[Nov 24th 2021] LabMeeting: Trucks as sensors: classification and mapping of abandoned waste by Deep Learning and embedded mobile computing

Barbara Toniella Corradini (University of Siena)

When: Nov 24th, 2021 – 11:00 – 11:45 AM
Where: Google meet link

Nowadays, all the activities relating to the collection and management of waste play a fundamental role in improving living conditions of citizens. Nonetheless, there are obstacles which make it difficult to set up an effective waste collection process, e.g., the ever increasing extension of urban areas: among the remarkable consequences of the above, one is the phenomenon of waste abandoned on the ground. The present work had as its main purpose to design and engineer an on-board smart computing system for garbage trucks, with the aim of automating the detection and positional mapping of waste abandoned on the ground. First, the hardware of the system was assembled: we will show what kind of difficulties arose and what constraints we had to face during the engineering phase. In a second stage, software for waste detection and mapping was implemented. It exploits YOLO architecture (i.e., the state of the art in DNN object detection and recognition) for the detection of abandoned waste on images taken from a camera module and tagged with the corresponding GPS positions, so that it is always possible to associate images and their coordinates. Thus, it is possible at glance for the operator to print the whole route of the truck and its points of interest on a map, modify the collection plan and reassign the task of abandoned waste collection to the appropriate waste collection unit. Results provided by the system are encouraging: the hardware showed high reliability, and very good results were obtained by the software, in terms of detection accuracy and speed.

Work developed in collaboration with Alia servizi ambientali S.p.A., the company which is responsible for waste collection in the central area of Tuscany.

 |  Category: Seminars