Rapid Prototyping of a Computer Vision Stack for AD Using MATLAB and Simulink

There are a large variety of AI learning frameworks. If you are interested in a particular convolutional neural network, you are restricted to the framework it was originally developed in. Often Docker containers are used to run different networks at the same computing hardware. When running different networks into a test vehicle, a standardized way of deployment is mandatory instead of maintaining different Docker containers with competing requirements to the GPU driver and libraries. It is a comfortable handling of the complete vision stack, including image acquisition, network inference, and all preprocessing and postprocessing steps.

MATLAB^® and Simulink^® provide many image processing functions and supports to run neural networks based on the Open Neural Network Exchange (ONNX) format—an established standard in the community. Furthermore, the capability of C/C++ code generation is beneficial for in-vehicle usage.

In this presentation, see different deployment options using CPUs, GPUs, standard PCs, or embedded devices.