Radu-Laurenţiu Vieriu, PhD

Research & development

As a teaching bonus, I got in contact with many bright and enthusiastic minds, some of which I got the pleasure to supervise during their bachelor theses. Bellow, I enumerate some of them:

  • Samuel Giacomelli (2016): together with Samuel, we have addressed a multi-label classification problem, targeting facial action units in still images. The solution contains traces of deep learning :)

  • Roberto Molinari (2015): Roberto studied the possibility of "frontalizing" a face captured with a standard RGB camera, given as input the orientation of the head and a few facial landmarks. He proposed a method of his own that builds the "frontalized" face by merging various texture patches from the face, that have been warped into their frontal position using the head pose information. Roberto defended his thesis in 2015.

  • Davide Todeschi (2014): with Davide, we have investigated the use of specialized regression trees for head pose estimation from range images. He successfully defended his thesis ("3D Head Pose Estimation in tempo reale" in March, 2014

My research focuses on applied machine learning for pattern recognition and human behavior analysis. Here are some research projects I've been working on:

Linking safety perception in Streetscape images with mobile phone activity patterns: this work investigates the influence of safer-looking places on the population demographics of two Italian cities (Milan and Rome). Our findings are in-line with Jane Jacobs' natural surveillance theory and Oscar Newman's defensible space theory.

More details here.


Between March and October, 2013, following Russel & Norvig's AI book, I prepared and held a course in general Artificial Intelligence at the Faculty of Cognitive Science, University of Trento. I designed the course for undergraduate students (3rd year) and included 4-hour lecture presentations, homework assignments, lab and several exam sessions.

Bellow, you can find most of the resources I used during the course:

Facial Expression Recognition under a Wide Range of Head Poses : this is an extension of previous work on head pose estimation, that essentially decouples the facial expression problem from the appearance variation due to the rotation of the head. This is done by building a head pose - invariant face representation, given as input the point cloud corresponding to the face, the two eyes' positions and the head pose.

More details here.


Robust Real-Time Extreme Head Pose Estimation: in this work we combine a head pose detector trained on range images with a head pose tracker powered by an improved ICP, in order to get the best of both worlds: fast recovery from lost track (thanks to the detector) and accurate informed head pose prediction (thanks to the tracker)

More details here.


While in Trento, my work was split into research & development and supervision. I've also done some teaching in the past. Bellow, you can find evidence of all three: