prikaz prve stranice dokumenta Detekcija grana za rezidbu na RGB-D slikama
  Download
PDF 9.73 MB
master's thesis
Detekcija grana za rezidbu na RGB-D slikama
Osijek: Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, 2024. urn:nbn:hr:200:609181
Bošnjak, Andrej
Josip Juraj Strossmayer University of Osijek
Faculty of Electrical Engineering, Computer Science and Information Technology Osijek
Department of Computer Engineering and Automation
Chair of Automation and Robotics

Cite this document

Bošnjak, A. (2024). Detekcija grana za rezidbu na RGB-D slikama (Master's thesis). Osijek: Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek. Retrieved from https://urn.nsk.hr/urn:nbn:hr:200:609181

Bošnjak, Andrej. "Detekcija grana za rezidbu na RGB-D slikama." Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, 2024. https://urn.nsk.hr/urn:nbn:hr:200:609181

Bošnjak, Andrej. "Detekcija grana za rezidbu na RGB-D slikama." Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, 2024. https://urn.nsk.hr/urn:nbn:hr:200:609181

Bošnjak, A. (2024). 'Detekcija grana za rezidbu na RGB-D slikama', Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, accessed 24 March 2025, https://urn.nsk.hr/urn:nbn:hr:200:609181

Bošnjak A. Detekcija grana za rezidbu na RGB-D slikama [Master's thesis]. Osijek: Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek; 2024 [cited 2025 March 24] Available at: https://urn.nsk.hr/urn:nbn:hr:200:609181

A. Bošnjak, "Detekcija grana za rezidbu na RGB-D slikama", Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, Osijek, 2024. Available at: https://urn.nsk.hr/urn:nbn:hr:200:609181

Please login to the repository to save this object to your list.
Metadata
TitleDetekcija grana za rezidbu na RGB-D slikama
Title (english)Detection of branches for pruning in RGB-D images
AuthorAndrej Bošnjak
MentorPetra Pejić (mentor)
MentorJana Dukić (sumentor)
Committee memberPetra Pejić (predsjednik povjerenstva)
Committee memberRobert Cupec (član povjerenstva)
Committee memberEmmanuel Karlo Nyarko (član povjerenstva)
GranterJosip Juraj Strossmayer University of Osijek
Faculty of Electrical Engineering, Computer Science and Information Technology Osijek
(Department of Computer Engineering and Automation)
(Chair of Automation and Robotics)
Osijek
Defense date and country2024-09-30, Croatia
Scientific / art field,
discipline and subdiscipline		TECHNICAL SCIENCES
Computing
Artificial Intelligence
Abstract
U ovom diplomskom radu predstavljen je prikupljeni podatkovni skup BRANCH koji sadrži RGB-D slike stabala voćki prije i poslije rezidbe. Predstavljen je i algoritam koji iz više dubinskih slika stabla rekonstruira 3D model stabla u obliku oblaka točaka. Također, opisana je i neuronska mreža temeljena na PointNet++ arhitekturi. Algoritam za rekonstrukciju 3D modela podrazumijeva odabir oblaka točaka za rekonstrukciju, uklanjanje pozadinskog šuma i preklapanje oblaka točaka pomoću RANSAC i... More ICP algoritama. RANSAC se koristi za inicijalnu globalnu registraciju, a ICP za dodatno poravnavanje točaka. Evaluacija algoritma za rekonstrukciju 3D modela obavljena je na sintetičkom modelu. Korištene su metrike Intersection over Union - IoU i chamfer distance. Rezultat metrike IoU, za vrijednost praga od 1 cm, je 63.39%, a za vrijednost od 5 cm, rezultat je 94.66%. Metrika chamfer distance predstavlja prosječnu udaljenost između sparenih točaka skupa i ona iznosi 2.897 cm. Uspješnost izvedbe algoritma rekonstrukcije modela na BRANCH podatkovnom skupu iznosi 74.29%. Preklapanjem rekonstruiranih oblaka točaka drveća prije i nakon rezidbe dobivene su oznake točaka koje pripadaju orezanim granama. Ove oznake, zajedno s oblacima točaka drveća prije rezidbe koriste se kao ulaz za učenje neuronske mreže. Neuronska mreža je temeljena na PointNet++ arhitekturi, na specifičnom modelu za segmentaciju dijelova objekta. Model mreže, program za učitavanje podatkovnog skupa i proces treniranja mreže su prilagođeni za potrebe diplomskog rada. Mreža je trenirana na skupu za učenje, a testirana na podatkovnom skupu za testiranje na kojem postiže točnost od 90.08%. Također na cijelom testnom skupu postiže preciznost 69.76%, odziv 67.85% i F1 68.79%. Upotreba kvalitetnijih senzora za prikupljanje slika sa što manje gubitka informacija te optimizacija algoritama za rekonstrukciju i klasifikaciju prijedlozi su za poboljšanje predložene metode za detekciju grana za rezidbu. Less
Abstract (english)
In this thesis, the collected dataset BRANCH is presented, which contains RGB-D images of fruit trees before and after pruning. Additionally, an algorithm is introduced that reconstructs a 3D model of a tree from multiple depth images in the form of a point cloud. The thesis also describes a neural network based on the PointNet++ architecture. The algorithm for 3D model reconstruction involves selecting point clouds for reconstruction, removing background noise, and aligning point clouds... More using the RANSAC and ICP algorithms. RANSAC is employed for initial global registration, while ICP is used for further alignment of points. The evaluation of the reconstruction algorithm was performed on a synthetic model. The metrics used were Intersection over Union (IoU) and chamfer distance. The IoU result, with a threshold value of 1 cm, is 63.39%, and for a threshold of 5 cm, the result is 94.66%. The chamfer distance represents the average distance between paired points in the set, which is 2.897 cm. The algorithm's performance on the BRANCH dataset achieved 74.29%. By overlapping the reconstructed point clouds of trees before and after pruning, point labels corresponding to pruned branches were obtained. These labels, along with the point clouds of trees before pruning, are used as input for training the neural network. The neural network is based on the PointNet++ architecture, specifically tailored for object part segmentation. The network model, dataset loading program, and training process were adapted for the purposes of this thesis. The network was trained on the training set and tested on the test set, where it achieved an accuracy of 90.08%. Additionally, on the entire test set, it achieved a precision of 69.76%, recall of 67.85%, and an F1 score of 68.79%. The use of higherquality sensors for image collection with minimal information loss, as well as the optimization of reconstruction and classification algorithms, are suggested for improving the proposed method for branch pruning detection. Less
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:200:609181
Study programmeTitle: Graduate University Study Programme in Computer Engineering
Study programme type: university
Study level: graduate
Academic / professional title: sveučilišni/a magistar/magistra inženjer/inženjerka računarstva (sveučilišni/a magistar/magistra inženjer/inženjerka računarstva)
Type of resourceText
File originBorn digital
Access conditionsOpen access
Terms of useLicence In copyright icon
RepositoryFaculty of Electrical Engineering, Computer Science and Information Technology Osijek
Created on2024-10-27 17:04:27
accessibility

closeAccessibilityrefresh

If you wish to permanently save changes, click on Save, if not - your settings will be reset when you restart the browser.