Neurodata Lab

History

2020: The first place in international contest by contactless determination of pulse

On March 10, 2020 the Russian company Neurodata Lab reported that it won first place in international contest by the contactless determination of pulse organized within the international CVPR2020 conference.

Neurodata Lab developed the algorithm which won first place in tender on the contactless recognition of physiological signals (1st Challenge on Remote Physiological Signal Sensing (RePSS)) organized at CVPR2020 - an annual international conference on computer vision and image identification. In 2020 it passes in Seattle, the USA, and traditionally involves a great number of the researchers and engineers working in this area.

The purpose of tender consisted in determination of the algorithm which is most precisely distinguishing the mid frequency of pulse of the person according to video as it is the most fundamental problem in the field of contactless recognition of physiological signals. The technologies capable it is contactless to read out such signals, are demanded in a number of the industries - in particular, in health care, telemedicine, marketing researches, and fitness.

Our algorithm on the basis of convolution neuronets determines pulse by a 10-second fragment of video. The feature of the task set at tender was that people on videos at which it was necessary to define pulse, talked and could move freely in a frame. Also on video bad lighting was frequent that complicated recognition of pulse. For fight against these difficulties we aligned (stabilized) persons on video, detected areas of a mouth and eyes and noted them as irrelevant for measurement of pulse. Besides, to video frequency filters were used to decrease in impact of non-periodic changes of lighting and digital noise, - told TAdviser provided Neurodata Lab.

More than 120 commands and individual participants took part in tender. They had to test the accuracy of the algorithms at determination of pulse of the person according to video. For this purpose participants trained previously the algorithms at the special videos provided by organizers and then tested on test data. The algorithm developed by Neurodata Lab showed the largest accuracy among all participants and the smallest discrepancies with real values of pulse. Analysis results were loaded by the leading Data Scientist of the company Mikhail Artemyev under the nickname Mixanik.

Data of tender were rather difficult and close to real life application of technology therefore the fact that our algorithm showed good results, shows that our technology will work with high accuracy and in commercial applications, commented Mikhail

2019: Hit in the three of the best products according to emotions in the world

On September 30, 2019 the Neurodata Lab company reported to TAdviser that it entered into the three of the companies in^[1] of the British scientists who found out whether the artificial intelligence will be able to recognize emotions more precisely, than the person. The top of smart algorithms included technologies of startups from the USA and the Netherlands.

Both algorithms, and people had to determine six emotions by a look: joy, grief, surprise, rage, fear and disgust. Developments tested on 938 videos which then also showed to people. A part of emotions on video was played, other emotional manifestations were spontaneous.

It turned out that the medium accuracy of recognition of emotions on the person the person does not exceed 73%. For comparison, the medium accuracy of Neurodata Lab technology for seven classes of emotions was 57% (probability of accidental guessing for six emotional classes - about 16%). At the same time some algorithms will recognize joyful and sad looks even better or also well, as well as people. The played emotions and to artificial intelligence, and people were simpler to be recognized, than spontaneous.

Actually in the world there is no 'reference' expression of this or that emotion. Your smile will seem to someone joyful, and someone will take it for manifestation of politeness. Therefore even among people there is an inconsistency in what emotion they see on video. Our technologies proved to be very well, we managed to bypass large corporations and the companies which are present at the market more than 10 years, comments on a research Georgy Pliyev, the founder and the managing partner of Neurodata Lab company

The company uses algorithms for recognition of emotions and other behavioural characteristics of the person for the analysis of customer service, mass recruitment and digital medicine. Neurodata Lab is known for the projects with robotic company Promobot and also products for client analytics in contact center of Rosbank.

In addition to the analysis of emotions, Neurodata Lab develops systems for complex analytics of data on the person. In particular, systems for recognition of difficult cognitive statuses according to video and audio, non-contact methods of determination of pulse, breath and other physiological parameters, collecting and the analysis of affective data.

Notes