Swarm-based Object Tracking UAV Navigation System in Indoor Environments

This project focuses on developing a software algorithm for object tracking using unmanned aerial vehicles (UAVs) in indoor environments. The aim is to enable drones to detect, segment, estimate depth, and track dynamic objects—specifically humans and cars—while maintaining a consistent distance through automated controls. The system integrates state-of-the-art deep learning models, including YOLOv5 for object detection, Mask R-CNN and U-Net via Detectron2 for segmentation, and MiDaS for monocular depth estimation. Object tracking is implemented using a combination of DeepSORT and geometric techniques, supplemented by PID control for drone navigation. Data is prepared via web-scraping and manual image collection, with preprocessing and annotation using tools like Roboflow and PyTorch. The final output aims to provide a robust, real-time tracking mechanism suitable for constrained indoor drone navigation, demonstrating promising results across detection accuracy, depth estimation, and tracking reliability.