APPLY Case Studies in A.I.

We have used computer vision, predictive & behavioural analysis to develop A.I. solutions for 7 industries. Here are most of our case studies.

AdEx - Real-time visual media analytics 

A.I. and Neural networks automatically analyse content from video streams (cable, satellite, IPTV networks or OTT platforms).

AdEx is capable to:

  • automatically detect ads from TV stream real-time

  • identify and detect various objects, subjects and scenes

  • extract text from screen and speech-to-text

AdEx right now is used to track & analyse TV ads. Which means that media buyers now can see if their ad have been shown at the right time, with AdEx you can compare your competitor ads side by side. This technology also is used to track politician activities in election period. 

In short - finally there is transperancy!


Industrial grade robotics adopted for nonlinear tasks.

One of the bigger challenges for faster and more wide integration of robotics in everyday manufacturing, processing and servicing tasks is entropic and unpredictable environment of human world.

For us it is easy to figure out which object is to be picked first from box of randomly arranged different shape and size objects, for traditional robotics - not so much.

We are talking about AI vision enabled robot, which can autonomosly analyze contents of container by determining shapes and positional relations of different objects for most effective processing of those objects. System also is capable of classifying objects based on size, weight and visual appearance of objects. For each type of object robot encounters there can be different routine performed.

This product is intended to make significant steps towards changing status quo. 


Smarter robots for metalworking

When working with large scale raw metal pieces it is normal to have some variations in dimensions. When human worker is welding or cutting such piece there is no problem at all. However, usually there has to be some checking and some additional programming to perform, so robotic manipulators can work on such metal piece. This can be time consuming and since there is human evaluation involved, mistakes happen.

So, in collaboration with ABB Robotics we are adapting standard robotic manipulator for automatic evaluation of raw metal pieces and adapting blueprint to all its irregularities.

System can scan, for example, metal beam and detect that it is wider in middle than at both ends and can modify plasma cutting profile to take that into account.

Finance & Insurance

Automatic medical receipt submission for insurance company

We developed a solution for one of the largest insurance companies in Europe “If P&C Insurance” LTD.  The solution is combines computer vision and optical character recognition. 

The problem. 

So far, clients collected their medical receipts for long period of time and then went on to submit them. The main reason for developing a new and innovative solution was the submission process - it took a lot of clients precious time. 

The solution.

Now the client takes a photo of the receipt, uploads it to the insurer's system and that is it. 

Our A.I. system:

  • recognises a receipt in the uploaded image;

  • understands what kind of information is in the receipt;

  • fills out the form itself.

The best part is that A.I. can extract the information from all kinds of receipts - nobody taught A.I. all the possible receipt types, this process alone would be very time consuming and unefficient.


Vision system for casino table accounting control.

This product is part of Casino Management System (CMS). It performs visual observation of any card table and track chip exchanges between dealer and players.

As part of CMS it tracks whether dealer makes correct payouts to players by analysing both card combinations and chips value, both by chip count and nominals.

Also it builds statistics for each dealer - their performance under different circumstances.


Vison system for fortune wheel type game monitoring.

System for monitoring of fortune wheel type game. It visually observes wheel and determines game start, wheel spinning speed, direction, game stop and winning sectors as well as uploads all the data to server for further processing.

System also can work with any type of video feed, so it does not neccessary have to be located near the wheel itself.


Artificial Intelligence enabled dairy product contamination control system.

Currently de facto industry standard in dairy product quality control was developed in 19th century. It is simple and fool proof - take a piece of dairy product, put in petri dish with some food for bacteria and fungus to grow. Wait for 5 days and there you have it.

Drawback - you need to store huge amount of processed and ready for consumption food before it can hit the shelves, thus shortening shelf life significantly, spending a lot of money on refrigeration. And we are not talking about logistics yet.

This product will change that 5 day wait process to couple hour process, buy using patented contaminat cell biomarking method and automated microscopy platform.

This platform scans whole sample, finds actual cells of microorganisms and provides exact counts of them. Visual recognition is done by utilising neural networks. In fact this system can be adapted to wide variety of microscopy visual analysis cases.


Interactive face motion game to support children with speech and face disabilities.

CheeksUp - 3D facial expression recognition system, that gives feedback to user in real time through game like animations.

The idea is to engage children in otherwise boring and repetitve tasks of facial muscle exercising. All the exercises are adopted from tradition speech and facial muscle rehabilitation medicinal practices.

This cloud enabled system also includes patient & therapist profile system, thus can track detailed activity and progress of patients, generating useful medical history data.


Detecting fire before it before it starts, or spreads

A company that owns multiple office buildings needed a solution that can detect potential fire break outs at a very early stage and if a smoke, fire, or flame originates from an area where it should not. We used computer vision and here's why. A camera can see what the human eye cannot, such as infrared and ultraviolet light levels. The AI can detect the difference between a flame from a candle, or a fireplace, where they should be, or from a smoldering curtain or carpet where it should not be.  Much faster than standard smoke or flame detectors.

And combining the AI with advanced smoke detectors, determine the difference between cigarette smoke or a bed on fire.  The result - a sophisticated system for hazard alerts.

Automobile and Human counting