Abstract
|
|
This project studied traffic cam image colorization and weather conditions to determine whether correlations exist. The goal was to construct a model which could predict color characteristics of images based on the current weather condition as reported by the National Weather Service.
|
Introduction
|
|
A model which could predict the expected colorization of webcam images based on weather conditions could aid in image recognition by providing a better base-line from which to measure events. This could allow for consistant accuracy under a variety of weather conditions.
|
Methods
|
|
This study gathered webcam images and weather condition data from around the Pittsburgh area for a two week period. Images from 3pm on each day were used, and their respective levels of red, green, and blue, normalized by the average for the given camera, stored in a database. Catagorical weather condition data was used as the explanatory variable for models to attempt the prediction of red, green, and blue average and standard deviation.
|
Results
|
|
One-way ANOVA tests were conducted for each of the response variables. It was found that over 23% of the variation in the average amount of red in an image could be explained by the weather condition. It was also found that 22% of the variation in amounts of blue could be modeled. No significant model could be created to predict levels of green.
|
Privacy
|
|
The privacy implications of this study lie first in the existence of such webcams in public places. Drivers are certainly not consulted as to whether images of them may be broadcast, nor are they even warned that these webcams exist. That it appears possible to construct models based on weather data to predict image colorization, signals the increasing accuracy of automated recognition of items and events taking place in images. This leads to questions of surveillance by not only the government, but any party.
|
References
|
|
Camera: https://www.nb.net/~finals/allcams.htm
Project Paper (PDF)
|
