Hrishikesh Garud

Developed and implemented Multinomial Logistic Regression and Random Forrest for image classification.

Sidewalk detection to study walkability in the United States of America. Project funded by CDC. Implementing Deep learning in Semantic segmentation and Ensemble learning using Deep learning, Random Forests and Naive Bayes classification.

Real time human object tracking using Convolutional Neural Net and Recurrent Neural Net which incorporates visual attention for faster and better performance.

Computed aspects of affine 3D Geometry of a scene, captured by a still camera sensor, from a single perspective image by using line segment detector and calculated homography projections of the scene by obtaining vanishing points along each of the coordinate axes. Recreated a 3D scene using LabelMe, OpenCV, Python.

Benchmark custom made hardware accelerators by implementing an activity recognition algorithm using Convolutional Neural Net and Recurrent Neural Net.

-- Overall objective: Map a construction site using two aerial and one ground based robotic agent to enable autonomous navigation.
-- Team objective: Implement SLAM on two aerial agents using a visual-inertial approach with the help of Raspberry Pi, BNO055 IMU, Raspberry PiCam.
-- Accomplishments: 3-D point cloud, history pointcloud, real time localization.
-- Software stack: ROS, C++, Python

A comparative study involving development and implementation of the Expectation-Maximization algorithm for the following machine learning classifiers for Face and Non-Face images:
◆Simple Gaussian
◆Mixture of Gaussian
◆T-Distribution
◆Mixture of T-Distribution
◆Factor Analyzer

Graph based segmentation of Foraminifera:
◆ Morphological operations to get finer edge map
◆ Lazy Snapping to classify edge and non-edge pixels
◆ Contour extraction to get number of chambers and the mid-point of chambers

Used deep learning coupled with with recurrent neural networks (GRU cells instead of LSTMs) using visual attention to achieve an accuracy of 78% in producing a LaTex source of a mathematical formula given its PNG image. Technologies used: Convolutional neural net, recurrent neural net, Visual Attention, Python, Tensorflow.

Used machine learning and prediction algorithms to estimate a person’s rate of respiration based on high dimensional data. Algorithms used: PCA, LASSO, Linear Regression using stochastic gradient descent, Artificial Neural Networks, Hidden Markov Model and ARIMA. Technologies used: Python, Keras, Tensorflow

Project aimed at providing ample and accurate data on air pollution levels in Indian cities, particularly Pune. Currently there are only 7 air quality sensor nodes in and around Pune. We aim to install 30 such nodes in pune which will provide us with data that is highly localized for a particular location. Next step will be to upload this data on a website to make this readily available for experts as well as general public. Another aspect of this project involves layering of Google Maps with… Show more

Project aimed at providing ample and accurate data on air pollution levels in Indian cities, particularly Pune. Currently there are only 7 air quality sensor nodes in and around Pune. We aim to install 30 such nodes in pune which will provide us with data that is highly localized for a particular location. Next step will be to upload this data on a website to make this readily available for experts as well as general public. Another aspect of this project involves layering of Google Maps with the pollution data.That is, colour coding of the routes between 2 points to show the pollution in real time.
Technologies used: Flask, JavaScript, HTML, CSS, MySQL, Google APIs, Eagle Design Suite. Show less

This project was in association with BlackBerry and Maharashtra State Police.
This project aims at curbing child trafficking in Maharashtra. I was team lead of the Android App development team. The app is a part of the front end of the project which allowed users to report suspicious activities by uploading photographs to a remote server. The salient features of the app are HTTP Client Server connectivity, MySQL Database Connectivity, JSON encoding, Camera and GPS access and development of… Show more

This project was in association with BlackBerry and Maharashtra State Police.
This project aims at curbing child trafficking in Maharashtra. I was team lead of the Android App development team. The app is a part of the front end of the project which allowed users to report suspicious activities by uploading photographs to a remote server. The salient features of the app are HTTP Client Server connectivity, MySQL Database Connectivity, JSON encoding, Camera and GPS access and development of intuitive GUI. Show less

A patient’s past medical history plays a vital role in prescribing the appropriate medicine, in optimum dosage. There is no existing system which stores the medical records of patients, securely, and make it available to doctors easily. Hence why not build a system that offers this.
By using individual patient RFID cards, accessing the database will be secure and will require patient authorization.
Technologies used: Arduino, RFID tags, Python, MySQL, Eagle board design… Show more

A patient’s past medical history plays a vital role in prescribing the appropriate medicine, in optimum dosage. There is no existing system which stores the medical records of patients, securely, and make it available to doctors easily. Hence why not build a system that offers this.
By using individual patient RFID cards, accessing the database will be secure and will require patient authorization.
Technologies used: Arduino, RFID tags, Python, MySQL, Eagle board design suite.
Resposible for setting up the entire backend system functionality using Arduino for reading the RFID tag ID and Python to enable communication with the MySQL database. Show less