BTech Student Rishabh Bhattacharya Wins Indian Navy Tech Competition

At the 'Swavalamban 2024' seminar, an event dedicated to innovation and self-reliance, the Indian Navy announced a competition that challenged individuals to create technological innovations aimed at solving operational issues. The contest highlighted several areas for improvement, such as an app for load balancing, a method for drone swarm coordination without central oversight, enhancements in maritime awareness, and AI-driven speech separation in conversations, besides real-time tracking of aerial objects. Rishabh Bhattacharya, a dedicated scholar pursuing a dual degree in BTech and Master of Science in Computer Science and Engineering by Research, found his interest piqued by the task of enhancing real-time tracking and navigation for flying entities. Drawing from his research showcased at the IEEE International Conference on Robotics and Automation (ICRA) 2023, Bhattacharya set his sights on developing an algorithm capable of optical flow tracking with unmatched precision. His project aimed to tackle the variances in lighting, swift movements, and intricate textures, proving particularly beneficial for drone technologies or embedded systems application.

The significance of his work lies in the ability of autonomous systems, like drones, to navigate swiftly changing environments and dodge obstacles seamlessly. Optical flow, a technique in computer vision, plays a crucial role in understanding object motion through pixel analysis across successive images. The goal to reach sub-pixel accuracy presents a steep challenge, given the unpredictable and rapid movements of flying objects, necessitating sophisticated detection and tracking tools for real-time functionality.

To bolster the effectiveness of his algorithm, Bhattacharya amalgamated the Flying Objects dataset from Sekilab with a UAV dataset sourced from Kaggle, alongside a synthetic dataset he developed. This rich combination aimed to provide a thorough grounding for training and validating the algorithm's optical flow tracking prowess. The announcement of this dataset's future public availability marks a step forward in this research domain.

Delving deeper into the technicalities, Bhattacharya introduced a novel approach to enhance the algorithm's object detection feature. Leveraging a technique from his research paper titled "GDIP: Gated Differential Image Processing for Object Detection in Adverse Conditions," he aimed to boost performance in difficult scenarios such as foggy or dimly lit environments. This methodology was incorporated into an advanced YOLOv8 model, which underwent training over 50 epochs with the combined dataset. The outcome demonstrated remarkable efficiency and accuracy, making it suitable for real-time application even under challenging conditions. During the innovation marathon, Bhattacharya had the chance to present his winning solution to Navy officials, including admirals and commanders, who showed keen interest in applying this technology within Navy operations. Bhattacharya reflected on how his involvement in the Machine Learning Lab and his prior research laid the groundwork for this solution. He emphasized the value of interdisciplinary learning and its application in solving practical challenges, highlighting the profound impact such exposure can have on technological advancements.