Rajat Sharma

The aim of this thesis to regress bounding boxes for object detection using distance regression from deep learning models in Autonomous driving.

In this work my task includes Explore Carla driving simulator Create requirements specific synthetic datasets (RGB, LIDAR, DEPTH) Train training deep neural networks on these datasets and do evaluation.

Research study in high quality surface reconstruction using deep learning. Review of different Deep Learning approaches based on volumetric fusion for making high-resolution TSDF volume. Review of up-sampling point clouds techniques based on deep learning methods. Development of a new approach for Efficient Point Cloud Up-sampling and Normal Estimation using Deep Learning for Robust Surface Reconstruction (Also publishing this work as part of conference).

Developed 3D Reconstruction pipeline using android app service requiring uploading data, processing and returning back to android app in real time on Azure Cloud Platform. Azure stack used as follow: Azure VMs , Azure Storage (Containers, Files), Azure Function Apps (for azure storage triggering), Azure Notification Hubs (real-time communication with android App), Azure Batch Service. Develop Android prototype for uploading data to azure storage, and real time communication.

Working as a part-time student assistant (40 hours per month). Support for software development for interactive virtual experiments. Achievement highlights: Development of backend using python and Django-channels for real-time application. Checking for invalid experiments state charts using depth-first search. Adding replay mode functionality to view the already done experiment in future with play, pause, fast-forward functionality.

1. Develop and write software codes for web backend in Python, PSQL and frontend in JavaScript. 2. Handle Inter process communication between C++ server and web server using Apache Thrift and grpc. 3. Handle real time communication between web server and browsers using web sockets. 4. Adding new features to the product.