My Projects
Robotics
OpenCV
Brain Computer Interfacing
Motor Imagery based classification of EEG data
Executed a project on Brain Computer Interfacing (BCI) at IIT Delhi in which the EEG signal recorded from the scalp of a subject was acquired and processed using signal processing, machine learning and other techniques to control a virtual scenario (moving a ball in a scene in this case) based on the motor Imagery data from a subject. I designed and coded the simulation program in pygame which displays Right and Left arrow at short random intervals. I also recruited and prepared subjects for EEG recording, performed the setup for scalp preparation like setting the EEG cap, electrode attachment and impedence checking, etc. I trained the model from the data in MATLAB using EEGLAB and BCILAB. The subjects were asked to think about moving their left or right hand based on the arrow on the screen. The EEG was recorded during these tasks, which was then filtered and processed and trained upon to build a model
Deep Learning
Reinforcement learning
Built a Deep-reinforcement learning system that learns to play the catapult game using OpenAI gym and deep neural Network
GAN (Generative Adversarial Network)
A GAN that generates images learned over mnist and fashion_mnist dataset
Music Generation
A neural network that can generate music on its own using LSTM. Built 2 models, one that generates music in abc format and other that generates in wav format
Anime vs real
A neural network that can classify real photographs and drawn images
Sign lang mnist
A neural network to classify images containing pictures of hand making signs from sign_lang_mnist
BBC text class
A neural network for classification of BBC text
Pneumonia
A neural network that can detect cases of pneumonia from a chest X-ray
Imdb reviews
A DNN that classifies a given imdb review as positive or negative
Sentiment in tweets
A neural network to classify tweets for sentiments
Poetry generator
A neural network that automatically generates poetry similar to Shakespeare's poems
Chatterbox
A chat box which you can chat with. It work based on a deep neural network and can post replies to your posts.
Fashion mnist
A DNN to classify images containing fashion items from the fashion mnist database
Handwriting recognition
A DNN to recognize handwritten text
Human or horse
A DNN to classify a given image as human or horse.
Defected vs. healthy
A DNN that classifies a given image as that of a healthy or defected artifact
Trigger Word Detection
Constructed a speech dataset and implemented an algorithm for trigger word detection (also called keyword detection, or wakeword detection) that allows a system to wake up upon hearing a certain word
Neural Machine Translation
Built a Neural Machine Translation (NMT) model to translate human readable dates (“25th of June, 2009”) into machine readable dates (“2009-06-25”) using an attention model
Emojify
Implemented a model which inputs a sentence (such as “Let’s go see the baseball game tonight!”) and finds the most appropriate emoji to be used with this sentence using word vector representations
Music Generation with LSTM
Implemented a model using LSTM to generate Jazz music
Dinosaur Island – Character level language model
Built a character level language model to randomly generate new dinosaur names based on names learned from a dataset of dinosaur names
Face Detection with CNN
Implemented triplet loss function and performed face verification and face recognition using 128D encodings generated with a pre-trained ConvNet model
Neural Style Transfer – Art Generation
Implemented Neural style transfer algorithm to generate novel artistic images using the algorithm by Gatys et al
Cat vs Non-Cat classification using DNN
Constructed a deep neural network to classify images as cat and non-cat
Car Detection with YOLOv2
Constructed CNN to perform object detection on a car dataset using YOLO model as well as dealing with bounding boxes to filter the most fitting box
Other
Open3D
An Api for Open3D which returns a list of identically colored connected components grouped together based on their color in a mesh of connected vertices.
Telegram
-----Muxxer
An Api written in python that collects messages and posts from multiple channels and converges them to one single channel. A very useful api in cases where a user has to join large number of channels. This api removes the need of joining multiple channels while bringing the content at one place.
-----Channelmuxbot
A telegram bot that helps Muxxer by adding and removing channels in a list of channels. It provides a clean interface to the user to make it easy to add/remove channels.
Sound Localization
Designed and performed a complete experiment to study the degree to which people can localize a source of sound among normal, blind, and patients treated for blindness.
Isolated Digit Recognition using HMMs
Designed a speech recognition system using HMMs that can recognize digits from 0 to 9 spoken by the user.
Automatic collision prevention system
Designed a system that prevents collision of vehicles by monitoring other vehicles on road relative to the user’s vehicle with the help of multiple sensors mounted on the vehicle.
Thermal Video Stabilization
To efficiently estimate and correct motion in a thermal video using Harris Corner detector and Lucas-Kanade method.
Biometric door locking and unlocking system
Designed a system to lock and unlock using finger print scanner and microcontroller with the facility of multiple finger print recognitions.
Automatic switching of lights
Designed a circuit that switches on the lights of a room when a person enters the room and switches them off when the room is empty. It also counts the number of people present in the room at a time.
Alcohol detector based ignition control system
Designed a control system that controls the ignition of a vehicle based on the alcohol level detected. If the driver is drunk, the system will prevent the vehicle from starting. It was designed for the “TI MCU DESIGN CONTEST 2012”.
Analysis of behavior of Air Horn (Acoustics)
Analyzed the behavior of an air horn and its contribution in improving the performance of a driver including how an airhorn is able to produce high levels of amplitude without active amplification of the sound. Also studied different shapes of air-horn and which shape results in most efficient output and why.
Quadrature Amplitude Multiplexer
Designed a circuit to successfully transmit and receive two message signals by using Quadrature Amplitude Multiplexing. The circuit was constructed on TI Analog Design kit.