Pablo Barrera González: Research
Research Work
There are also some videos with important results of these research projects.
METS (Multi-Eye Tracking System) and tridimensional tracking
METS is the main application developed for my PhD Thesis. Its aim is to track several objects at same time using two or more digital cameras.
Using two or more cameras allows a direct estimation of the 3D position of the object, providing we are able to localise the same object in each view. Matching procedure can be complex what translates in a high computational cost. An alternative consists in work directly in the 3D space, make hypothesis of the object position and, later, validate each hypothesis using the information retrieved with the cameras.
As it is shown in the video, the system is able to localise four different objets in statical positions. They appear and disappear and the system is able to identify the number of objects correctly. The system localises each new objet as soon as it appears. It speed is due too the use of an abduction technique that places new hypothesis in the most probable regions of the space. Otherwise the localisation of new objects is to slow, as it is shown in this video where the abduction technique is not used.
Surveillance Applications
Tridimensional tracking can be applied to visual based surveillance systems. With this work, four cameras connected to a remote host are used to localise the tridimensional of a person inside a big room. Using the same techniques developed in Mets, we are able to use the information of the position of each person to take different actions.
This is a video surveillance application. It detects when a person is too close to a restricted area and then it fires an alarm. If it is approaching another person who does not meet the requirements in colour or size, the alarm is not activated. The last version is also able to track several person at same time.
In this case the system tries to locate people who have fallen to the ground using the tridimensional position of the object. As the estimation is a real 3D positions it is straightforward to calculate the height of the person. When the system detects a fallen person it activates an alarm.
Visual Attention
This is a navigation system that uses visual attention techniques and 3D reconstruction. The robot is able to move the cameras to localise visual landmarks and potential obstacles such as walls. Once they have be localised, the system add them to an internal 3D representation of the world that it is used to navigate safely through it. The system also move the cameras to ensure that objets added to the representation of the world remains in place.
