Sota Shimizu, Ph.D.
Associate Professor,
Advanced
Research Institute for Science and Engineering, Waseda University
17
Kikui-cho, Shinjuku-ku, Tokyo, 162-0044, Japan
Home
Information
Research
Papers
Links
Contact
Current research
interest:
The human eye has a 120-degree wide field of view horizontally (by a single eye). Its visual acuity is the highest in the central field of view and decreases rapidly towards periphery [1]. Thus, it means that there exists an explicit attention region in the field of view. By combining eye movement, we can observe an object as more in detail as possible and as with small amount of information as possible. My current research interest is to know a mechanism of the human brain-vision system comprising such a smart bio-sensor.
[1]
S. Shimizu, Wide-Angle Foveation for All-Purpose Use, IEEE/ASME
Transactions on Mechatronics, Vol.13, Issue 5, pp.587-597 (2008.10)
PDF
1. Bio-inspired Fovea
Vision Sensor:
Figure 1 shows a
picture of a developed fovea sensor, where a special made wide-angle lens
is attached with a commercially-available area sensor. This bio-inspired artificial
vision sensor is applicable for all-purpose use, e.g., surveillance, end
scope, robot control, and etc., just by a single sensor. Figure 2 shows a foveated input
image from this sensor. One
of the most remarkable advantages of this unique vision system is to
reduce the number of data drastically in the entire field of view. Moreover, the fovea sensor
realizes both wide-angle field of view and high resolution in the central
field of view, simultaneously.
Although a couple of methods have been proposed to acquire such a
foveated image, an optical approach is the best for the highest resolution
in the central field of view.
We can extract a log-polar image, which is useful for pattern
recognition due to its rotation and scale-invariant properties, from the
foveated input image very easily.
one of the most beautiful and rational ways to acquire such an
image, on earth. Fig.3 shows
the extracted log-polar image.
A new fovea vision sensor is being developed currently in our
laboratory.
Fig.1 Fovea Sensor Fig.2 Foveated input image
Fig.3 Extracted
log-polar image
2. Eye Movement
Analysis:
It is quite essential for fovea vision to know how the human look at and see the object. Thus, eye movement data are measured using an eye-tracking device and are analyzed paying attention to gaze decision-making.
Fig.4
Experimental scene using an eye-tacking device
3. Mobile Robot Navigation based on
Multi-functional Use of Fovea Sensor:
Visual information acquired from the fovea sensor can be applied for various tasks simultaneously and cooperatively. Paying attention to all-purpose use of the foveated image, a mobile robot is navigated autonomously using a stereo camera head which can change view directions of 2 fovea sensors. Figures 5, 6, and 7 show a picture of the mobile robot, the stereo camera head, and an example of multi-functional use of the foveated input image. In case of Fig.7, the central and peripheral fields of view are used for 3D measurement and localization, respectively.
Fig.5 Mobile robot Fig.6 Stereo camera head fovea sensors
Fig.7 An example
of multi-functional use of the foveated input image
4. Space-variant Data
Processing:
Figure 8 shows 3 images having different distributions of non-uniform spatial resolution. We, human beings, can recognize that these three images are the same intuitively due to our flexible and robust brain structure. However, we often need to use different ways in order to reduce a loss of both their high spatial resolution and wide field of view when they are compared computationally using signal processors.
Fig.8 Comparison of images with non-uniform spatial resolution