We offer ready to use and tested solutions:
Queue Management
The idea of this product is to provide better planning of cashiers’ work in retailers. CCTV cameras observe customer flow: how fast customer goes through the shop, how many enters and leaves. Just before the POS area, the key task of cameras is to register the size of the queue - recognizes the number of people waiting in the queue to pay for their goods.
Furthermore, the system gets information on how fast each cashier works - how many products she/he scans, break time between subsequent receipts.
Based on the above information, the system predicts the waiting time in each queue in real-time. Historical data gathered during the system run is used to anticipate the number of cashiers needed per day.
Depending on the number of cameras around the shop we can create a heat map of customer traffic, how long they spend in each section of the shop, and how fast they move inside. The system can predict demand on the number of open cash registers just based on the number of customers enters the shop.
Parking management
The main purpose of the parking management system is to control traffic in parking spaces. Our system uses cameras for the monitoring parking area to detect empty parking spots. This can be used to direct drivers in real-time towards empty spaces which will minimize the time spent looking for one and reduce traffic congestions. Information from the system is presented on display boards placed on parking alleys.
By increasing the number of cameras, the system can be further equipped with additional features, such as:
- Traffic jam detection
- Number plate recognition
- Cars parked in ban area
- Pedestrians in prohibited places
Field workers can be equipped with a tablet application to access all information, alarms, cameras view, and directional view (map view of parking from current position). Current occupancy of parking can be presented online on a map view on the website.
Face detection and recognition
Face detection and recognition are becoming more and more popular as an application of computer vision.
We can arm any CCTV system with face detection and recognition- no matter what make or video register currently used.
Our solution works in parallel - we read the stream directly from a camera, process it, and present the results in a suitable form. For example, we can trigger action, push a message to the mobile phone, or open the door.
Hand detection and recognition
Hand detection and recognition algorithms are dedicated to safety and security. Assume you have a big machine with dangerous parts and no one should put a hand inside, a detection algorithm can initiate an alarm when a hand appears too close to the machine.
Recognition can be used to verify the identity. We examine movements of a hand and based on this we can verify the identity of someone known to the system.
Virtual interactive systems
Virtual interactive system allows us to interact to computers or systems.
It recognizes poses, movements or gestures, such as waving your arms, when you need help. A camera equipped with specialized algorithms can recognize and trigger an alarm to the security team.
Other uses can be drawing a circle around an unattended bag and ask the CCTV system to monitor that particular bag, and when it is removed (taking away by someone) sends an alarm and saves a short clip.
Artificial Intelligence is our focus area. We use our knowledge of neural networks and genetic
algorithms to solve challenges, no matter small or big.
Genetics algorithms & neural networks mix
The concept of Genetic algorithms and neural networks has its roots in artificial intelligence. Both methods are computing systems inspired by the nature.
Genetic algorithms are commonly used to create high-quality solutions for optimization and searching. They are relying on bio-inspired operators such as mutation, crossover and selection.
Neural networks are designed to works similarly to the human brain’s neural network and are mainly used to recognize patterns. Combined each other accelerates the learning process to solve a challenge quicker and cheaper.
Deep learning – neural networks
Deep learning is part of a broader family of machine learning methods based on artificial neural networks.
Deep learning has added a huge boost to computer technology. With this method, a lot of new applications of computer vision techniques have been introduced and are now becoming part of our everyday lives. These include self-driving cars, face recognition, human action recognition, etc.
We apply deep learning to improve existing solutions, automate monotonous tasks and fetch unseen conclusions.
Machine learning
Machine learning helps computers understand what they see. It is a hot topic at the moment, not only since Google’s AlphaGo AI managed to beat the human world champion in the game of Go. Computer vision is seen as a key factor for automation in various fields like autonomous cars, intelligent agricultural machines, robotics. Machine learning has wide applications in the area of data processing, like fraud detections, anywhere where the size of data is big and relations between them are hard to understand by humans.
We apply machine learning methods to design more sophisticated data processing methods and to deliver a deeper understanding of data.
Computer vision is our key area. We are experts with decades of hands-on experience. AI principles have introduced revolutionary changes in precisions and possible applications. Together with powerful hardware, areas of usage are becoming wider with every new GPU introduced on the market. Imagine that 768 cores of GPU can monitor hundreds of parking spaces (NVIDIA laptop card, introduced in October 2016, has 768 GPU cores).
Behind the scene of computer vision, there are many classic algorithms we have implemented ourselves and we can deploy to any hardware:
- Dense optical flow
- Haar’s features
- Histogram of Oriented Gradients with extension to allow detection of objects no matter of their orientation
- Colour tracking
- Dynamic scene tracking
- Kalman filter
- Advanced image transformations
- Contrast enhancement
- Adaptive binarisation
We combine classical computer vision with a more sophisticated one based on neural networks. We use pre-trained neural networks to save time and money. Sometimes a network needs to be trained on a specific dataset to get better prediction. In such a case we compile a collection of images, tag them and retrain the neural network.
We offer our expertise to choose the best-suited image processing pipeline, combining classic computer vision algorithms, machine learning, and deep learning principles.
Realistic images can be created by AI. There are a lot of examples, like the creation of non-existing faces based on real people, animation of facial movements. Fake news, especially with short movies with people acting like the real thing could be dangerous.
We are focusing on two aspects of such uses:
- Is it possible to validate images or movies as realistic or manipulated by specialized algorithms?
- Can we use AI to speed up the processing of 3D transformations?
We invite everyone to visit the web page: agnes3d.pl
Someone may ask, why aircraft - because one cofounder loves it!
We are starting our journey with AI in visualisations and we hope it will be one of our top priorities in the next few months. We believe AI should be used only for good things.
Besides of use of AI, we can pack solutions in beautiful and easy-to-use wrapping. All HTML-based, interactive and dedicated to solved challenges, flat, 3D or holographic.
We started working as a team in 2018. It began as a result of the first contract we won. We decided to establish a dedicated company focusing on AI and its applications for businesses.
The name of our company is related directly to one of the cofounders: Tom’s eyes. Years ago he started investigating neural networks when the world was fascinated by information sharing through World Wide Web. He saw more and now there is a time to deploy his knowledge in the real world. He holds a Ph.D. in technical sciences.
We are data engineers with decades of experience in image processing, data architecture and processing, performance and tuning. As we are a small team we work closely with IT companies that provide the infrastructure needed to realize our visions. Typically we start with a proof of concept which shows customers what can be made real.
During past years we have learned how to create end-to-end solutions from data receptions, through data processing up to advanced visualisations in on-premise, cloud, or hybrid environments. We join our past experiences to provide the best and most effective solutions.
AI evolves quickly and every project is challenging. We are doing our best to deploy the newest and verified algorithms.