Section: New Results

Head Detection for Eye Tracking Application

Participants : Thanh Hung Nguyen, Antitza Dantcheva, François Brémond.

Keywords: computer vision, head detection, eye tracking

Head detection [77] uses RGBD sensor (Kinect 2 sensor) which is supported by SUP platform of STARS team. For the eye tracking, we use the open-source library (OpenBR) which has good performance in our test.

Until now, the head detector was working well when people were standing but not good enough when people were lying or bending as you can see on Figure 5 and Figure 6 . This experiment was realised on simple datasets where mostly people was close to camera and walking. The main reason is the lack of samples in the learning process for the challenges cases (lying, bending). So at this moment we are collecting the head samples for it.

Figure 5. Performance of head detection on simple dataset

Figure 6. An example of experiment