This was a final project for my course in Computer Vision (CS 283, Fall 2013).
Title: Segmentation and Identication of Traversable Surfaces from Aerial Images for UGV Guidance and Support
The identification of traversable terrain from aerial images is essential for the development of autonomous robotic networks capable of operating over a wide area within a rapidly evolving environment. We propose the usage of SVM classifiers in order to identify roads and buildings within an urban environment and use the information to provide pathfinding for an agent on the ground. Such an approach works very well when utilizing Gabor filters in conjunction with color information to build a feature vector for each pixel within an image. We find that our pathfinding algorithm works well at identifying and using traversable terrain, even identifying side roads that were not previously registered in the database, while still avoiding buildings. We envision that our platform could be used as a foundation for the development of UAV-UGV networks acting outside of direct human control.
Project report can be viewed here.