Ramya (Girish) Balasubramaniam

Developed a computer vision-based application that detects and extracts cars in an image and classifies them according to make & model and body type (sedan, cross-over, etc.) using a faster r-CNN a combination of Regional Proposal Network (RPN) and Convolutional Neural Network (CNN).

A novel effort proposed to eliminate the pilferage of food during it's transportation using the Global Positioning System. Made a Video presentation supporting possibility of eliminating hunger by improving food security and minimizing wastage using food surveillance.
We won 3rd place in Envisioning 2020, Imagine Cup 2010

Developed convolution neural network (CNN) based political-bias identification and ideology classification system
Highlights:
1. Trained on huge dataset labeled using weak supervision unlike conventional annotation
by human experts, avoiding unnecessary financial and time-related commitments.
2. Clustering method (k-means) was used to group words based on political ideology
3. More than twenty different hyperparameters were tuned to get high accuracy scores

‘Multimodal Biometric System’ using the Iris Recognition and Speaker Identification. The first phase and the second phase of the project has already been implemented. The project includes the process of iris recognition using the Db-3 and Db-4 wavelets for encoding, Daugman’s Algorithm for segmentation and pitch period estimation of speech signal using wavelets coding. The Biometric system combines the decision taken by the Iris Recognition System and the Speaker Identification System by using… Show more

‘Multimodal Biometric System’ using the Iris Recognition and Speaker Identification. The first phase and the second phase of the project has already been implemented. The project includes the process of iris recognition using the Db-3 and Db-4 wavelets for encoding, Daugman’s Algorithm for segmentation and pitch period estimation of speech signal using wavelets coding. The Biometric system combines the decision taken by the Iris Recognition System and the Speaker Identification System by using the Decision Theory.

Selected as the best project of the University for the year 2009 Show less

Performed ‘Simulation of the Wireless Channel’ using the Jake’s model and developed a model and coded the same on the MATLAB. For this purpose, random numbers belonging to different distributions were studied in detail and the same was generated with self-written codes, the relationship between PSD and Autocorrelation function was studied along with the relationship between the white and the colored noise, subsequent to this the PSD of the colored noise was generated according to the Jake’s… Show more

Performed ‘Simulation of the Wireless Channel’ using the Jake’s model and developed a model and coded the same on the MATLAB. For this purpose, random numbers belonging to different distributions were studied in detail and the same was generated with self-written codes, the relationship between PSD and Autocorrelation function was studied along with the relationship between the white and the colored noise, subsequent to this the PSD of the colored noise was generated according to the Jake’s model. A ‘SISO model’ was created, which can be extended into a MIMO model. Show less

Worked on detection of quadratic phase coupling between line frequencies and for estimation of their coupling strengths from acoustic signals. Project was done in collaboration with the IRD (Industrial Research and Development Department) for which honorarium was awarded by IIT Delhi. Analysed Wav File, generated data using C code. Found to be twice as efficient as MATLAB code