High speed road boundary detection with CNN-based dynamic programming

Abstract

Analogic CNN-based road boundary detection algorithm for fast and optimal driving of autonomous vehicle is proposed. The CNN is a massively connected analog parallel array processor. In the previous study, the dynamic programming which is an efficient algorithm to find the optimal path is implemented with the CNN algorithm [1]. Due to the parallelism of the CNN structure, the processing speed of the CNN-based dynamic programming is expected to be very fast. The proposed study is an extension of previous optimal path finding algorithm for high speed road-boundary detection. If the road-edge images are utilized as the space variant weights and if the goal and the start lines are positioned at the top and the bottom of the image, respectively, the optimal path finding concept can be exploited for the road boundary detection. The proposed road boundary algorithm is described and simulation results are included. © Springer-Verlag Berlin Heidelberg 2002.