Advancing Defense and Security with Deep Learning-Based Detection and Tracking

Surveillance and defense systems increasingly require intelligent tools to identify and monitor individuals in real time across a wide range of complex and dynamic environments. To address this critical need, an AI-poared framework proposed which is specifically designed to detect and track military, paramilitary, and non-military personnel with high precision and operational efficiency. The system leverages the YOLO model family, specifically its nano and small variants, which offer a balance between detection speed of computation and accuracy, making them ideal for real-time applications. To maintain consistent tracking across video frames, the detection outputs are integrated with four advanced multi-object tracking algorithms: ByteTrack, BoT-SORT, OC-SORT, and DeepSORT. A custom-annotated image dataset, named TriNet, uniquely including many hard-to-find Asian images. This shows real-world diversity, aiding effective model training and evaluation. Experimental results demonstrate strong detection and tracking performance across varied scenarios. Overall, this solution significantly enhances realtime situational awareness, contributing to smarter, more reliable surveillance systems for defense and security sectors.