Developed a project that implements Neural Radiance Fields (NeRF) to generate realistic novel views of 3D scenes. Leveraging deep neural networks, the project models the volumetric scene function by encoding both the density and color of spatial points as functions of their location and viewing direction. This approach enables high-quality, view-dependent rendering of complex scenes, showcasing a state-of-the-art intersection of deep learning and computer vision.

• Developed a Classical Structure from Motion (SfM) pipeline to reconstruct 3D structures from sequences of 2D images. This project integrates key techniques such as feature detection, matching, motion recovery, and 3D reconstruction.

• Utilized essential matrix computation, bundle adjustment, and triangulation methods to accurately estimate 3D points and camera positions, demonstrating the core principles of SfM in computer vision.

Implemented Zhang's camera calibration method from scratch to estimate intrinsic and extrinsic camera parameters. Used SVD, MLE, and DLT for homography computation, intrinsic parameter extraction, and non-linear optimization. The pipeline includes automatic checkerboard corner detection, sub-pixel refinement, lens distortion correction, and bundle adjustment for improved accuracy. Provides visualization tools for assessing calibration quality and image Un-distortions. 

DeepMelody is a project that focuses on generating personalized music recommendations by analyzing the prominence of individual instruments in audio tracks. Using machine learning techniques, the project aims to provide users with tailored music recommendations based on their instrumental preferences.

Autonomous Warehouse management system

• Led the team CON-SOL-E  to achieve the remarkable feat of becoming National finalists, surpassing over 9000  participants.

• Played a pivotal role in developing a sophisticated Warehouse Management System with Autonomous Mobile Robots (AMRs) governed by a centralized navigating system.

• Contributed to the design of AMRs and the development of a vision algorithm for object detection and tracking.

• Utilized fiducial markers (ARUCO Markers) for AMR localization, enhancing the system's accuracy and efficiency.  

• A group of at least 5 ground vehicles or aerial vehicles, with only one control unit. To have coordinated motion by moving together forward, backward, left and right (100m each direction). 

• Awarded 50,000 INR funding from the State Government for further research on Swarm optimization strategies and their implementation in real-world scenarios.

• Faculty Advisor: Prof. Harsh Kapadia Assistant Professor 

• This paper presents the use of the Extended Kalman Filter (EKF) to estimate the states of a moving object based on its measured position in space. By leveraging these estimated states, the EKF enables the prediction of the object's optimal trajectory using a motion model specifically defined for Kalman filtering.

• The project was completed with the valuable support and coordination of my classmate, Yash Battul.

Using Fiducial Marker