Research Experiences

 Graduate research 

Topic:State Chart Automata for Variable Structure System

Supervised by: Prof. Dr. rer. nat. Uwe Aßmann
Fakultät Informatik (nice brochure) (nice film)
Institut für Software- und Multimediatechnik (SMT)
01062 Dresden
Germany

 Problem Statement

– State charts are crucial while modeling a complex reactive system.
– Automotive systems, air traffic control systems, industrial
automation, etc all are reactive systems.
– Some common criteria of these systems are:
▶ Their environment is changing dynamically
▶ Not all future events can be predicted beforehand
▶ Some safety-critical systems can not be stopped to adapt to the change


Solution Concept
– Solution: We will change the state chart at run-time.
– Research Questions:
▶ What is the related research on this topic?
▶ Does any current state chart framework support runtime adaptivity?
▶ If no, how can we implement it?

Github Link

 

 

Undergraduate research 

Topic: Fish-eye video coding using affine motion model

Supervised by: Ashek Ahmmed, Assistant Professor
Department of Computer Science and Engineering
Rajshahi University of Engineering and Technology, Rajshahi, Bangladesh.

Overview:

In today's era, a 500$ mobile phone can capture videos up to UHD(3840 X 2160) resolution. People not only capturing these videos but also uploading them on facebook, youtube, and many
Other sites. In every minute 100 hours of videos are uploaded in youtube and over 6 billion hours of video are watched in each month. The typical video resolution range In youtube is from 240p(426 X 240) to 2160p(3840 X 2160), but recently 8k which has the resolution of 7680 X 4320 is also available for some videos. Video applications have expanded from television to notebooks, smartphones, tablets, phablets, etc. over the last decade. For several purposes, various kinds of the lens are used such as a standard lens, macro lens, telephoto lens, wide- angle lens and specialist lens, etc. In surveillance, automotive driver assistance system and sports, a wide-angle of view need to be captured where a fisheye lens is used. Video coding makes the expansion of video-based applications easy. Raw videos contain a huge amount of data which is very hard to store and transmit and with the help of video coding it is possible. Video coding is the process of reducing the number of bits before transmission and storage. Video encoder generates the motion information and communicates to the decoder. The decoder then tries to generate the original video with the help of reference frames and motion vectors. The thesis work is based on the observation of the characteristics of fisheye video sequences which have high distortion rate and nontranslational nature. Traditional codecs such as H.264/AVC, HEVC uses translational motion model wherein the cases of high distorted areas they divide smaller blocks and try to predict, but error rate remains high because those parts of the sequences are non-translational. Dividing more blocks increases the bit rate still accuracy is low.

Video Coding

Compression is the practice of reducing the number of bits (0s or 1s)
in a le. A typical video le contains images and audio. All the
images and audio are a combination of zeros and ones. All these
components are compressed in video compression.

Importance of video coding

1. Total 400+ hours of videos are uploaded on youtube every minute
   and 100 million hours of video watch time daily on Facebook. It is
   roughly estimated that in 2019, 80% of total storage on the internet
   will be videos.
2. In limited bandwidth video signals should be passed. Video
   compression allows the effie cient utilization of bandwidth by reducing it's sizes.

Fish-eye lens 

A fish-eye lens is an ultrawide angle lens to create a full panoramic or
hemispherical image.

Importance of fi sh-eye lens

1. In many applications, such as surveillance, automotive driver
   assistance systems, and recreational sports there is a need to preserve very large fi eld of view(FOV), typically well beyond the common 60 to 90 degrees. It is possible to achieve a FOV of 180 degrees or more using a sheye lens.
2. Fisheye lens video cameras are also used in virtual reality applications.

Example of a fi sh-eye frame 

Problems with traditional codec

1. From the the above image of fi sh-eye, we have seen that they have high radial distortion and do not follow the law of perspective.

2. Traditional codec use translational motion model to compress and
decompress fish-eye videos like the typical video, so error rate remains
high.
3. Our proposed approach is to use a ne motion for compressing
fish-eye video sequence.

Drawbacks of traditional codecs while compressing fish-eye videos 

The edges of the the fish-eye frame is distorted. Traditional codecs can
not efficiently compress those parts of the image.

Motion estimation

1. Motion estimation is the process of determining motion vectors that
describe the transformation from one 2D image to another; usually
from adjacent frames in a video sequence.
2. In block matching algorithm best-matched block is found by searching
a grid adjacent to the pixel.
3. Then rst frame and the motion vector is sent. Thus the size of the
second frame is reduced.

Continue reading