![](bg3.png)
or body and it is difficult to observe the surgical field
with a single camera constantly.
Other attempts have also been made to record
surgery with a surgical field camera placed between
the eyes of a doctor. The camera of such recording
systems were not high resolution and did not pro-
duce good video quality (Matsumoto et al., 2013;
Murala et al., 2010) because of the limited hardware
system. In addition, doctors had to perform surgery
with interference by the surgical camera itself and
its code. Nair et al.(Nair et al., 2015) recorded the
surgery by putting a high-resolution camera (GoPro
Hero 4) on the doctor’s head. The doctor’s head
moves greatly during the surgery, and camera cannot
always shoot the surgical fields, and video is always
shaking. Therefore, video recorded by it can be of-
fending for viewers.
In our proposed system, multiple cameras are at-
tached to multiple lights mounted the surgical lamp,
and the surgery is recorded by them. While one of the
lights illuminates the surgical field, we assume that
the surgical field can be observed one of the attached
cameras. Therefore, the surgical field is recorded
without disturbing the surgery.
2.2 Multiple Camera Switching
In recent years, multiple cameras are introduced in
any place, such as office environments, sports stadi-
ums, and downtown areas. Instead of its convenience,
it is difficult to extract only the necessary information
from the huge amount of video sequences from a lot
of cameras. Therefore, camera self-control technol-
ogy, such as the automatic viewpoint switching video
generation, and highlight video generation technol-
ogy, are regarded as important issues (Chen and Carr,
2014).
Liu et al.(Liu et al., 2001) interviewed pro editors
to gain knowledge of video editing, and implemented
the camera switching rules. Based on the rules, they
switched the viewpoint of three cameras shooting the
speaker, the audience, and the entire in conference
video. Doubek et al.(Doubek et al., 2004) observed
moving objects using multiple fixed cameras in an
office environment. Selection of a camera was per-
formed based on the score of each camera, and the re-
sistance coefficient was introduced so that the switch-
ing of a camera may be performed only when the
score changed significantly. However, such camera
switching strategies may occur frequently if cameras
with competing scores existed, which may reduce the
overall QoV.
In order to suppress camera switching frequency,
Jiang et al.(Jiang et al., 2008) proposed a cost func-
tion calculated based on the size, posture, orientation,
etc. of the target, and controlled the frequency of the
camera switching while considering QoV. Daniyal et
al.(Daniyal and Cavallaro, 2011) calculated the visi-
bility score of an object using a multivariate Gaussian
distribution model, and used the partial observation
Markov decision process for the camera switching to
maximize the visibility score while suppressing the
camera switching frequency. Although these methods
selected the optimal view using past sequential infor-
mation, the switching should be conducted using not
only past but also future information. Compared to
the conventional switching method, our method uses
both past and future frames to switch cameras so that
a higher QoV can be generated.
Also, QoV and user-specified weights for camera
switching may change depending on the target scene,
because they detected the event for calculating QoV.
For that reason, the hyper-parameter of the method
which determines camera switching depends on the
surgical scene, and it is difficult for non experts to de-
termine. On the other hand, in the proposed method,
since the minimum frame during that camera switch-
ing does not perform is specified and optimization is
performed for the entire frame, the camera switching
frequency does not change depending on the target
surgery scene.
3 MULTI-CAMERA RECORDING
As shown in Figure 2, we attached multiple cam-
eras to multiple lights mounted the surgical lamp.
Thereby, as long as the surgical field is illuminated
by one of the lights, our proposed camera recording
system shoots the surgical field. Compared to the
previous camera recording system (Matsumoto et al.,
2013) which attached cameras to the doctor’s head,
our system does not bother the doctors during surgery
while maintaining visibility of the surgical field.
4 PROPOSED METHOD
Figure 3 shows the overview of the proposed method
which consists of two components: camera scoring
and camera switching. The multiple surgical videos
are captured from the our capturing system (Section
3). To switch between camera sequences to generate
the best video quality, the frame in each sequence has
to be scored. The score represents how the surgical
field can be seen in the image. In our methods, first,
the score is estimated against each frame in each se-
quence. Next, the frame is selected sequentially using
the score.
Surgery Recording without Occlusions by Multi-view Surgical Videos
839