Identifying the duplicate questions is challenging because the sentence composition and word selection vary among persons. This project implements an LSTM based architecture that can detect the duplicate question if they have the same intent with an accuracy of 82.65%.
Poster Link: [Link]
Github repo: [Github]
A deep learning model with the combination of Siamese Network and U-Net to detect road surface for autonomous driving from RGB and depth camera image.
Achieved 88% accuracy in the KITTI benchmark dataset. In this project, we collected training images using semantic segmentation, RGB, and depth cameras of the CARLA simulator. This dataset eliminates the requirements of explicitly labeled training images.
Developed a fake TV script generator using Recurrent Neural Network (RNN).
Github repo: [Link]
Analyzed the positive and negative sentiment of the people of Los Angeles regarding COVID-19 from Twitter data using bidirectional LSTM, CNN-LSTM, and Resnet.
Github repo: [Link]
StreetBit app is a part of the Pedestrian Safety research project, conducting at The University of Alabama at Birmingham (UAB), USA. This app communicates with Bluetooth beacons installed in roadside corners and provides timely alerts to distracted pedestrians while they are about to enter into the intersection in a seemingly distracted situation.
I developed the full iOS app and a portion of the android app. Moreover, I also developed the full backend of the system in Amazon EC2.
Poster link: [Poster]
A framework for delivering edge computing platform that can perform calculations of different edge applications in the context of connected autonomous vehicles. Edge computing platforms were powered Nvidia Jetson Nano and Intel Neural Compute Stick 2.
Tested the feasibility of the framework using an object detection algorithm (YOLOv3).
Paper link: [Link]
AVGuard stores all the interactions among different components of autonomous vehicles and different stakeholders such as autonomous vehicle, manufacturing company, insurance company, and cloud service provider into public digital ledger which can help to retrieve the related events of an incident.
The goal of this project is to design and implement a cloud-based storage system. The project contains most of the services of a standard cloud-based storage service such as login, registration, upload, download, share, and delete files in that account.
For the Desktop client, we used Java, AWS java SDK, OkHttp, and JAVA NIO.
For the Web client, we used PHP, MySQL, AmazonEC2, and Amazon RDS.
For Object storage, we used AWS S3 Bucket
Project homepage: Link