Vision is one of the main potential sources of information for robots to understand their surroundings. For a vision system, a clear and close enough view of objects or events, as well as the viewpoint angle can be decisive in obtaining useful features for the vision task. In order to prevent performance drops caused by inefficient camera orientations and positions, manipulating cameras, which falls under the domain of active perception, can be a viable option in a robotic environment. In this paper, a robotic object detection system is proposed that is capable of determining the confidence of recognition after detecting objects in a camera view. In the event of a low confidence, a secondary camera is moved toward the object and performs an independent detection round. After matching the objects in the two camera views and fusing their classification decisions through a novel transferable belief model, the final detection results are obtained. Real world experiments show the efficacy of the proposed approach in improving the object detection performance, especially in the presence of occlusion.

Designing Collaborator Robots for Highly-Dynamic Multi-Human, Multi-Robot Teams, Office of Naval Research (ONR) PI: Monica Nicolescu, co-PI: David Feil-Seifer, Amount: $656,511, April 1, 2016 - March 31, 2019

DURIP: Humanoid Platforms for Human-Robot Collaboration, Office of Naval Research (ONR) PI: Monica Nicolescu, co-PI: David Feil-Seifer, Mircea Nicolescu, Amount: $312,000, May 1, 2014 - June 30, 2015