In a remarkable achievement for the Indian academic community, Rishabh Bhattacharya, a third-year student at the International Institute of Information Technology (IIIT) in Hyderabad, has secured first prize at a prestigious Indian Navy event. His innovative algorithm, designed to enhance navigation and real-time tracking of aerial objects such as drones, has earned him a cash award of Rs 3 lakh. The accolade was announced during Swavalamban 2024, an event focused on innovation and indigenisation organized by the Indian Navy.
Bhattacharya’s optical flow tracking algorithm has been praised for its ability to achieve sub-pixel accuracy, a critical feature that allows for precise motion estimation and tracking. The algorithm also effectively navigates environmental challenges such as poor lighting conditions, rapid object movements, and intricate textures.
The Swavalamban seminar, held in October, featured a nationwide competition aimed at addressing various operational challenges through technological innovation. Participants were invited to tackle problem statements that encompassed a range of topics, including swarm drone coordination and maritime situational awareness, with Bhattacharya choosing to focus on the navigation and tracking of flying objects.
Drawing inspiration from his prior research presented at the IEEE International Conference on Robotics and Automation (ICRA) 2023, Bhattacharya emphasized the importance of resilience in technological solutions. “One of the criteria laid out was for the solution to demonstrate resilience to varying lighting conditions, rapid movements, and complex textures while maintaining efficiency on platforms like drones or embedded systems,” he said.
Creating an algorithm capable of functioning effectively under these constraints was no easy feat, particularly given the inherent unpredictability of flying objects. Bhattacharya explained the complexities involved in tracking such objects necessitated sophisticated detection and tracking mechanisms that could operate in real-time.
To overcome a lack of comprehensive datasets, he innovatively merged a flying objects dataset from Sekilab—containing planes, helicopters, and birds—with a user-generated UAV dataset from the platform Kaggle. By employing semantic separation techniques, he generated a synthetic dataset that could simulate varied motion scenarios, making the dataset versatile and rich for training purposes.
The combined dataset, totaling 7.7 gigabytes, is set to be released publicly, providing a valuable resource for the wider research community. In a bid to enhance algorithm performance under challenging environmental conditions, Bhattacharya integrated a framework he had developed earlier, known as Gated Differential Image Processing (GDIP). This framework optimizes object detection models, such as YOLOv8, making them more adept at operating in low-visibility situations like foggy weather.
The model underwent training with the comprehensive dataset over 50 epochs and was fine-tuned for real-time applications, achieving a processing speed of approximately two milliseconds per frame. Rigorous testing confirmed its reliability in varying lighting conditions, complex textures, and unpredictable movements.
Bhattacharya credits his success in part to his experiences at the Machine Learning Lab, where he worked under the guidance of Dr. Naresh Manwani. He recalled engaging discussions on research papers that helped shape his final solution. The seminar provided an invaluable opportunity for Bhattacharya to present his work directly to Navy admirals and commanders, who expressed genuine interest in potentially integrating his solution into operational frameworks.
“Meeting Navy officials who appreciated and discussed my work was an inspiring moment,” Bhattacharya remarked, reflecting on the significance of his accomplishment and the potential impact of his research on future maritime operations.
