Ensemble of Regression Trees (ERTs) have been successfully used for face shape alignment in order to recognize facial expressions (driver drowsiness, human reactions, etc.). They can also be employed to align the shape of various objects ranging from living animals (e.g., fish) to distinct human organs displayed in CT or MRI scans, ultrasound images, etc. In ERTs, a shape is defined as a number of landmarks. For example, in human faces, specific landmarks are defined around the eyes, the nose or the mouth, in fish shape alignment landmarks are defined at the mouth, the eyes, the gills, etc. From the distance of specific landmarks facial expressions, shape malformations, etc., can be recognized or object dimensions can be measured. From the position of specific landmarks localized image processing can be applied to extract further features of the object. ESDALAB has implemented and introduces you to the ERTs accelerated in hardware, in order to increase frame processing speed and monitor objects like human organs from video streams in real-time.
Associate Professor in the Department of Electrical and Computer Engineering at the University of Peloponnese
