AI image analysis and prediction to improve the weather news prediction accuracy:
Funding Agency: Weathernews Inc., Tokyo, Japan
Me & my research group at Visual Learning and Intelligence Laboratory has used AI image analysis and prediction to improve the weather news prediction accuracy. A collection of advanced preprocessing algorithms and AI models are developed for analysing the images in order to forecast the weather conditions.
Following algorithms and models have been developed and carried out the technology transfer:
Precipitation Nowcasting
Volcanic Eruption Classification – Ash Detection
Cloud segmentation on real-time RGB/Infrared data using deep learning techniques.
Road Scene Vehicle Detection in adverse weather conditions
Image Dehazing for road scene analysis tasks such as vehicle detection, vehicle classification and road line detection
Design and Development of Real-Time Transportation Safety Monitoring System for Smart Cities
Executive Summary:
The project aimed at the development of methods for analyzing traffic flows especially in the crowded scenario, methodologies to analyze various anomalous events like accidents, snatch thefts, and violence during religious processions, the design of machine learning models and the design of techniques for representing such anomalous events. The software is developed which works in real-time on surveillance scenarios. The deep learning- based models have been introduced to determine traffic violators (helmetless driving, rushing at stop signals, wrong side driving, illegal turns, etc.) and deep learning-based methods for recognition of human poses in various surveillance scenarios for person re- identification are the significant works competed as a part of this collaborative project.
A real-time and scalable system for person re- identification that can identify potential anti- social elements and track their movements is an off-shoot of this collaborative project.
Salient Research Achievements:
• Detection of anomalous events:
Gaussian mixture model (GMM) is used to form a universal attribute model consisting of multiple actions to identify relevant attributes, also called action vectors. They contain actions of anomalous activity. For snatch theft detection, we have achieved over 99% classification accuracy. Due to the dynamic nature of this feature representation, this approach can also be used for other anomalous actions such as accident detection.
• Traffic violation detection (motorcyclists without a helmet):
Deep learning approach such as a convolutional neural network (CNN) to identify motorcyclists in dense traffic videos is explored in this work. After identification of motorcyclist, another CNN is used to detect head region to classify among motorcyclists with and without a helmet with over 90% classification accuracy at 52 ms/frame.
• Person re-identification in surveillance videos:
Using deep features (VGG16), a graph is constructed in such a way where each person is a node in which the edge between these nodes is calculated from a similarity measure to find the closest k neighbors of each node (person). A graph kernel is then used to classify among multiple persons.
This is a collaborative project work between Indian Institute of Technology, Hyderabad, India and University of Tokyo, Japan.
Funding agency: DST-JSPS
Duration: June, 2018- Mar. 2020.
On the Indian side:
Principal Investigator: Prof. C Krishna Mohan, IIT Hyderabad, India.
Co-Investigator: Prof. B H Shekar, Mangalore University, Mangalore, Karnataka, India.
On the Japan side: Dr. Masaki Ito, University of Tokyo, Japan.
Some memories down the lane of this Project
Research Diary
Continued…
Research Diary
Prof. C. Krishna Mohan
Dean (Public & Corporate Relations) &
Professor, Department of Computer Science
& Engineering, IIT Hyderabad
Highly energy-efficient LED technology contributes to saving the Earth’s resources from 20% to 40% compared to conventional technology as about one-fourth of world electricity consumption is used for lighting purposes. Nevertheless, the current LED technology is limited with selected color options. Indeed, with current technology, we can’t achieve the “True Color” from a single material. Our target is to tune the light emission through the nanoscale molecular aggregations.
Therefore, the research collaboration has been established between our Organometallics Lab at IIT Hyderabad and Prof. Dr. Osamu Tsutsumi’s lab at Applied Chemistry, Ritsumeikan University, Japan, through the generous support of the JICA Friendship Project in November 2015. The purpose of our interdisciplinary research collaboration is to advance fundamental understanding to solve problems at the nanoscale molecular aggregation level.
We have been working on this cutting-edge problem to identify the single source material to emit blue light or even direct white light using organo gold precursor or equivalent materials. One of the most challenging tasks in this research area is to identify the blue or white emitting system without chromophore’s contribution. As a result, the n-alkyl chain stimulated paradigm luminescence shift with unique emission in N-heterocyclic carbene gold(I) chloride complexes have been demonstrated for the first time in the crystalline phase. The research collaboration has been further strengthened through the JICA PhD, RU JASSO internship, and RU Post-Doc fellowships.
Four M1 students Mr. Shinya Nakamura, Mr.
Masaya Yamane, Mr. Shohei Sugiyama, and Mr.
Ozaki Kazuhisa, were invited through JICA from RU, Japan, to work in Organometallics Lab.
Similarly, Dr. Katam Srinivas, Dr. Vaddamanu
Moulali, and Mr. Nandeshwar Muneshwar Giridhar were visited the RU for multiple times to work on this research problem. Through this intensive collaboration, four master students from RU and one post-Doc from IITH were trained. Besides, Mr. Kumar Siddhant, a master's student from Organometallics Lab, has been selected under the JICA Ph.D. program to investigate this research problem and he has been pursuing his Ph.D. in RU. The outcome of this collaboration includes the additional research collaboration with Prof. Shigeyuki Yamada’s lab, KIT, Kyoto, Prof. Nobuyuki Mase’s Lab, Shizuoka University, Hamamatsu along with several lectures, several conference presentations, several publications in very high impact peer revived journals, the Journal cover page highlight in America Chemical Society, the Best Poster Award, and the workshop on Photo functional Gold Molecules and Nano Materials 2019 (PGMNM2019).
Some throwback moments exchanged between the two Research Group