Face to Face
Recognition, just like 3-D, has long been a favorite Russian direction in technology. The majority of successful projects at Russian computer companies have involved one of these two areas, and sometimes both.
Sep 04, 2004
Recognition, just like 3-D, has long been a favorite Russian direction in technology. The majority of successful projects at Russian computer companies have involved one of these two areas, and sometimes both. Abbyy, Parallel Graphics, Cognitive, Real-Time Geometry, Neurok are just a few examples, and even Russian-created anti-virus programs use the same basic algorithms as 3-D technology.
Barebones technology
The story of A4Vision (full name: Application for Vision) begins back in 1998. Two of its founders, recent graduates from the Bauman Moscow State Technical University Artem Yukhin and Andrei Klimov, created a device described in its patent as a zero-contact 3-D optical scanner, or in other words, a 3-D camera. This device was initially built to make machines see, such as independently operating robots moving across an uneven surface. When the year 2000 arrived and the internet boom was in full swing, they like many other young developers at the time began to seek investors. However, an investor found them first. The first million dollar investment came to A4Vision the following year, in 2001.
At first, A4Vision did not have any commercial product to its credit, only barebones technology. They, along with their venture investors, had to come up with something to sell. Initially they planned to create 3-D images of products in online stores based on A4Vision's technology. This proved unfeasible and the idea quickly died. Then the company began to work on software for cosmetic surgeons that modeled patient's outward appearance after the operation. This idea might have been successful, but then came September 11 and a dramatic jump in the demand for security technologies. The company immediately knew what it had to do. Their cosmetic ideas were easily converted into a way to fight terrorism via physical recognition. This also happened to be a problem Yukhin had dealt with before, in his MA thesis.
Security technology, as the term is currently understood, includes everything from protecting computers from hackers to detecting bacteria. This broad category has many smaller divisions and sub-divisions. A4Vision focuses on biometric and identification systems. There are currently three main ID methods: fingerprints, the iris, and the face, which is A4Vision's specialty. These aren't the only ID methods out there; there are many, far more exotic approaches. For instance, not long ago researchers at Fujitsu-Siemens demonstrated an identification system based on the pattern of veins in the palm, which are believed to be unique and unchanged throughout a person's life. However, the main means are still fingers and eyes. Sometimes, according to Yukhin, fingers just don't cut it, however. Some East Asians have much fainter fingerprints than Europeans, unsuitable for ID purposes.
As A4Vision Vice-President, Alexei Gostomelsky, told Expert, the facial ID methods used by the company's direct competitors are all based on two-dimensional images, on photographs. The only method that can be applied to photos is defining the light pattern reflected from facial features. The main problem with this approach is scale. The more the scales of different images vary, the more difficult it becomes to compare them. Lighting and other so-called masking elements (glasses, facial expression, etc.) also make for insurmountable problems.
A4Vision's technology captures an image of the face's surface using structured illumination. Using this data, it creates a 3-D model of a person's appearance down to the last millimeter. Then a computer algorithm creates a model of hard tissues in the head, in effect "wiping away" the soft tissues of the face and revealing the skull. The modeled skull is incorporated into the model of the face's soft tissues. The image is then ready. The device does more than simply compare the locations of various points on the skull and face; it also compares the curvature of various surfaces and zones. Someone has yet to figure out how to change the angles of the face and cheekbones in order to fool this identification device. Glasses, however, remain a recognition problem, but as they are a surface at a certain distance from the face, the program could note the characteristic projection in front of the eyes and request via speaker that the glasses be removed. Alternatively, this part of the face could simply be discarded from the recognition process.
Now, A4Vision faces the task of creating a recognition system of substantially better quality than those of its competitors. Only by discrediting earlier, two-dimensional attempts will A4's technology become of commercial interest. The A4Vision system currently has the advantage of recognizing faces very quickly, in "real time" according to the company's representatives. This means as fast as video recording, 10-25 images a second, which competing scanning systems can't manage. Here Russian computer science and math reign supreme. However, to start using the facial model system, people have to stand in front of a special camera. Thus, it is only possible to create a database of criminal offenders only by taking them to a special location. And while this may work for criminals, you can't add terrorists, who only appear in bad and blurry photos, to such a database.
