Developing autonomous robotic systems that are able to assist humans in daily tasks is one of the grand challenges in modern computer vision and robotics research. In this talk, I consider the task of a robot searching for and retrieving objects of interest from an unknown environment using only a monocular camera. Both efficient detection of the object and accurate pose estimation are essential to the task. I will discuss Active Deformable Part Models (ADPM), an active learning approach for part-based object detection that dramatically speeds up the detection by reducing the number of part evaluations. Building on top of the detection hypotheses, object foreground can be further segmented for estimating the object 6-DOF pose using only shape information. I will discuss the resulting accurate shape-based pose estimation method that enables autonomous robotic grasping using a single image.
Menglong Zhu is a Ph.D. candidate in the Department of Computer and Information Science at the University of Pennsylvania. His current research focus is 2D and 3D object recognition and pose estimation. He is also interested in semantic localization, text recognition, and action recognition. He obtained his Master’s degree in Robotics from the University of Pennsylvania in 2012, and a Bachelor’s degree in Computer Science from Fudan University, China, in 2010.
Host: Kris Kitani