I show this result first because 9 of 10 persons see an axe. If you see an axe in a head You're normal in some standards. This disparity map is computed for this left image:
Those result are enough for obstacle tracking. The algorithm was not fully described and that's why i try to obtain better results by randomly changing the parameters. this algorithm is described in the internet and was made by Dong-Hu Nie, Kyu-Phil Han and Heng-Suk Lee. I haven't received any code from the creators(even a email :) ), i have reconstructed it from the paper. There's many secret parameter's but it's a real fun finding them out step by step. But i want it to be more detailed in a controlled environment. That will give me more accuracy in the real world. When i will have noises and other weird things that i haven't considered yet. This genetic algorithm is ready for GPU coding but i have revived a great help from one in a kind stereo vision specialist: Xuefeng Chang. He wrote a really mind blowing real time stereo vision algorithm that i am currently trying to understand and code by myself, mayby i'll be able to enhance the speed. but i need a PC that can handle OpenCL, that's why for now i have to pause my work and find some work that will generate enough funds for a nice and quick nvidia graphics card. I need also funds for the PRE-Flight simulator. After gathering that money i will keep You updated on that page.
