Week 2 — Warmth of Image
Title: Weather Condition Prediction from Image
Team Members: Berk GÜLAY, Samet KALKAN, Mert SÜRÜCÜOĞLU
E-mails Respectively: berkgulay.cs@gmail.com , abdulsametkalkan@gmail.com , mertsurucuogluu@gmail.com
In previous post, we mentioned about some related works and datasets. This week, we did research about those studies and datasets in details.
Our biggest problem was collecting data. We took advice from our lecturer to collect data. He gave us a AMOS site. AMOS is a collection of long-term timelapse imagery from publicly accessible outdoor webcams around the world. There are images of millions and webcam of thousands. But we didn’t enter to the site. Because there is a problem in the site.
We looked for other ways to collect data. We mailed some people and institutions which have done work like ours to share their datasets with us but they even didn’t answer us. As a result of our lengthy research, we found a dataset that has been classified. Images that we found have 5 classes: cloudy, foggy, rain, snow and sunny. There are approximately 150.000 images and this is enough for our work.
There are some sample images below:
Also we found some related work previous week. We investigated those works in detail: How they did, which features used, which algorithm used etc.
In general, they used neural network and support vector machine(SVM). Also they used hog feature, color histogram, contrast, sharpness as feature.
We decided to use above methods and features. Next week we are going to start code it.
See you next week.
