Flexible neural network for fast and accurate road scene perception

Abstract Accurate object detection on the road is most important requirement of autonomous vehicles. Extensive work has been accomplished for car, pedestrian, and cyclist detection; however, comparatively, very few efforts have put into 2D detection. In this article, a dynamic approach investigated to design perfect unified neural network that could achieve best results based our available hardware. The proposed architecture CSPNet feature extraction in an end-to-end way. net extracts visual features by using backbone subnet, pyramid (FPN). order increase flexibility, auto-anchor generating method applied layer makes suitable any datasets. For fine-tuning net, activation, optimization, loss functions are considered along with multiple check points. trained tested benchmark KITTI dataset. Our extensive experiments show model superior others, where other nets output low accuracy pedestrian detection, achieves 99.3% recall rate