Scope

The main goal is to advance the field of computer vision by overcoming existing limitations in motion understanding. The project has three main objectives. In the first one we will develop new deep architectures for learning robust visual object representation based on a few examples. Such representation will allow to detect similar-looking objects. The second objective focuses on the development of methods for minimal scene understanding required for optimal object tracking, particularly in the presence of distractors. In the third objective, we will develop deep models capable of accurately tracking objects that undergo transformations or decomposition. Apart from advancements in computer vision research, the methods developed are expected to have a significant impact on various applications, spanning autonomous navigation, industrial production, and healthcare. These innovations will contribute to the creation of safer, more efficient, and intelligent systems that benefit society as a whole.

Robust object tracking and motion estimation is fundamental in diverse applications such as autonomous navigation, microbiology, first-person and ambiental activity analysis.

Workpackages

The project is divided into four methodological work packages (WP1-WP4) and two supporting work packages (WP5, WP6):

• WP1: Visual representation learning of general objects based on a few examples
• WP2: Visual object tracking of general monolithic objects in presence of distractors
• WP3: Tracking of general objects under complex transformations
• WP4: Challenge-specific datasets and operational demonstrators
• WP5: Dissemination and exploitation
• WP6: Project management

Publications