top of page

COMPUTER VISON
DEEP LEARNING

ABOUT US

We are a technology company focused on the application of artificial intelligence in industry. We design and supply solutions for data collection, data analysis, optical control stations based on neural networks and a system for reactive, scheduled and predictive maintenance.

We have laid the foundation of the FRAVEBOT system. He is now in the independent company FRAVEBOT s.r.o.

ABOUT US
COMPUTER VISION with DEEP LEARNING

We use DEEP LEARNING for deep vision neural networks for machine vision tasks. These methods are suitable for complicated tasks with high variability of observed objects. Tasks can be divided into Object detection or Anomaly detection.

Object detection can accurately find the search object by which the network is taught. This is the so-called learning with the teacher, where we explicitly determine the patterns that represent a specific object. This method is suitable for finding known defects or objects that we expect in the scene (eg components).

Anomaly detection uses a comparison of the scanned image with a mathematical standard composed only of OK images. It is not necessary to specify a network for specific patterns. This method is used mainly in the search for unknown defects on continuous surfaces (surface of products, deviation from the fabric pattern, painting defects).

As a data source we use cameras with GeniCam interface working in different spectra. The resulting recognition can combine several inputs. A typical example is a combination of a camera and a telecentric, where the detected visual defect is confirmed by looking at the profile of the tested object.

deeplearning
robots
ROBOTIC APPLICATIONS

We create robotic applications with robots, cobots. We design our own autonomous robots for agriculture.

We use ROS2 for management, we train applications using Reinforcement Learning.

We use data from LiDAR or 3D cameras for navigation.

Object detection is realized by neural networks.

We perform training on synthetic datasets, which drastically shortens the development cycle.

PCs with NVIDIA GPUs or NVIDIA Jetson minicomputers serve as control computers.

We control peripherals via a Siemens or Beckhoff PLC.

bottom of page