Undergrad Research Project - Masking vegetation and sky in images for improved 3D reconstruction

Spring 2017

Elizabeth Yan
John Dolan
Project description

The goal of this project is to detect and mask specific areas like vegetation and sky and moving objects like vehicles and people in a set of images to improve the 3D reconstruction of other objects. This is a continuation of the Fall 2016 project, the addition for this semester is detecting and removing moving objects like vehicles and humans. Programs like VisualSFM are able to make a 3D model of an object from a set of images that have been taken from different viewpoints. The Navlab group has used VisualSFM to make 3D models of accident scenes. The program works best when the scene is static. Parts of the scene that change, like clouds and vegetation moved by wind or pedestrian and vehicles moving around, can cause significant problems. It is therefore desirable to detect these areas and objects in the images and mask them. Lisa already learned how to use the 3D reconstruction methods used by the Navlab group. She also knows the basics of running a texture detection program and masking out areas. She will improve the texture detection to find areas of sky and vegetation. She will learn to use human and vehicle detection methods to be able to identify the location of those objects in the image and mask them out. The detection and masking algorithm should be validated by testing it on several of our accident data sets and other data sets.

Return to project list