Computer Vision and Pattern Recognition in Sri Lankan research and industry
View(s):Computer vision attempts to make the computers “see” as human beings do. This is mainly done through recognising patterns in images and videos. Although humans do these effortlessly, they are complex tasks for the computer. Computer vision and pattern recognition previously seemed alien ideas to Sri Lanka, and as a matter of fact, to the consumers of the whole world. This is no longer the case. Now, these applications are commonplace.
When a camera detects faces and focuses well to produce a sharp portrait picture, a social network suggests names to tag people in a picture, we virtually travel through cities using Google street view, and we hear about self-drive cars. Computer vision is now here to stay. In all these applications, computer vision attempts to make the computer see as human beings do. Whenever a website suggests that we visit a page, or Amazon suggests books, we can see that that our browsing patterns have been recognised. Pattern recognition is about recognising patterns, as we humans do.
How Sri Lanka could benefit
Are these seemingly high-tech fields of artificial intelligence relevant to Sri Lanka? They are very relevant due to several reasons. First, in order to develop such an application, we do not need to make a large investment.
Second, developing an application of this nature needs top-quality brain power, which we have, judging from the activities of our students. Third, due to the explosion of digital data, such as statistics, pictures and videos, the need for such applications is unprecedented.
Why do computer vision and pattern recognition applications not require large investments? These applications, essentially, are pieces of computer software. To solve a problem in this area, an engineer would follow several steps. First, the client or the engineer would identify an interesting problem, e.g., analysing pictures submitted to a social networking site and identifying who the close friends of a profile owner are. Then he would look around to see if there are similar applications, usually by searching through recent academic publications.
Then the engineer will break the problem into several sub-problems, and line up the most appropriate technique to solve each sub-problem. For example, a sub-problem in the “close friend” problem would be to detect faces. Another would be recognising faces. Then the engineer will integrate these sub-solutions to form the final solutions. He or she may implement the solution in an appropriate computer language to run in an efficient manner. What we observe here is that we did not require a large factory, a large workforce, raw material, and expensive machinery. What is required are the problem definition and an intelligent engineer. This makes the computer vision and pattern recognition businesses easy to start. Although easy to start, it needs top brain power. Fortunately, Sri Lanka seems to have it. The work of final-year undergraduates, graduate students, and several high-tech organisations in Sri Lanka is evidence that we have the capacity.
Local projects
Let’s take a few example projects from the University of Moratuwa,financially supported by the university as well as the National Science Foundation, the National Research Council and the Senate Research Council. A couple of years ago, a group of final year project students developed an application that could use the regular CCTV images of the Southern Expressway to recognise license plates of Sri Lankan vehicles. This was quite a challenging problem, as the images were poor to the point that even human eyes could not read the plates with one image, as the cameras were not meant to be used for license plate recognition. The students developed a learning-based method that integrated information from approximately 25 frames to read the licenc e plates.
Another student developed a system for automatically solving square-piece jigsaw puzzles by observing visual similarities between pieces, and counting similarities along loops. Due to the work of a PhD. student, we now have a system that recognises activities, e.g., biking, driving, and tennis in videos. The student used a “convolutional neural network”, to learn the motion and static data. This method surpassed the state-of-the-art methods for activity recognition. Most of these are applications of machine learning in vision. I am sure that other universities too can get engaged in these very exciting pieces of work.
In addition to universities, several organisations in the industry, apply these intelligent techniques or are working on computer vision and pattern recognition products. Hi-Tech Solutions apply their solutions in the important apparel industry, for detecting defects, and intelligent counting.
Zone24x7, a US-based company with a technology centre in Sri Lanka, uses computer vision to develop a product that would digitise cheques enabling paperless banking by facilitating remote cheque deposits, using smart phones. I am sure that there are many other organisations who are successful in this area. As a result, there will be new job opportunities too in this exciting field. These organisations, manned by our engineers, are evidence that Sri Lanka is ready to embark on the computer vision and pattern recognition business.
A bright future
In order for us to have a bright future, in the computer vision and pattern recognition area, we need several things. Sri Lanka’s own organisations must approach those who are capable of providing solutions, and give them an opportunity to expand. It is important to understand that the proof-of-concept solution is expensive compared to what is to be deployed in numbers. Therefore, there must be an investment to get started. To show the industry that there is a solution, the problems must be solved in a research environment. Universities must be able to attract more research funding from government agencies and industry. Engineers of this calibre are very expensive. Funding must be sufficient to have them working on research problems. Giants such as Microsoft, Google, and Facebook have massive computational infrastructure, which makes training of these systems simple. We too must contemplate on investing on or having access to powerful computational infrastructure.
Computer vision and pattern recognition, from the examples we saw, will revolutionise the way digital information is handled and made use of. As Sri Lanka has talent that can generate income in this area, government and organisations must take steps to promote and invest. Then Sri Lanka will be able to be in the forefront of the knowledge era.
(The writer is a BSc Eng Hons with first class honours graduate of the University of Moratuwa, while also holding a MESc and PhD from Western University, Canada. His research interests are in the general area of computer vision. Currently, he is a Senior Lecturer with the Department of Electronic and Telecommunication Engineering, University of Moratuwa).