropometric systems.
• To enhance the accuracy, efficiency, and precision of anthropometric measurements.

Main Methods:

• Incorporated Channel-Attention: Multiple Bidirectional Convolutional Attention Modules (CBAMs) into the ResNet101 backbone of PSPNet for improved spatial and channel feature fusion.
• Optimized network efficiency by replacing common convolutions with group convolutions in ResNet101 residual blocks.
• Developed a corner-based feature point design for sub-pixel level coordinate extraction and applied regional constraints for complexity reduction in stereo matching.

Main Results:

• The proposed CBAM-based segmentation and corner-based stereo-matching significantly outperformed state-of-the-art systems in accuracy.
• The system demonstrated improved efficiency due to reduced model parameters and computational cost.
• Achieved compliance with national standards (GB/T 2664-2017, GA 258-2009, GB/T 2665-2017) and textile industry standards (FZ/T 73029-2019, FZ/T 73017-2014, FZ/T 73059-2017, FZ/T 73022-2019).

Conclusions:

• The novel approach significantly enhances the performance of anthropometric systems.
• The integration of attention mechanisms and refined stereo matching offers a robust solution for accurate human body measurement.
• The system's compliance with multiple industry standards validates its practical applicability.