Fallen Tree Detection and Free Urban Space RELEAF

The vegetation of urban streets and neighbourhoods is often impacted after storms with fallen trees while nearby vacant areas currently have no known use for replantings. In this paper, we describe a web based AI system which allows users (i.e., citizens) to automatically identify fallen trees from their camera images, geolocate the areas and provide a realistic set of nearby planting sites or locations where there is space enough for a new tree. The device developed in this study uses an object detector model (YOLOv8) to quickly confirm that a tree has recently fallen, using photos taken at street level. We utilise a simple semantic segmentation and rule engine to find suitable planting strips and parcels using satellite or map images. The project is implemented using a cloud supported application with a React frontend, firebase authentication and firestore to allow synchronous real time data exchange between users and authorities. Our implementation provides a dashboard where red markers represent hazardous areas and green markers represent potential planting spaces, allowing staff to manage incidents from report to clearance and replacement tree planting. Our evaluation of the platform using curated image sets and pilot testing with users indicates that the system significantly reduces the amount of verification required by individuals, as well as provides, at the same time, practical site specific assistance for greening urban areas.