To conclude by Bachelor Computer Science at the University of Leiden I wrote a bachelor thesis. I chose a topic very relevant to robotics, monocular depth perception. The task is to take a single image and, for every pixel, calculate the distance of the object to the camera. This is ofcourse a very difficult task, we as humans usually use both our eyes to understand depth through disparity.

At the time of writing of this thesis, the state of the art used deep convolutional neural networks. My thesis uses multiple of these deep convnets and fuses them using different fusion methods, e.g., average, median and weighted average. These methods are studied to understand if they result in an increase in depth perception performance.

The PDF below provides a more comprehensive overview of my thesis. It can also be downloaded here and the code can be viewed on my GitHub here.