Evolution and challenges of computer vision and deep learning technologies for analysing mixed construction and demolition waste

Adrian Langley,Matthew Lonergan,Tao Huang,Mostafa Rahimi Azghadi
2024-09-20
Abstract:Improving the automatic and timely recognition of construction and demolition waste (C&DW) composition is crucial for enhancing business returns, economic outcomes, and sustainability. Technologies like computer vision, artificial intelligence (AI), robotics, and internet of things (IoT) are increasingly integrated into waste processing to achieve these goals. While deep learning (DL) models show promise in recognising homogeneous C&DW piles, few studies assess their performance with mixed, highly contaminated material in commercial settings. Drawing on extensive experience at a C&DW materials recovery facility (MRF) in Sydney, Australia, we explore the challenges and opportunities in developing an advanced automated mixed C&DW management system. We begin with an overview of the evolution of waste management in the construction industry, highlighting its environmental, economic, and societal impacts. We review various C&DW analysis techniques, concluding that DL-based visual methods are the optimal solution. Additionally, we examine the progression of sensor and camera technologies for C&DW analysis as well as the evolution of DL algorithms focused on object detection and material segmentation. We also discuss C&DW datasets, their curation, and innovative methods for their creation. Finally, we share insights on C&DW visual analysis, addressing technical and commercial challenges, research trends, and future directions for mixed C&DW analysis. This paper aims to improve the efficiency of C&DW management by providing valuable insights for ongoing and future research and development efforts in this critical sector.
Computer Vision and Pattern Recognition,Artificial Intelligence
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to improve the automatic and timely identification of construction and demolition waste (C&DW) components, especially when dealing with mixed and highly contaminated materials in a commercial environment. Specifically, the paper focuses on using technologies such as computer vision, artificial intelligence (AI), robotics, and the Internet of Things (IoT) to develop an advanced automated mixed C&DW management system. Although deep - learning (DL) models have shown potential in identifying homogeneous C&DW piles, few studies have evaluated the performance of these models in dealing with mixed materials in a commercial environment. Therefore, the focus of the paper is to explore the challenges and opportunities in this area and provide valuable insights to support continuous research and development work in this field. The specific objectives of the paper include: 1. **Review the development process of C&DW management**: Introduce the environmental, economic, and social impacts of C&DW management in the construction industry. 2. **Evaluate existing C&DW analysis techniques**: Discuss the application of non - deep - learning methods in C&DW assessment and reporting. 3. **The evolution of sensor and camera technologies**: Review the development of various sensor and camera technologies used for C&DW analysis. 4. **Advances in deep - learning algorithms**: Outline the deep - learning techniques used for C&DW analysis, especially the advances in object detection and material segmentation. 5. **Creation and management of C&DW data sets**: Discuss the collection, creation, and innovative methods of C&DW data sets. 6. **Technical challenges and future directions**: Explore the challenges faced by deep - learning technologies in C&DW management and potential future development directions. Through these objectives, the paper aims to support the improvement of C&DW management efficiency, thereby improving business returns, economic outcomes, and sustainability.