Evidence record · UDINUS · SDGs
- Unit
- Rektorat UDINUS
- Period
- 2025
- Goals
- Record
- 3badbc75-cc76-4c7e-80ea-b01932bfe40b
- Published
- 2026-10-30
Universitas Dian Nuswantoro (UDINUS), through participating researchers Prakoso, Erik; Syukur, Abdul; and Affandy, conducted research on a non-contact method for estimating toddler height from monocular images. The activity addresses stunting as a chronic nutritional problem that inhibits the growth and development of children under five and requires early detection through accurate height monitoring. The research responds to manual measurement methods that are often inaccurate and impractical for active toddlers.
The activity proposed a computer vision and deep learning model for estimating toddler height from a single image. Its architecture integrated Mask R-CNN through Detectron2 for precision instance segmentation, MediaPipe Pose for three-dimensional pose keypoint estimation and bounding box adjustment, and DenseNet121 as the main backbone for feature extraction and height prediction through a regression approach. The model was trained using transfer learning and fine-tuning techniques on personal toddler images containing variations in posture and lighting conditions.
Image preprocessing included exposure adjustment, denoising, sharpening, background removal, and resizing to improve input-data quality. Model performance was evaluated using Mean Absolute Error, Root Mean Squared Error, and tolerance-based accuracy within five centimetres. The source does not provide numerical evaluation results. The proposed integration aims to improve the accuracy and robustness of toddler height prediction from a single image and to offer a more efficient and non-invasive approach to stunting detection.
In relation to the stated goal, the activity connects toddler height monitoring with early detection of a chronic nutritional problem affecting children under five. UDINUS presents this institutional research activity as an application of computer vision, deep learning, image processing, pose estimation, instance segmentation, and regression-based prediction for non-contact toddler height estimation.