Key Points
We detect pear flower buds in sequential images taken from below and use keypoint matching between images to identify the same buds. This reduces duplicate counting when images are captured by a moving camera and enables the number of buds to be estimated.
Abstract
As a camera moves beneath pear trees and looks upward, flower buds appear in a sequence of images. These are the buds that will develop into flowers. Simply detecting the buds in each image and adding up the results can count the same bud more than once.
We first detect the flower buds in each image. We then look for matching visual points between consecutive images and use these matches to work out where the same buds appear in the next image. This reduces duplicate counting. In experiments, combining bud detection with this matching step produced smaller counting errors than using bud detection alone.
Publications
- Takumi Hanakawa, Shintaro Nakatani, Jaehwan Lee, Eiji Morimoto, Masashi Nishiyama and Yoshio Iwai,
Counting of Pear Flower Buds in Images by Judging Acquisition Conditions and Matching Keypoints, Proceedings of 43rd IEEE International Conference on Consumer Electronics (ICCE), CT02-3-2, pp. 1 - 6, January 2025. [Slides] - Takumi Hanakawa, Masashi Nishiyama, Yoshio Iwai, Eiji Morimoto, and Shintaro Nakatani,
Counting Pear Flower Buds Using Worm's-Eye View Images, Proceedings of IEEE 11th Global Conference on Consumer Electronics (GCCE), OS-AIP, pp. 544 - 547, October 2022. [Short movie] - 花川 拓海, 中谷 真太朗，森本 英嗣，西山 正志，岩井 儀雄,
梨の仰視画像の撮影条件判定を用いた花芽数カウント手法の検討,
動的画像処理実利用化ワークショップ (DIA), OS3-04(IS3-04), pp. 229 - 236, March 2024. [研究奨励賞] [Slides] [Short movie]
