Bio-Inspired Computer Vision: BAYESIAN INFERENCE OF MODELS ON EVENT BASED OPTICAL FLOW TO SOLVE APERTURE PROBLEM

This project is one part of the module: Bio-inspired computer vision and was implemented by Florian Jäger and Weijie Qi under Prof. Dr. Marianne Maertens and Prof. Guillermo Gallego.

The folder contains the following:

-Event segmentation folder

In this part we try to split the motion to solve the aperture problem with two objects, based on the data generated in the optical flow part using event segmentation.

-Optical flow folder:

This part of the project includes an improved version and improved understandability of the optical flow. Please run this part of the project first so that you understand what data is used the Event Segmentation part

-Slides:

The presentaion of the results of the project

What we want to solve:

-When there are two moving objects, particularly when they have overlapping part in the space, how to solve the second aperture problem?

Our solution:

-We want to use the normalized and unnormalized velocities to sum up the detected velocity.

-We put an alpha matte towards all pixels and model each pixel as a combined effect from the object 1 and object 2.