Optical 3D Object Recognition for Automated Driving

Abstract

In this contribution, we propose an system setup for the detection and classification of objects in autonomous driving applications. The recognition algorithm is based upon deep neural networks, operating in the 2D image domain. The results are combined with data of a stereo camera system to finally incorporate the 3D object information into our mapping framework. The detection system is locally running upon the onboard CPU of the vehicle. Several network architectures are implemented and evaluated with respect to accuracy and run-time demands for the given camera and hardware setup.