Lane Detection Based on Improved RESA in Power Plant

Abstract

Lane detection is one of the important tasks of the environmental patrol work in power plant. In order to improve the detection accuracy of lane, this paper proposes a tensor fusion structure RCFPN, and takes the lane detection model RESA as baseline. After the backbone feature extraction network of RESA model, RCFPN is added to construct the improved network. The experimental results prove that RCFPN has an effect on improving RESA's precision. RCFPN can not only improve the precision of RESA model, but also can be flexibly integrated into other lane detection models and other target detection models. The average detection accuracy of CULANE was increased from 75.31% to 77.76%. The F1 score, accurary, FP, FN are better than the original model in the Tusimple data set.