The DeepEyes Algorithm
The DeepEyes algorithm is an artificial intelligence methodology in the true sense of the word: It works with autonomous deductive reasoning and includes both machine learning and deep learning technologies. This makes it unique.
Like any artificial intelligence the algorithm evolved over time. In the early beginnings of our research, our goal was to simply find a quick and accurate solution for remotely measuring the human pulse, by using a video cam. We experimented with algorithms we already knew, but always found that they were too slow and returned a large error count.
Less Code, More Math
So we came up with the idea to radically change our approach: For accurate real-time results we researched various mathematical and statistical models and eventually came up with the optimal combinations which we hard-coded using low-level languages. The outcome was impressive: Our newly developed – “DeepEyes” as we called him – algorithm produced highly accurate results in real time. Therefore we kept evolving the scope of the models over time. Pulse analysis was not sufficient anymore, we developed more application cases, for example for analyzing movements of facial muscles which is one of the most complex fields of study. Our newly detected combination of algorithms made it possible to detect tiniest aberrations and we started to research human emotions through the analysis of micro expressions.
After extensive testing, we saw even more potential for the use of the technology and started to experiment with targets other than the human face. First we experimented with human gestures and motions, but soon we realized that any kind of object that can be recognized and analyzed with a camera is a good recognition target. We even combined various camera types (e.g. infrared, near infrared and off the shelve cameras) and started combining the results. This showed good results, also.
Over the time, we developed an algorithm for the hidden relationships between events and behavior and evolved models to predict results with high accuracy.
What we are proud of is that this our algorithm makes it possible to execute highly complex calculations with low computing power and with unprecedented precision. Also, the technology works independently, without neural networks of any kind. The technology can be used on any small device with a microcontroller.
This represents an invaluable advantage: although the technology can access any data base, neural network, internet data source, it can also work standalone, independently, without internet access. Thus, application possibilities become endless. Artificial intelligence and video recognition will become a technology open for all.