2019/11/15 by Sohini Roy Chowdhury, Chowdhury, Sohini Roy, Lars Tornberg +23
Computer Science · Engineering · #Advanced Neural Network Applications #Artificial Intelligence (cs.AI) #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Image and Object Detection Techniques #Machine Learning (cs.LG) #Vehicle License Plate Recognition
paper · pdf · doi:10.48550/arxiv.1911.06486
openalex publication_date 2019/11/15 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
Traffic sign identification using camera images from vehicles plays a\ncritical role in autonomous driving and path planning. However, the front\ncamera images can be distorted due to blurriness, lighting variations and\nvandalism which can lead to degradation of detection performances. As a\nsolution, machine learning models must be trained with data from multiple\ndomains, and collecting and labeling more data in each new domain is time\nconsuming and expensive. In this work, we present an end-to-end framework to\naugment traffic sign training data using optimal reinforcement learning\npolicies and a variety of Generative Adversarial Network (GAN) models, that can\nthen be used to train traffic sign detector modules. Our automated augmenter\nenables learning from transformed nightime, poor lighting, and varying degrees\nof occlusions using the LISA Traffic Sign and BDD-Nexar dataset. The proposed\nmethod enables mapping training data from one domain to another, thereby\nimproving traffic sign detection precision/recall from 0.70/0.66 to 0.83/0.71\nfor nighttime images.\n