Some Problems and Intelligent Multimedia Solutions for Smart Cities
India offers a rich landscape of problems for information technology based solutions. Every problem is an opportunity for quality science, powerful technologies and social enterprises – i.e., good academics, good money and much satisfaction, all at the same time.
This happy state of affairs is a new phenomenon. Solving large, geographically distributed problems has become increasingly feasible with increasing availability of diverse and distributed problem related information and data, made feasible by sensors embedded in things of day to day use with easy access to high speed communication networks comprising Internet of Things, and availability of powerful embeddable computers. Since the data understanding component has not kept pace with the growth of data generation and communication, there is much need for intelligent multimedia systems. Emergence of such systems has already begun to impact many problems that were earlier considered very difficult if not insurmountable.
This talk will review a cross-section of such problems, ubiquitously present in India, and some examples of solutions developed by my research group at UIUC as well as by institutions across India that are part of the ITRA ecosystem. Some of these solutions have led to products and services already in place, while others are in the process of being transferred through start-ups or existing industry.
If such problems are solved by students as a part of their research/other projects, there will be no reason for them to look for a job; they can create plenty by themselves, for themselves and for others.
Prof. Santanu Chaudhury
IIT Jodhpur (IIT Delhi)
Prof. Sumantra Dutta Roy
Motion Segmentation and Tracking
The talk will dwell on a popular robust technique for motion segmentation, and then go onto some popular techniques for visual tracking, particle filters and Eigen Tracking. The first part of the talk will focus on a method based on practical Engineering heuristics, which work well for a particular task. This can be useful for automatic object segmentation of moving objects from videos. The second part of the talk will focus on a survey and demonstration of some popular algorithms for state estimation and tracking in videos, which will consider objects, which not just change their position, the talk will examine objects, which change their appearance as well.
Prof. Rahul Banerjee
The LNMIIT, Jaipur
Cyber Physical Systems
Dr. Ujjwal Bhattacharya
Issues of Automatic Document Analysis
Electronic documents are available either as a pixel image or as a text file. In the present lecture, we shall deal with the analysis of document images. The topic Document image analysis deals with suitable algorithms and methods that are applied to images of documents to obtain a text file from pixel image file. Since an arbitrary document may have a very complex structure containing tables, images, drawings; formatted in one or multiple columns; oriented in one or multiple directions and so on, the algorithms dealing with them need to be extremely sophisticated. Due to the extensive research during the past several decades, the problem is currently mostly solved as far as the input document is a good quality, more or less well-behaved one. This is true for a large number of scripts of the world. A popularly known product of document image analysis research is the Optical Character Recognition (OCR) software that recognizes individual characters in a scanned document. State-of-the-art OCR algorithms are capable of providing nearly 100% accuracy on reasonably clean documents. However, the situation deteriorates excessively when the input document is a degraded one. Due to the recent advances in deep learning methods, somewhat major breakthroughs could be achieved in cases of degraded document image analysis tasks. In the present lecture, we shall give special attention to such degraded document analysis tasks which has major impact on our society considering the preservation aspect of old treasures stored either in public or in personal libraries
Dr. Anoop M. Namboodiri
Recognizing People in Images: Challenges and Opportunities
With the ubiquity of surveillance cameras, detecting and recognising people in images has become an important problem for a wide variety of applications. In the context of smart cities, these applications include efficient deployment of resources, crowd control, security, etc. However, solutions to the recognition problem faces challenges in terms of its accuracy, efficiency and cost. More importantly, these applications also raise serious concerns of privacy and safety, especially with vulnerabilities in the digital networks. This talk will focus on approaches to improve accuracy, efficiency and privacy of human recognition systems from images.
Dr. Phalguni Gupta
NITTR (IIT Kanpur)
Challenges in Fingerprint based Biometric System
Fingerprint is one of the best known and well accepted biometric traits. It is an impression that gets developed on a surface touched by the lower skin of a human finger. Print pattern of a fingerprint contains a lot of black lines called ridges. A ridge can either terminate or join with other ridges at the end. Both of these points are of special interest and are called minutiae points. Location along with the direction of the minutiae point contributes towards the individuality of a fingerprint. Despite being one of the most popular and widely accepted biometric trait, performance of a automatic fingerprint based recognition system is not 100\% accurate. Especially for large-scale deployments. There exist several challenges which one faces while designing an efficient fingerprint based biometric system. Some of the well known challenges are (i) designing an efficient technique to extract true feature points, (ii) Efficient technique to get stable core point (iii) Effective technique for indexing a big fingerprint database (iv) A good and distinguishable fingerprint matching technique (v) Developing an appropriate measure to determine the quality of fingerprint (vi) In presence of multiple instances of the fingerprint an efficient score fusion strategy to unify the scores for better performance (vii) Designing recognition systems for the rural population where quality of fingerprint is very poor (viii) Efficient and accurate technique to segment fingerprints automatically from a digital slap image that contains fingerprints of four fingers in a single image.
Dr. Anil Kumar Tiwari
Image and Video Processing
Signal compression is highly desired for its efficient storage and for transmission purposes. Image and video signals are two such examples and compression methods used with such signals are generally of two types: 1) Lossless and 2) Lossy. In lossless compression, as name suggests, decoder reconstructs the exact original signal at the expense of low compression ratio. On the other hand, lossy methods give high compression ratio but causes loss of information. Mostly, signals are compressed using lossy methods. Image files with .jpg extension and video files with .mpg extension are compressed by lossy methods. However, typically, lossless compression methods use to be a component of lossy methods. Besides being a component of lossy methods, lossless methods have specific application areas such as medical imaging, images of historical importance, imagesfor art and culture etc. In this talk. we shall discuss various methods used for lossless compression and will also discuss about the challenges in development of new methods.
Dr. K.S. Raju
Intelligent Internet of Things
One of the most promising potential technology with interdisciplinary is Internet of Things. IoT is all about information visibility – without any problem in maintaining privacy, security and confidentiality. Thus, IoT is the network of things, with device identification, embedded intelligence, and sensing and acting capabilities, connecting people and things over the Internet. Around 60 billion IoT devices Edge, Gateway) are expected in use by 2020. IoT technology would be applied for many applications such as smart cities, infrastructure [Old and heritage buildings and bridges, Oil and Natural GAS pipelines, Big structures (steal and civil)] maintenance and prediction of SHM, Intelligent Health care Intelligent Transport, Intelligent building systems and Smart power grid etc. Having lot of opportunities, they also pose challenges in terms of system robustness and reliability, real-time data handling and transfer, efficient fault tolerance techniques with interoperability are the major issues. The solutions for these issues could be efficient system architecture with HW-SW co-design and reconfigurable computing system concepts to incorporate machine learning as well as multiple standards support in a single device. Proposing an interdisciplinary solution based on the application oriented / domain specific analytical model along with data driven model is an essential investigation to solve these kinds of problems. Another important issue would be how to handle heterogeneous communication standards with efficient reliable communication while maintaining low costs using SDR and cognitive radio concepts. SDR would help us to support multiple standards; CR would help us in maintaining spectral management. One of the proposed solutions to the above problems is that intelligent IoT system design. In this talk we would discuss how to build intelligent IoT, particularly communication and data acquisition issues and their solutions as per the application demand.
*Speakers from Nvidia, IBM and MATLAB will also be joining.