At a time when cities and towns are losing their green cover to unplanned urbanisation, researchers at the Indian Institute of Information Technology (IIIT) in Hyderabad have come up with a new method to scientifically count trees and generate density maps to denote the same. A ride across the city with a basic camera to record videos is all that is required to generate a tree count and density map within minutes.
The computer-vision-aided tool for object detection uses a machine-learning algorithm that can automatically detect and count the number of trees abutting roads. The result is a colour-coded map of the entire city to explain the extent of tree cover on specific routes.
Arpit Bahety, 23, a research fellow at the Center for Visual Information Technology lab in IIIT-H, tested the tool by capturing footage on a GoPro camera mounted on a helmet tied to the rear seat of his Honda Activa. The same was done in Surat too, using a Samsung mobile phone camera.
A Machine Learning (ML) model was developed where trees can be detected at once. This can also avoid duplication, which is common when many trees are enumerated manually as is the practice.
This shows the density of trees on a road, and how it translates to the map. (Source: Arpit Bahety/IIITH)
“Trees can be detected with a camera mounted on any vehicle, but it has to be facing the left. The vehicle needs to move closer to the left side of the road. Based on the trunk of the tree, the camera will detect it. When the video is fed to the system after processing, we can generate a map that gives a big picture of the entire city. We know which areas have more trees or fewer trees, etc,†he told IndianExpress.com.
More than 50 trees per kilometer would mean a very good tree count, while less than 20 trees per km is a low tree count. On the map generated by the system, dark green represents a very good tree count, and areas with the lowest tree count are denoted in black.
The system, developed by the team of researchers, obtained an accuracy of 83 per cent for the streets of Hyderabad and Surat.
A GoPro camera is mounted on a helmet tied to the rear seat of his Honda Activa. (Source: Arpit Bahety/IIITH)
Bahety worked under the guidance of Prof C V Jawahar in VCIT labs for the project and credits the support of Dr Rohit Saluja, Dr Anbumani Subramanian, and also Prof Ravi Kiran Sarvadevabhatla who were part of the team.
Going ahead, the team hopes to test their tree-counting system in different cities to understand its efficiency at a large scale.
The research idea stemmed from a request from the National Institute of Urban Management to generate an AI-enabled solution to facilitate city planning and urban afforestation. The team believes this tool could help assess efforts of afforestation in urban areas.
