Skip to main content

Research Repository

Advanced Search

3D plant phenotyping system using photometric stereo

Bernotas, Gytis


Gytis Bernotas
Research Fellow in Computer Vision and Machine Learning


Tracking and predicting the growth performance of plants in different environments is critical for future crop development, which is under dual pressure from population expansion and global climate change. Automated approaches for image capture and analysis have allowed for substantial increases in the throughput of quantitative growth trait measurements compared to manual assessments. This thesis presents the development of PS-Plant - a low-cost and portable 3D Plant Phenotyping (PP) platform based on an imaging technique novel to PP called Photometric Stereo. The suitability of photometric stereo for imaging vegetation was investigated through empirical leaf tissue reflectance measurements which showed that Near-InfraRed (NIR) wavelengths are better suited for this purpose than the visible light. This was further affirmed during PS-Plant validation procedure that involved estimation of the surface area and angle of calibration objects and leaf tissue under white and NIR illumination, achieving state-of-the-art results - a mean relative error of 1.09% for area and a mean absolute error of 1.28 degrees for angle estimations. During this project, a large-scale PS-Plant database of annotated Arabidopsis thaliana leaves of 41 plant individuals grown in different environmental conditions captured in over 470 sessions was made available to the public, which is believed to be the first of its kind. The work introduced six image modalities captured the PS-Plant database, which were used to train two state-of-the-art instance segmentation Neural Network (NN) architectures for automated leaf segmentations. It was concluded that grayscale mages attained the highest scores for both architectures when compared to other investigated modalities and attained a mean symmetric best dice score of 64.1 and 80.3% for two of the investigated neural networks. Further, the differences in performance were established for data splitting technique for training and validating the NN architectures with time-series images of growing plants. The proposed data acquisition and processing pipeline was utilised in a use-case study to extract, in an automated manner, traits of varying complexity from PS-derived data, including the tracking of 3D plant growth and diel leaf hyponastic movements of plants grown under a matrix of different conditions. The work has shown that PS-Plant is a powerful new phenotyping tool for plant research that provides robust data at high temporal and spatial resolutions. The system is well-suited for small and largescale research and will help to accelerate bridging of the phenotype-to-genotype gap.


Bernotas, G. 3D plant phenotyping system using photometric stereo. (Thesis). University of the West of England. Retrieved from

Thesis Type Thesis
Publicly Available Date Aug 23, 2019
Keywords 3D, Arabidopsis thaliana, computer vision, leaf angle, leaf reflectance, low-cost, segmentation, machine learning, near-infrared (NIR) LED, Photometric Stereo, plant phenotyping, Raspberry Pi
Public URL


You might also like

Downloadable Citations