Evaluation of Vision Transformers for the detection of fullness of garbage bins for efficient waste management

Parakram Singh Tanwer¹SHISHIR MAHESHWARI²Sushree Behera¹Amit Chauhan³Sunil Kumar Telagamsetti^4*

• ¹International Institute of Information Technology, Bangalore, Bangalore, Karnataka, India
• ²Motilal Nehru National Institute of Technology Allahabad, Allahabad, Uttar Pradesh, India
• ³Thapar Institute of Engineering & Technology, Patiala, Punjab, India
• ⁴University of Gävle, Gävle, Sweden

Efficient waste management is crucial for urban environments to maintain cleanliness, reduce environmental impact, and optimize resource allocation. Traditional waste collection systems often rely on scheduled pickups or manual inspections, leading to inefficient resource utilization and potential overflow issues. This paper presents a novel approach to automate the detection of garbage container fullness from images using machine learning techniques. More specifically, we explore three transformer-based architectures, namely, vision transformer, Swin transformer, and pyramid vision transformer to classify input images of garbage bins as clean or dirty. Our experimental results on the publicly available Clean dirty containers in Montevideo dataset suggest that transformer-based architectures are effective in garbage fullness detection. Moreover, a comparison with existing methods reveals that the proposed approach using the vision transformer surpasses the state-of-the-art, achieving a 96.74% accuracy in detecting garbage container fullness.