Tree Image Dataset for Biophysical Parameter Estimation using Stereoscopic Vision

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本数据集包含251组立体图像对，拍摄自肯尼亚基安布郡基耶尼森林的补植林区，内容为完整树木；另有90组立体图像对，拍摄自肯尼亚涅里县德丹·基马蒂理工大学的休闲公园，内容为树干。此外，还提供了两组校准图像对，分别包含30组与29组图像，以供有需要的用户自行开展相机校准工作。本数据集可用于研究利用立体视觉技术提取树木生物物理参数的相关方法。 本数据集以单个Zip压缩包形式上传，解压后将得到5个文件夹与3个CSV文件。所有完整树木与树干图像的文件名均遵循[CAPTURE_TYPE][SEQUENCE_ID]_[IMAGE_ID]-[STEREO_ID]的命名规范。其中，CAPTURE_TYPE可取值为FT或T，分别代表“完整树木（full tree）”与“树干（trunk）”；SEQUENCE_ID表示图像在拍摄序列中的位置；IMAGE_ID包含图像拍摄时刻的时间戳；STEREO_ID可取值为LEFT或RIGHT，分别对应左视图图像与右视图图像。 以下为本数据集内文件夹与文件的详细说明： 1. full_trees文件夹：包含251组立体图像对（共502张图像），拍摄自基耶尼森林的野外数据采集场景，内容为完整树木；另有251张对应的分割掩码图像，总计753张图像。这些树木的冠幅直径（crown diameters, CD）与树高（tree heights, TH）的真值标注存储于"full_trees.csv"文件中。 2. trunks文件夹：包含90组立体图像对（共180张图像），拍摄自德丹·基马蒂理工大学的野外数据采集场景，内容为树干；另有90张对应的分割掩码图像，总计270张图像。这些树干的胸径（diameters at breast height, DBH）的真值标注存储于"trunks.csv"文件中。 3. Calib-v1文件夹：包含29组校准图像对，可从中提取相机校准参数，该参数仅适用于full_trees*文件夹内的图像。 4. Calib-v2文件夹：包含30组校准图像对，可从中提取相机校准参数，该参数仅适用于trunks文件夹内的图像。 5. "full_trees.csv"文件：存储了基耶尼森林中251棵树木的冠幅直径与树高真值标注。full_trees文件夹内图像中的每棵树木，其冠幅直径与树高数值均可在该文件中查询。 6. "trunks.csv"文件：存储了德丹·基马蒂理工大学中90棵树干的胸径真值标注。trunks文件夹内图像中的每棵树干，其胸径数值均可在该文件中查询。 在采集本数据集内的图像之前，已使用知名的OpenCV库完成相机校准。若需了解本文档与YAML文件中各相机参数的含义，请参考OpenCV官方文档：https://docs.opencv.org/3.4.15/d9/d0c/group__calib3d.html