Dynamic thresholding algorithm for robotic apple detection

Elie Zemmour, Polina Kurtser, Yael Edan

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

7 Scopus citations

Abstract

This paper presents a dynamic thresholding algorithm for robotic apple detection. The algorithm enables robust detection in highly variable lighting conditions. The image is dynamically split into variable sized regions, where each region has approximately homogeneous lighting conditions. Nine thresholds were selected so as to accommodate three different illumination levels for three different dimensions in the natural difference index (NDI) space by quantifying the required relation between true positive rate and false positive rate. This rate can change along the robotic harvesting process, aiming to decrease FPR from far views (to minimize cycle times) and to increase TPR from close views (to increase grasping accuracy). Analyses were conducted on apple images acquired in outdoor conditions. The algorithm improved previously reported results and achieved 91.14% true positive rate (TPR) with 3.05% false positive rate (FPR) using the NDI first dimension and a noise removal process.

Original languageEnglish
Title of host publication2017 IEEE International Conference on Autonomous Robot Systems and Competitions, ICARSC 2017
EditorsLino Marques, Alexandre Bernardino
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages240-246
Number of pages7
ISBN (Electronic)9781509062331
DOIs
StatePublished - 29 Jun 2017
Event2017 IEEE International Conference on Autonomous Robot Systems and Competitions, ICARSC 2017 - Coimbra, Portugal
Duration: 26 Apr 201728 Apr 2017

Publication series

Name2017 IEEE International Conference on Autonomous Robot Systems and Competitions, ICARSC 2017

Conference

Conference2017 IEEE International Conference on Autonomous Robot Systems and Competitions, ICARSC 2017
Country/TerritoryPortugal
CityCoimbra
Period26/04/1728/04/17

Keywords

  • Apples detection
  • Dynamic thresholding
  • Object detection
  • Robotic harvesting

Fingerprint

Dive into the research topics of 'Dynamic thresholding algorithm for robotic apple detection'. Together they form a unique fingerprint.

Cite this