← All projects

2022

Fog Classification of Trail-Cam Footage in Hawai'i

I worked on a machine learning project that classifies trail-camera images as foggy or clear for climate research workflows in Hawaii.

  • Machine Learning
  • Artificial Intelligence
  • Climate Research
  • Blender
Trail camera machine learning project screenshot

Overview

I worked with a partner to classify trail-camera footage as foggy or clear. We focused on long-term climate tracking by helping people sort large image collections without relying on expensive fog sensors.

Synthetic data

I used Blender to generate more than 100 foggy and clear scenes with physically based path tracing. Those synthetic images became training data for a model tested against real-world images.

Results

I trained a logistic regression model on Blender-generated imagery that reached an average AUROC of 0.74 on real-world test data, showing that synthetic data could capture useful visual features of foggy conditions.

Gallery