Motion-blur-free video shooting system based on frame-by-frame intermittent tracking
Inoue et al. Robomech J
Motion-blur-free video shooting system based on frame-by-frame intermittent tracking
Michiaki Inoue 0
Mingjun Jiang 0
Yuji Matsumoto 0
Takeshi Takaki 0
Idaku Ishii 0
0 Department of System Cybernetics, Hiroshima University , Higashi-Hiroshima , Japan
In this paper, a concept of frame-by-frame intermittent tracking to achieve motion-blur-free and high-brightness images when video shooting fast moving scenes is proposed. In our tracking concept, two control methods are applied alternately at hundreds of hertz, according to the open or closed shutter state of the camera. When the shutter is open, the target's speed in images is controlled at zero and visual feedback is transmitted to achieve motion blur reduction, and when the shutter is closed, the camera returns to its home position. We developed a prototype of our motion-blur-free video shooting system, which consists of our tracking method implemented on a high-speed two degrees-of-freedom tracking vision platform that controls the pan and tilt directions of the camera view by using high-speed video processing in order to reduce motion blur. Our motion-blur-free video shooting system can capture gray-level 512 × 512 images at 125 fps with frame-by-frame intermittent tracking. Its performance is verified by the experimental results for several video sequences of fast moving objects. In the experiments, without a decrease in the exposure times our method reduced image degradation caused by motion blur.
Target tracking; High-speed vision; Active vision; Motion deblurring
Introduction
Motion blur is a well-known phenomenon that occurs
when shooting images of fast moving scenes. The
degradation degree of images caused by motion blur depends
on the duration of the camera exposure, as well as on
the apparent speed of the target scenes, and the
camera’s exposure time is often decreased in order to reduce
motion blur. However, a trade-off exists between
brightness and motion blur in image shooting, because it is
difficult to obtain non-blurred bright images with a
decreased exposure time, since less light is then
projected onto the image sensor. This trade-off is extremely
aggravated in highly magnified observations of fast
moving scenes in various application fields, such as flowing
cells in microscopic fields, precise inspection of products
on a moving conveyor line, and road surface and tunnel
wall inspection from a moving car, because the
apparent speed of the scene increases in the magnified camera
view and the light that is projected on the image sensor
diminishes when the magnification is increased.
Motion deblurring is a frequently used approach for
reducing this image degradation resulting from motion
blur in image shooting. In many studies, approaches [
1,
2
] were developed that apply blur kernels that express
motion blur in the input images; the blurred images are
restored by deconvolving the images with the estimated
blur kernels. These approaches include single-image
deblurring methods [
3–5
] and multi-image deblurring
methods [
6–8
]. In the former, the blur kernels are
estimated from a single image using parametric models for
maximum a-posteriori estimation and in the latter the
illposed problems in deconvolution are reduced by
estimating the blur kernels from multiple images. Several papers
have reported motion deblurring methods that consider
the camera’s egomotion while the shutter is open, which
is estimated by using gyro sensors and accelerometers [9]
or the camera’s geometric location [
10
]. However, most
of these methods adopt a software-based approach for
image restoration and do not consider the acquisition
of blur-free input images. There are limitations to the
extent to which images can be improved, in particular
when significant changes in the target scene occur in the
images captured while the camera shutter is open.
To reduce motion blur resulting from camera shake,
a large number of digital cameras with camera-shake
reduction systems have been developed that can
stabilize input images by shifting their optical systems
mechanically. These image stabilization systems are
categorized into two types according to the approach that
is applied: the lens-shift approach, which shifts a
floating lens to move the optical axis [
11, 12
] and the
sensorshift approach, which shifts the image sensor [13]. These
image stabilizers can stabilize input images by
controlling the optical path with a floating lens or the position
of the image sensor with the camera’s internal sensors,
such as its gyro sensor. The camera-shake reduction
system is a camera-stabilization approach that uses the
camera internal sensors for reducing motion blur resulting
from camera shake; it is unsuitable for shooting blur-free
images of fast moving scenes when the camera is fixed,
because the internal sensors cannot detect any apparent
motion in the captured images.
Many high-speed vision systems operating at 1000 fps
or more have b (...truncated)