Tensor-Centric Processor Architecture for Applications in Advanced Driver Assistance Systems

Yu-Sheng Lin, Wei-Chao Chen, Trista Pei-Chun Chen

Abstract: We describe our implementation of a tensor-friendly AI processor architecture for an advanced driver assistance system. We implement modern video-based lane-departure and blind-spot detection algorithms on our previously proposed architecture called VectorMesh, and demonstrate deployment of our technology using open neural network exchange compilers and FPGAs in a commercially viable and cost-efficient manner.