Tree Image Dataset for Biophysical Parameter Estimation using Stereoscopic Vision

Published: 5 July 2023| Version 4 | DOI: 10.17632/nx3ggv7pxf.4
Contributors:
Cedric Kiplimo,
, Billy Okal

Description

This dataset contains 251 stereoscopic image pairs of full trees obtained from a replanted section of the Kieni Forest in Kiambu County, Kenya and 90 stereoscopic image pairs of tree trunks obtained from recreation parks at the Dedan Kimathi University of Technology, Nyeri, Kenya. In addition, two sets of 30 and 29 calibration image pairs have also been provided for the benefit of those who prefer to perform camera calibration on their own. This dataset can be used to study the use of stereo vision for extracting tree biophysical parameters of trees. The dataset is uploaded as a single Zip file, which upon extracting unfolds into five folders and three CSV files. All filenames for images of full trees and trunks follow the convention [CAPTURE_TYPE][SEQUENCE_ID]_[IMAGE_ID]-[STEREO_ID]. CAPTURE_TYPE can be FT or T representing 'full tree' and 'trunk' respectively. SEQUENCE_ID refers to the position of the image in the capturing sequence. IMAGE_ID contains the time stamp for the time the image was captured. STEREO_ID can be LEFT or RIGHT depending on whether it is the left or right image. Below is a description of the folders and files in the dataset: 1. full_trees: Contains 251 stereoscopic image pairs (502 images) of full trees captured during field data collection at Kieni Forest, and 251 corresponding segmentation masks (to make a total of 753 images). The ground truth values for the crown diameters (CD) and the tree heights (TH) of these trees are contained in the 'full_trees.csv' file. 2. trunks: Contains 90 stereoscopic image pairs (180 images) of tree trunks captured during field data collection at the Dedan Kimathi University of Technology, and 90 corresponding segmentation masks (to make a total of 270 images). The ground truth values for the diameters at breast height (DBH) of these trunks are contained in the 'trunks.csv' file. 3. Calib-v1 (folder): Contains 29 calibration image pairs from which calibration parameters can be obtained. These parameters should be applied only to the images in the 'full_trees*' folders. 4. Calib-v2 (folder): Contains 30 calibration image pairs from which calibration parameters can be obtained. These parameters should be applied only to the images in the 'trunks' folder. 5. full_trees.csv: Contains the ground truth values for the CD and TH of the 251 trees measured at Kieni Forest. Each tree in the images inside the 'full_trees' folder has its CD and TH value in this file. 6. trunks.csv: Contains the ground truth values for the DBH of the 90 trees measured at the Dedan Kimathi University of Technology. Each tree in the images inside the 'trunks' folder has its DBH value in this file. Before capturing the images in this dataset, camera calibration was performed using the well-known OpenCV library. For the meaning of the individual camera parameters in the text and YAML files, please refer to this article by OpenCV: https://docs.opencv.org/3.4.15/d9/d0c/group__calib3d.html

Files

Steps to reproduce

1. Build a stereo camera 2. Write python code for capturing left and right image pairs (Use the OpenCV library 3. Capture the images 4. Use an image labeling tool to create the masks

Institutions

Dedan Kimathi University of Technology School of Engineering

Categories

Binocular Vision, Stereopsis, Forest Inventory

Licence