So far, the company has not succeeded in creating a camera that can capture images at a distance. The current camera only works at a distance of 1.5 meters, but as A4Vision informed Expert, they are in the process of developing a camera that can catch faces from up to ten meters away and will be able potentially to scan crowds. A4Vision in its current form could never capture moving images in a crowd of people.
It also seems likely that the recognition algorithm will need some fine tuning in the future. When the system is dealing with several hundred people, it works like a charm. However this does not necessarily mean that it will be just as successful with a database of several tens of thousands of records. The system has yet to undergo mass testing.
First money
It is still not clear whether this system will ever reach the market, but it is too early to ask. Products based on recognition technology have just begun to sell. A4Vision is not selling its products directly to customers, but is working through specialized integrators. The basic product the company offers based on its technology is a so-called Software Developer Kit (SDK), a collection of licensed programming modules that can be built into other programs. A4Vision currently has three integrator clients in Europe and one in the US.
In addition, A4Vision is working with the Swiss company Logitech, a well-known producer of computer hardware peripherals such as the optical mouse and the wireless keyboard. A4 developed a special camera for Logitech that automatically focuses on the face of the person in front of it by following the movements of the head. This kind of smart camera will become part of video conferencing programs. In February 2003, Logitech announced that it had licensed this technology and was beginning production. A4Vision had just earned its first money.
Legally, A4Vision is an American company, though only 6 out of 59 employees work in the US: the CEO and CFO, two sales reps, and two troubleshooters. That's it. The majority of the company's employees work in the company lab in the Zamoskvorechye section of downtown Moscow. Between the US and Russia, A4Vision also has a Swiss subsidiary with five employees, in tribute to the company's predominately Swiss-Italian investors.
The company has just gotten its first deals. In two to three years, A4Vision plans to have sales of $10 million or greater and start breaking even. The total investment on the part myQube and its investment partners, Logitech and private investors, has reached $6.5 million. The investment group now holds about 85% of the company's stock, with myQube in possession of 40%.
Apparently, myQube currently considers A4Vision one of the most promising companies in its portfolio. It would be nice to think that the hot security market will help A4 get a leg up, break even, and eventually sell for tens or hundreds of millions of dollars to someone like Oracle or IBM. Maybe someday, with any luck at all, A4 will be making its IPO on the NASDAQ.
Expert (http://www.expert.ru/)
Barebones technology
The story of A4Vision (full name: Application for Vision) begins back in 1998. Two of its founders, recent graduates from the Bauman Moscow State Technical University Artem Yukhin and Andrei Klimov, created a device described in its patent as a zero-contact 3-D optical scanner, or in other words, a 3-D camera. This device was initially built to make machines see, such as independently operating robots moving across an uneven surface. When the year 2000 arrived and the internet boom was in full swing, they like many other young developers at the time began to seek investors. However, an investor found them first. The first million dollar investment came to A4Vision the following year, in 2001.
At first, A4Vision did not have any commercial product to its credit, only barebones technology. They, along with their venture investors, had to come up with something to sell. Initially they planned to create 3-D images of products in online stores based on A4Vision's technology. This proved unfeasible and the idea quickly died. Then the company began to work on software for cosmetic surgeons that modeled patient's outward appearance after the operation. This idea might have been successful, but then came September 11 and a dramatic jump in the demand for security technologies. The company immediately knew what it had to do. Their cosmetic ideas were easily converted into a way to fight terrorism via physical recognition. This also happened to be a problem Yukhin had dealt with before, in his MA thesis.
Security technology, as the term is currently understood, includes everything from protecting computers from hackers to detecting bacteria. This broad category has many smaller divisions and sub-divisions. A4Vision focuses on biometric and identification systems. There are currently three main ID methods: fingerprints, the iris, and the face, which is A4Vision's specialty. These aren't the only ID methods out there; there are many, far more exotic approaches. For instance, not long ago researchers at Fujitsu-Siemens demonstrated an identification system based on the pattern of veins in the palm, which are believed to be unique and unchanged throughout a person's life. However, the main means are still fingers and eyes. Sometimes, according to Yukhin, fingers just don't cut it, however. Some East Asians have much fainter fingerprints than Europeans, unsuitable for ID purposes.
