Academic Innovations

An end-to-end computer vision pipeline for real-time detection and tracking of small objects such as drones, even when object visibility is degraded due to poor illumination, small object size, or adverse weather (fog, rain, haze, low light). The system uses deep neural networks and vision transformers to detect and track small objects in aerial imagery under challenging conditions. It includes modules for image enhancement, atmospheric correction, and multi-scale target detection using attention-based CNNs. A Kalman filter and optical flow tracker ensure stability during occlusions or motion blur. The system processes real-time data on embedded GPU platforms and fuses multi-sensor inputs for improved accuracy. Its modular design allows deployment on UAVs or ground stations, supporting applications in defense surveillance, maritime safety, and environmental monitoring with high precision and efficiency.