Computer Vision Based Tomato Fruit Segmentation and Volume EstimationBLAŽEVA, BLAGICA (Avtor) Perš, Janez (Mentor) Mihelj, Matjaž (Komentor) deep learningMask R-CNNtomato fruit segmentationsphere fittingpoint cloudThis thesis focuses on developing a method for tomato fruit segmentation from images of tomato plants and volume estimation of the segmented tomato fruits. The proposed method for tomato fruit segmentation is a deep learning method, which consists of a pre-trained Mask R-CNN model. To evaluate the model's performance, we assembled and annotated an image dataset of tomatoes in different sizes and colors, shown in their natural growing environment. Additionally, we trained a Mask R-CNN model for segmentation of the leaves and branches to exclude them from the tomato fruit segmentations for further ripeness analysis. We performed a pixel-wise evaluation of the segmentation for both models, using metrics such as precision, recall, and F1 score. The volume estimation method is based on fitting a sphere to the tomato fruit point clouds, using the least squares method. The point clouds were obtained using an RGB-D camera. The proposed method showed good estimation results in a controlled environment, with an error of 5 - 15 \% compared to the results measured with the water displacement method.20222022-09-09 13:15:00Diplomsko delo/naloga140019VisID: 61006COBISS_ID: 121474563sl