This paper aims to describe an optical flow detection method to be used for navigational tactics similar to those found in natural agents. These tactics are computationally not complex yet lead to useful behaviour, they make use of information generated by a correlation-based optical flow detection and a study of both is made. To be able to gauge the performance of the detection two proxy-tasks are used, one where the agent judges distance of surfaces and one where it tracks its own movement. The experiments are run in a simulated environment to have full control over relevant parameters such as retinal structure, textural properties of the world and others. An important factor is that the algorithm should run on a graphics processor as these are common in contemporary mobile low processing power platforms, such as smartphones, paving the way for a more autonomous agent.
To read the full paper, which was written in 2014 to finish my studies of Cognitive Artificial Intelligence at Utrecht University, just click here.
If you are feeling particularly adventurous you could download the accompanying soure code and compile it by running "make" after making sure you have installed opencv, freeglut and glew on your pc. Running linux or some other posix-compliant-like system will most likely increase your chances of success by a great deal.