RGB-D fusion models for construction and demolition waste detection

Area of Science:

• Environmental Science
• Computer Vision
• Robotics

Background:

• Urbanization generates substantial construction and demolition waste (CDW), leading to land occupation and ecological damage.
• Effective recycling of CDW is essential to mitigate its negative environmental impact.
• Current waste classification methods relying on color images lack the accuracy needed for efficient management.

Purpose of the Study:

• To develop an advanced waste classification system for construction and demolition waste (CDW).
• To enhance the accuracy and efficiency of CDW sorting through improved feature fusion techniques.
• To evaluate the performance of novel RGB-depth fusion models for real-time waste detection.

Main Methods:

• An RGB-depth (RGB-D) detection platform was constructed using a color camera and a laser line-scanning sensor.
• Three feature fusion models (RGB-D concat, RGB-D Ci-add, RGB-D Ci-concat) were proposed, based on the Mask R-CNN instance segmentation network.
• The models were evaluated for their ability to classify and segment CDW objects using combined RGB and depth image data.

Main Results:

• The proposed RGB-D fusion models, particularly RGB-D Ci-add and RGB-D Ci-concat, demonstrated superior performance compared to RGB-only models.
• Mean Average Precisions (mAPs) increased by 1.33%–1.72%, and classification accuracy improved by 1.92%–2.27%.
• All developed models met real-time online detection requirements, indicating practical applicability.

Conclusions:

• The RGB-D Ci-concat model exhibits excellent comprehensive performance, making it suitable as a final detection model for robotic sorting of CDW.
• The integration of depth information significantly enhances the accuracy of CDW classification.
• The developed platform and models offer a promising solution for improving the efficiency of construction and demolition waste recycling.