
Intelligent Systems Research Group
Komo3D
Context-Sensitive Mobile 3D Multi-Sensor System
Team:
Johannes Dornheim, M.Sc.
Prof. Dr. Franz Quint
Funding: Baden-Württemberg Stiftung
Duration: 2019 - 2021
Overview
In Komo3D an innovative mobile 3D multi-sensor system is implemented, which can capture and evaluate image data as appropriate to the situation. The main characteristic is a 3D multi-sensor system for capturing the image data in connection with a distributed system architecture consisting of components with different capacities in terms of real-time capability and computing power.
The 3D data is obtained by multi-sensor image acquisition on a mobile system in conjunction with a field-programmable gate array (FPGA) for real-time image evaluation and classification. Taking into account the limited computing power of the mobile unit, it is supported by a tiered edge / cloud computing system.
Machine learning processes based on convolutional neural networks (CNN) are used for object detection and situation recognition. The neural network is implemented directly on the mobile unit, while the configuration or reconfiguration of the network takes place via edge and cloud computing.
Partners & Funding
The work in this project is carried out with our partners
Komo3D is funded by the Baden-Württemberg Stiftung.