Solutions

V&R are experts for algorithmically and technically demanding problems in the field of capturing and analyzing sensor data in 2D and 3D.

Exemplary subject areas

Intelligent systems on wheels: We develop software solutions for intelligent, mobile systems on wheels. Our expertise lies in precise data acquisition and interpretation in both 2D and 3D to enable informed decisions.
Scientific precision and solidity: We work scientifically and realize on a solid basis.
Robust software architecture: With a keen sense for robust software architecture, we design durable systems that meet the highest requirements.

Optimal sensor integration: Thanks to our in-depth knowledge of the latest sensor technology, we make optimal use of it to provide you with efficient, high-performance systems.
Efficient visualization and system understanding: We place great importance on understanding our systems and use efficient visualization methods to gain insights and meet your requirements in the best possible way.
Customized image processing solutions: With expertise in image processing, we offer tailor-made solutions to meet your specific requirements and achieve first-class results.

Background

V&R is a spin-off of the Aktives Sehen (Active Vision) Working Group (AGAS) at the University of Koblenz-Landau. Both partners and founders, Professor Dr. Dietrich Paulus and Dr. Johannes Pellenz, as well as most of the permanent employees, have many years of experience in this working group.

This background as well as the continuous reference to research together with more than 10 years of experience within practical projects ensure an innovative and well-founded competence in 2D and 3D image processing.

SLAM

SLAM, short for Simultaneous Localization and Mapping is a concept from robotics in which a robot simultaneously determines its pose and creates a map of its environment. V&R is an experienced expert in this field and has successfully demonstrated the performance of its systems in international competitions (e.g. KITTI Odometry Benchmark, HILTI SLAM Challenge .

Realization & Applications

Use of inertial sensors (gyroscope, accelerometer, magnetometer) - Tightly-coupled inertial SLAM
Continuous-time SLAM at the state of the art
Compatibility with current sensor technology e.g. Velodyne HDL*, Ouster, HESAI
Optional integration options for further sensors (e.g. GPS [RTK])
Creation of maps: Depending on the sensors, environment, etc., accuracies down to the millimeter range can be achieved
Localization: A machine should continuously estimate its own position without necessarily being dependent on GPS

Sensor calibration

The simultaneous use of several sensors requires extrinsic calibration of all sensors in addition to the intrinsic calibration of the individual sensors. In addition to static calibration using markers visible in all modalities, we also use dynamic calibration methods that continuously estimate and improve the transformation while the sensors are in use. This results in greater robustness if, for example, the position or orientation of the sensors changes during use.

Applications

Project Tango: V&R Vision & Robotics GmbH has been working on improving the calibration software for the Google Project Tango. The Tango device is able to capture a 3D model of the environment and simultaneously track the 6D pose of the device. This allows the creation of systems for indoor navigation in museums or shopping centers, but also augmented reality applications such as the placement of virtual furniture in an empty apartment. Thanks to its experience in the field of image processing and the interpretation of 3D laser scans in real time, V&R Vision & Robotics GmbH was able to support the Tango project in the area of sensor data fusion and calibration.

Artificial Intelligence

V&R has developed its own efficient environment for data storage, annotation and training of neural networks. The structured approach right from the start enables solid results and transparent evaluations in relation to the training database. Artificial intelligence (AI) is a common tool for image processing tasks today. V&R has extensive experience with various AI frameworks, from conception to deployment and optimal use in operational environments. In the early stages of AI projects, the approach is often like looking for a needle in a haystack. Our in-house AI environment is used to quickly achieve technological breakthroughs and test initial approaches: We clarify the problem, select suitable neural networks or approaches and sift through and structure your data. Over the course of the project, we iteratively refine the selected approaches, optimize and expand the database. Once the desired accuracy and quality have been achieved, the deployment is carried out on suitable hardware with appropriate sensor technology.

Applications

V&R is involved in the development of AI solutions for the company TOMRA Systems. The work ranges from an early study with a technical breakthrough to the continuous development of the productive AI environment today. V&R has developed an AI-based image recognition system for a well-known manufacturer of agricultural machinery, which controls actuators with precise positioning. The work ranges from an early study with proof of technical feasibility to the provision of a suitable AI environment, the selection, adaptation and training of a suitable network and the provision of an executable solution on a Jetson platform for operation in the field.

V&R completed the research project “3D Auto Semantics” at the end of 2023. The project’s goal of being able to automatically assign semantics to raw data in 3D point clouds was achieved. This means that projects that require a semantically enriched 3D point cloud can now be processed efficiently.

Digital twin

Digital twins are virtual images of real systems or processes based on advanced technologies and precise data. V&R uses its expertise in sensor technology, image processing and AI to develop highly accurate digital twins. Our experience in integrating and calibrating various sensors enables us to ensure accurate data acquisition in real time. This data serves as the basis for modeling and simulating real processes.

Starting with a careful analysis of system requirements, we select suitable AI models and technologies to create an accurate digital representation. We continuously iterate and optimize to improve the match with the real system. The structured approach and the ability to obtain meaningful results already during the project support the development of high-quality digital twins. These virtual representations are used in various application areas, from predictive maintenance to the optimization of processes and products.