Shree Singhal
Back to projects

Project

Drone Tracking In-Air Laser Charging Capstone

  • Computer Vision
  • GroundingDINO
  • CoTracker3
  • PID
  • Capstone

A vision-guided turret that tracks a drone in flight and keeps a laser pointed at an onboard photodiode for wireless power transfer. The system fuses three CV models with a closed-loop PID actuator to maintain alignment in real time.

Overview

Three-layer computer vision pipeline combining GroundingDINO bounding boxes, segmentation masks, and CoTracker3 point tracking to keep focus on an onboard photodiode target. Drives a closed-loop PID turret actuation system that aligns laser placement with live tracking coordinates.

What I worked on

I built a three-layer CV pipeline: GroundingDINO produces bounding boxes from natural-language prompts, a segmentation pass refines the drone silhouette, and CoTracker3 maintains stable point-level tracking on the photodiode itself. The tracking coordinates feed a PID controller driving two-axis turret actuation, so the laser stays on-target even as the drone moves.

What I learned

Real-time CV is a latency problem first and a model-quality problem second. Most of the engineering effort went into shrinking the pipeline so the turret loop ran fast enough to be useful, and into deciding which layer of the stack should be the authoritative source of position at each moment.