In the early years of computer vision, things were very hard. Computers had a little processing power and researchers had to make an effort to get a novel frame grabber with processing capability and a fast specialized CPU. Working in 3D and with full color was mainly a dream. Things have change a lot from those days. Today my most modern TV set is named a smart TV; it can show stereoscopic movies with polarized light that I can see perfectly with glasses with different polarized filters for every eye. The effect is totally amusing and it makes me to remain with a little smile the old days when I tried to program a DCT algorithm in order to compress image sequences in a MPEG similar way using a very slower PC than this that I am using to write this discussion.
There had been a step of quality for scientific research in computer vision due to the modern electronic devices that today are cheaper and more widely spreaded. This can let that a researcher can think about applications and new challenges instead of optimizing the processing time.
Pattern recognition is today something that does not sound odd for us. The economic press is debating the value of the securing system with fingerprints of the new iPhones, in every thriller that we watch at the cinema the forces of law have an extremely fast face recognition software.
I think that the introduction of this technology in our lives has produced a change of the requirements of computer vision science. Today neither it is important to demonstrate than things are possible nor it is important to produce results in real time; today things must be made in a very robust way, and they must provide real value for the society.
However, there are many fields of computer vision science where we will need additional research yet. Low level computer vision has conquered a place at the markets with algorithms for “simple tasks”, but there are other fields where there is a long way to walk. We can recover and process easily 3D data and compare them with a database, however, our capability to interpret a scene is limited. Where computer vision abandon the mathematical algorithms of signal processing and enter the unexplored world artificial intelligence, things are more diffuse. Perhaps they are as diffuse as they were in those days that I have mentioned at the beginning of this paper. Today there are not a lot of real applications widely extended far from the old experts systems and the neural networks based on the multilayer perceptron, although I was working with other different things as Kohonen’s self-organizational maps then.
Recently there have been new advances in electronics for brain-like processing. This can be opening a new gate to the development of the artificial intelligence science and the high level computer vision science. Today, I am more oriented to innovation management than research, and with my current viewpoint I would assure that there will be a lot of business opportunities in this field in the next years. In order to provide value it will be necessary, as usual, to join three aspects in every technology organization: knowledge, funds and risk assuming.
I will start to make a team for this now, because I do not want to lose the train and I would advise to other organizations that they should follow the same way. However, we find always the same barrier, how can we join funds providers with risk assuming for a so challenging task?
We will have an answer in the next years but I am sure that it cannot be found looking at people that do not know what entrepreneurship means.