The next 3-5 years of computer vision

From 1966 to 2016, exactly the past 50 years, the last 50 years of computer vision development is very fast. Does computer vision become a very mature and perfect technology today? Not.

For example, under the blue sky, the highway has a big white truck, the computer may say, this is a white cloud. Everyone might have thought about it. This is a tragedy: a Tesla has not detected the truck, so that the high-speed Tesla driver died on the spot.

Although computer vision technology has not developed into a mature stage, but as the application market opens, new opportunities and technological innovations will follow.

In the next 3-5 years, there will be three changes in the field of computer vision that deserve attention

As a practitioner of the computer vision industry, tell me about my personal views on the next 3-5 years.

First of all, there are many open source software packages for computer vision, but as the problem of vision becomes more and more complex, and our demands for security become more and more high, complex problems must be solved by a professional team. For example, to solve the problem of camera motion trajectory, if you take a mobile phone from one room to another room, and then back to the original position, generated a trajectory. This data to the Orb slam analysis, the trajectory of it has changed, and the wall has a great overlap. If a robot uses this Orb Slam project, it will bump into a wall when going out.

Even now there are many open source projects, but the commercial computer vision system can be tested and compared in different platforms and different environments, which can achieve better average performance and no significant security risks. Complex problems and applications, we must find a professional team to solve. More than 1 billion next-generation computing platforms, a new generation of smartphones and drones, will emerge in this field. With this corresponding technology has tens of billions of of the market, the field of computer vision will be born a lot of listed enterprises.

The second change of concern is the chip. We know that computer vision often requires very complex algorithms to solve, applications are often mobile, such as mobile devices, mobile robots. In mobile devices, the chip must be the only way to carry out complex algorithms in a low-energy manner. Now in the industry, a large number of teams in the research and development of algorithms, some of the faster peers have begun to mature algorithm chip. In this way, in addition to the superiority of the algorithm, there must be a chip power consumption and cost issues.

The third change of concern is theoretical, that is, what kind of mathematical model will be produced after deep learning. The disadvantage of deep learning is also known, it needs to carry out a huge amount of information learning. Once the authority of this field, Davis Marr, predicted that the computer was a complex problem and that there would be a single theoretical framework to solve it.

Deep learning will not be the ultimate framework, and there will be more powerful mathematical models to emerge. It will also be more profound to the human influence, we should be deeply concerned about the theoretical changes, the latest technology into products to solve the actual problem.

This article first titanium media, by Zhang Lin according to Baoyingze in the 2016MIIC Conference of Speech finishing

The next 3-5 years of computer vision