The DeepEyes Algorithm
Born from almost two decades of research conducted by video analytics and neuro technology scientists as well as experts in predictive technologies we have developed a complex, self-learning algorithm that can reliably and accurately analyze video data in real-time without external communication like grid computing or neural networks. Instead the technology is an on-device neural network accellerator.
The technology operates on standard off-the-shelve hardware, even on a tiny micro-controller.
The uses of our technology are endless and we continue to push the limits of what is possible.
- A unique combination of mathematical and statistical algorithms
- Processing of information in RAM
- Use of low-level languages
Why This Technology is Special:
- Processing of large amounts of data per unit cycle.
- We solve complex classification tasks on low-end controllers (e.g. Arduino)
- Currently testing solutions that allow us to increase our speed 20 times. For example, 4000 polygons are classified in a few milliseconds.
- Most analytics tasks around our technology can be solved using a normal smartphone.
- Customization or OEM on request.
- Simple interface, easy integration and customization.
For specific customer requirements we have developed the following technologies:
Less Code – More Maths
The DeepEyes technology uses modified(!) Bayesian methods for the classification and analysis of images and data. The technology is adaptive, this means it can evolve models that dynamically respond to changes. The core algorithm consists of definitions of about 98 physiological parameters. It is able to work without external communication; that means that it does not need neural networks or grid computing. Depending on the specific requirements the algorithm is slightly altered or combined with additional technologies.
The Algorithm Includes (but is not limited to):
- Pulse, eye movement, respiratory rate
- Patterns of behavior/reactions to events
- Different variants of gestures/movements
- Head turns, activity of facial muscles on areas of the face
Individual Emotion Recognition | DeepMood
DeepMood can identify specific individuals´ emotions.
In a one on one setting (interview, conversation, interrogation) our Emotional Recognition Technology DeepMood compares facial expressions with the answers given orally – thus showing aberrations of body language and spoken replies.
Based on the unique DeepEyes algorithm DeepMood can tell the emotions of one specific person in an extremely performing and efficient way. In fact the technology works like a lie detector without the traditional wiring and without requiring a specialist polygraph examiner who interprets the results of the interview/interrogation.
DeepMood relies exclusively on the facial expression of the examined individual and is able to draw conclusions on the truthfulness or falsehood of statements.
- DeepMood works with simple HD cameras.
- DeepMood tells emotions like uncertainty, anger, surprise, well-being, confidence, mental disorder etc.
- 25% – 30% increase in accuracy as against traditional lie detectors
- Undemanding in terms of computer power
- Works efficiently on smart phones or pads
The DeepEye Technology DeepPose can accurately and completely track precise body movements through a normal HD camera. It accompanies the human body down to its fingers and toes and recognizes the distance from the camera/screen as well as the specific body movements.
All this is possible with a standard web-camera on a standard personal computer or, alternatively on a smartphone or pad. DeepPose does not need any additional equipment or accessories. The user can move freely without wiring. From now on, special motion sensors are obsolete. DeepPose can work standalone / without an internet connection.
Also, Gesture Recognition can be used in highl digitalized environments where commands are given through simple hand movements. Use scenarios are interactions with smart home appliances or IoT environments.
Crowd Recognition | DeepCrowd
The DeepEyes Technology DeepCrowd is a system for super-accurately recognizing patterns, outliers and anomalies in crowds in near-real-time.
DeepCrowd autonomously learns and recognizes patterns of behavior in a specific location and through its adaptive algorithm differentiates these from anomal patterns. Relying on massive numbers of data analytics and alerts to anomalies and potential threats unlike other products DeepCrowd can work standalone and is in a position to even identify subtle deviations. This makes it an ideal solution to early indicate potential threats in crowds.
In combination with DeepYou (?) the solution can identify and trace individuals by their faces, their body movements and way of walking and cross-reference identified persons against relevant databases.
Face Recognition | DeepYou
DeepYou identifies or verifies individuals under the most demanding conditions against specific databases or watch lists, including:
- by their faces, even if partly covered by scarves, beards, eye glasses, hats etc.
- by their faces even if they only cover as little as 15% of the screen
- by their faces from a wide range of viewing angles
- by their body movements and gestures
- Distance from 0,5 to 10 m
- Recognition of data that are several years apart (e.g. the database contains a photo of an 18 year old person, the camera recognizes the same person at the age of 25)
For privacy reasons the DeepEyes technology does not store complete pictures but rather privacy secure unrecognizable pixels as well as statistical information
General Emotion Recognition | DeepFace
In surveillance settings with several persons the DeepEyes technology DeepFace can identify general emotions like attention, surprise, disgust or indifference on human faces. Also,the technology can tell changes in mood as well as the duration of attention or the line of sight in an extremely accurate way. Additionally, persons can be categorized in terms of age, gender, clothing etc. This enables to effectively analyze customer behavior and adapt to changing demands and expectations. The technology works with data from normal cameras and its self-learning algorithm. DeepFace can classify customers in terms of age, gender and optional categories like clothing size or color and analyze customer-facing marketing and sales activities.
Data may be gathered online, at the point of sale or on any location where a camera can be mounted, e.g. on billboards, CCTV systems etc..
Anomalies Recognition | DeepShift
DeepShift can recognize even minor changes, anomalies or outliers on surfaces from raw materials and purchased parts to partially completed goods to finished goods or merchandise. Thus the technology can be used for inspecting surfaces and ensuring that they meet the specified quality standards.
The technology can be used for any kind of surface quality monitoring and can replace costly manual labour or high-tech machinery.
DeepVariant can be customized to any requirements including:
- Determining tolerances
- Parameter design and tolerance design
- Experimental design for dynamic characteristics
Advantages of Using Visual Recognition in Manufacturing:
- DeepVariant works with simple HD cameras.
- DeepVariant recognizes anomalies and outliers of any kind and can be easily customized to specific requirements
- Substantial cost reduction as compared to manual inspection
- Undemanding in terms of computer power