As A4Vision Vice-President, Alexei Gostomelsky, told Expert, the facial ID methods used by the company's direct competitors are all based on two-dimensional images, on photographs. The only method that can be applied to photos is defining the light pattern reflected from facial features. The main problem with this approach is scale. The more the scales of different images vary, the more difficult it becomes to compare them. Lighting and other so-called masking elements (glasses, facial expression, etc.) also make for insurmountable problems.
A4Vision's technology captures an image of the face's surface using structured illumination. Using this data, it creates a 3-D model of a person's appearance down to the last millimeter. Then a computer algorithm creates a model of hard tissues in the head, in effect "wiping away" the soft tissues of the face and revealing the skull. The modeled skull is incorporated into the model of the face's soft tissues. The image is then ready. The device does more than simply compare the locations of various points on the skull and face; it also compares the curvature of various surfaces and zones. Someone has yet to figure out how to change the angles of the face and cheekbones in order to fool this identification device. Glasses, however, remain a recognition problem, but as they are a surface at a certain distance from the face, the program could note the characteristic projection in front of the eyes and request via speaker that the glasses be removed. Alternatively, this part of the face could simply be discarded from the recognition process.
Now, A4Vision faces the task of creating a recognition system of substantially better quality than those of its competitors. Only by discrediting earlier, two-dimensional attempts will A4's technology become of commercial interest. The A4Vision system currently has the advantage of recognizing faces very quickly, in "real time" according to the company's representatives. This means as fast as video recording, 10-25 images a second, which competing scanning systems can't manage. Here Russian computer science and math reign supreme. However, to start using the facial model system, people have to stand in front of a special camera. Thus, it is only possible to create a database of criminal offenders only by taking them to a special location. And while this may work for criminals, you can't add terrorists, who only appear in bad and blurry photos, to such a database.
So far, the company has not succeeded in creating a camera that can capture images at a distance. The current camera only works at a distance of 1.5 meters, but as A4Vision informed Expert, they are in the process of developing a camera that can catch faces from up to ten meters away and will be able potentially to scan crowds. A4Vision in its current form could never capture moving images in a crowd of people.
It also seems likely that the recognition algorithm will need some fine tuning in the future. When the system is dealing with several hundred people, it works like a charm. However this does not necessarily mean that it will be just as successful with a database of several tens of thousands of records. The system has yet to undergo mass testing.
First money
It is still not clear whether this system will ever reach the market, but it is too early to ask. Products based on recognition technology have just begun to sell. A4Vision is not selling its products directly to customers, but is working through specialized integrators. The basic product the company offers based on its technology is a so-called Software Developer Kit (SDK), a collection of licensed programming modules that can be built into other programs. A4Vision currently has three integrator clients in Europe and one in the US.
In addition, A4Vision is working with the Swiss company Logitech, a well-known producer of computer hardware peripherals such as the optical mouse and the wireless keyboard. A4 developed a special camera for Logitech that automatically focuses on the face of the person in front of it by following the movements of the head. This kind of smart camera will become part of video conferencing programs. In February 2003, Logitech announced that it had licensed this technology and was beginning production. A4Vision had just earned its first money.
Legally, A4Vision is an American company, though only 6 out of 59 employees work in the US: the CEO and CFO, two sales reps, and two troubleshooters. That's it. The majority of the company's employees work in the company lab in the Zamoskvorechye section of downtown Moscow. Between the US and Russia, A4Vision also has a Swiss subsidiary with five employees, in tribute to the company's predominately Swiss-Italian investors.
The company has just gotten its first deals. In two to three years, A4Vision plans to have sales of $10 million or greater and start breaking even. The total investment on the part myQube and its investment partners, Logitech and private investors, has reached $6.5 million. The investment group now holds about 85% of the company's stock, with myQube in possession of 40%.
Apparently, myQube currently considers A4Vision one of the most promising companies in its portfolio. It would be nice to think that the hot security market will help A4 get a leg up, break even, and eventually sell for tens or hundreds of millions of dollars to someone like Oracle or IBM. Maybe someday, with any luck at all, A4 will be making its IPO on the NASDAQ.
Expert (http://www.expert.ru/)