Model-based selective image downsampling in remote driving applications

Paper i proceeding, 2022

The large-scale deployment of remote driving technologies can be limited by the large (upstream) link capacity necessary to stream the video from the onboard cameras to the remote driver. This paper proposes an algorithm that reduces the size of the upstream video, by selectively downsampling portions of the frame recorded by the onboard cameras. Portions of the frame to downsample are selected, which surround those road users (RUs) that are not likely to collide with the remotely driven vehicle. In this paper, we propose to select such RUs by using reachability analysis techniques. In order to limit the computational overhead introduced by the proposed algorithms, we resort to reachability analysis based on ellipsoidal sets. The algorithm is demonstrated in a simulation environment in a T-junction by showing a significative reduction of the frames size.