Geocoding of trees from street addresses and street-level images

Abstract

We introduce an approach for updating older tree inventories with geographic coordinates using street-level panorama images and a global optimization framework for tree instance matching. Geolocations of trees in inventories until the early 2000s where recorded using street addresses whereas newer inventories use GPS. Our method retrofits older inventories with geographic coordinates to allow connecting them with newer inventories to facilitate long-term studies on tree mortality etc. What makes this problem challenging is the different number of trees per street address, the heterogeneous appearance of different tree instances in the images, ambiguous tree positions if viewed from multiple images and occlusions. To solve this assignment problem, we (i) detect trees in Google street-view panoramas using deep learning, (ii) combine multi-view detections per tree into a single representation, (iii) and match detected trees with given trees per street address with a global optimization approach. Experiments for > 50000 trees in 5 cities in California, USA, show that we are able to assign geographic coordinates to 38 % of the street trees, which is a good starting point for long-term studies on the ecosystem services value of street trees at large scale.

Assessing climate risk to support urban forests in a changing climate

“…The use of big data methods based on remote sensing and integration with urban ecosystem accounting should be explored (e.g. Laumer et al, 2020;Hanssen et al, 2021 (Campbell et al, 2016). Therefore, future transdisciplinary vulnerability research and assessments on urban forest resilience to climate change will need to take into account more socio-ecological perspectives and approaches (Steenberg et al, 2017).…”

Remotely Sensed Tree Characterization in Urban Areas: A Review

“…Therefore, to complete a vegetation analysis with GLI, the images must be captured along sections that allow for analyzing the whole scene in order to detect and characterize urban trees [109]. Thus, for tree characterization and species identification, shorter section distances between consecutive images, e.g., 10-15 m with DGI [15,45,47,104,105] or 30 m with PPC or MLS are required [102] (Table 5).…”

“…This study highlights the importance of data sources such as GSV and OpenStreetCam for updating information on urban forests and their ability to offer ES. On the other hand, studies, such as the one developed by Laumer et al [105], advanced not only in tree recognition (91% of accuracy), but also in tree positioning, achieving 56% accuracy in the assignment of geographic coordinates' through GSV images and the Geocoding API from Google. Moreover, Branson et al [45] focused on tree species identification, in addition to the detection of individual trees, from GLI and achieved an accuracy of 70% in tree segmentation and an accuracy of 80% in species identification of urban trees of Pasadena, California.…”

Weakly Supervised Instance Attention for Multisource Fine-Grained Object Recognition with an Application to Tree Species Classification

“…The feature representations computed by deep networks independently trained for the aerial and ground views were concatenated and fed to a linear SVM for classification of 40 tree species. More recently, Laumer et al (2020) improved the existing street tree inventories where the individual trees were referenced by only street addresses with accurate geographic coordinates that were estimated from multi-view detections in street-view panoramas. The methods proposed in this paper are not specific to tree detection.…”