Research · i2bl Lab, UCLA
Ferrofluid Computer Vision
Real-time vision for closed-loop magnetic control · Nov 2025 – Present
I'm developing a real-time vision system that detects and tracks ferrofluid droplets and magnetic sources from a live camera feed, and goes a step further: it estimates each droplet's volume in microliters from vision alone, then feeds detections back to place the magnetic field and steer ferrofluid across a microfluidic chip.
Computer VisionReal-time InferenceClosed-loop ControlMicrofluidics
The loop
Vision in, magnetic field out
The system runs as a perception-to-actuation loop: see the fluid and the sources, decide where the field should go, move it, and watch the result, continuously.
Sense
Live Camera
microfluidic chip
Segment
CV Model
ferrofluid + sources
Localize
Spatial Detect
positions in frame
Actuate
Field Placement
algorithmic
Result
Trajectory
controlled flow
Demo
Live volume detection
Real footage from the lab rig: droplets are detected, tracked, and measured live, with the estimated volume in microliters drawn on every droplet.
Contributions
What I'm building
- Trained a detection model to localize and track ferrofluid droplets and magnets in live 720p video on an embedded camera pipeline, with per-droplet track IDs.
- Built automatic spatial-scale calibration: the magnet's known surface area locks a px²-to-mm² scale, so measurements survive camera repositioning.
- Combined detection with HSV segmentation inside each bounding box, magnet-exclusion masking, and occlusion handling that freezes the last known reading when a droplet passes under the magnet.
- Estimate droplet volume in µL via an interactive calibration routine: trimmed-median sampling of known volumes (5–25 µL) fitted with a power-law model (v = k·aⁿ), persisted to JSON, with per-track median smoothing for a stable readout.
- Live HUD reports droplet count, total visible volume, and calibration state in real time.
Stack
Tools
PythonOpenCVNumPyObject Detection + TrackingHSV SegmentationVolume CalibrationRaspberry Pi CameraMicrofluidics