Voxel Reconstruction (C++)

The master course Computer Vision consisted of several projects using OpenCV, one of which is the project described here. For this project, I worked together with one other student. The goal was to construct a voxel model of the foreground.

As our input, we had four short videos taken from different angles, both with and without the foreground model. After calibrating the four cameras, the foreground was separated from the background in the video containing the model. The first step to this was to create a background image by averaging the pixel values of each frame of the background video. The absolute difference between the background image and main video file was calculated and tested against a threshold based on the standard deviation within the pixels to determine whether a certain pixel belongs to the foreground or to the background. Erosion and dilation were used to improve the foreground image.

Finally, the voxel model was constructed by translating the pixel coordinates to world coordinates.