Vision and Robotics Seminar
Dept. of Solid Mechanics, Materials and Systems
The Iby and Aladar Fleischman Faculty of Engineering
will speak on
Biologically Inspired Vision System
for the Detection of Partially Occluded Objects
The use of a pseudo-retina methodology (space variant sensor with saccading) in computer vision is relatively new and seldom. The early work of (Schwartz 1977) and Yeshurun and Schwartz (1989), was followed by several groups starting primarily in the late 90s. The topic is currently gaining increased attention with the availability of some specially designed hardware. Much of this work
is based on the log-polar nature of biologic sensors such as the human eye. The lack of adequate hardware, dictating extra transformations (re-mapping), has probably influenced the rare use of pseudo-retina methods in computer vision.
Another factor that may have influenced the lack of popularity of retina-based methods is the semi-serial nature of the detection process. An important issue in autonomous robot vision system is the variability of the features, which are used in the computational process. A successful autonomous system must cope with the space of all possible views and light conditions of all the objects to be recognized. To help overcome the diversity in object appearance a new set of object representations has been developed. We suggest novel features that are collected within the cylindrical coordinates of a retina-like sensing. The features, which we have termed as Local Panorama Features (LPFs), are grouped in a special way such that partially occluded objects can be detected. The developed system uses a combination of colors and shapes to locate and recognize the object. The results, obtained from simulations, are promising both in terms of the robustness to variances, and the discrimination capability of the system.
The lecture will take place in the
Lecture Hall, Room 1, Ziskind Building
on Thursday, June 27, 2002