How It Works
The mobile computing unit uses a convolutional neural network to calculate the current position based on a camera image. Initially, the system automatically collects thousands of camera images and their corresponding positions. An existing network is further trained using this data to adapt it to the target environment. The trained network is then used on the target platforms to determine the positions of new images. To prevent the system from degrading over time, continuously updated information is collected on a central computing unit, the network is further trained, and the results are distributed to the mobile computers.
CNNLok System Components
The mobile computing unit is typically either a simple smartphone or an ARM- or Intel-based single-board computer equipped with a standard camera. The platform’s high flexibility enables many application scenarios that cannot be covered by traditional, infrastructure-based solutions. Customized motion models and specialized preprocessing of the collected data enable new data to be integrated into the existing positioning system. The continuous learning process regarding the environment is a highly computationally intensive task for the system, which is why a connection—for example, via a network or docking station—is required. Depending on the dynamics of the area, a central computer with powerful standard deep learning hardware—such as graphics cards or specialized vector processors—may also be required. It takes over the tasks of the mobile computing units when, for example, they need to be recharged.
Fraunhofer Institute for Integrated Circuits IIS
