AI-based detection system of resident's behaviors in automatic trash sorting booths: a background computing-based solution

• Published in: Chinese Automation Congress (Online) pp. 1756 - 1760
• Main Authors: Baojun, Gong, Wei, Zhang, Zhebin, Shi, Qiucheng, Huang, Dongping, Zhang
• Format: Conference Proceeding
• Language: English
• Published: IEEE 25.11.2022
• Subjects: Adaptation models; AI-based garbage sorting; Behavioral sciences; Computational modeling; Darknet53; Object detection; Object recognition; Plastics; Streaming media; YOLOv4 model
• ISSN: 2688-0938
• DOI: 10.1109/CAC57257.2022.10055608

In order to track the serious problem of garbage siege, four-categories garbage sorting systems have been increasingly installed and employed in urban communities. However, manual supervision of trash delivery behaviors is costly, inefficient, and unsustainable from the long term. AI-based automated garbage sorting booths based on computer vision technology become an attractive solution. Under the four categories sorting system, the sorting and treatment of perishable or organic waste belong to the most challenging part. For the delivery of perishable waste in automatic garbage sorting booths, the key of AI-based solution is to accurately identify residents' abnormal delivery behavior using object detection models. The composition of organic waste is extremely complicated, making it difficult to employ the object detection model in real applications. Therefore, this paper proposes a solution to use the YOLO model to identify non-organic objects such as garbage bags in the perishable barrels. The proposed solution in this study adopts the backend computing architecture and employs the multi-scale YOLOv4 model as the core, which can effectively identify the abnormal delivery of perishable waste. This system has been applied in dozens of intelligent waste delivery booths in Yiwu City, and has achieved satisfactory results.