USTS 75K: United States Traffic Sign Dataset

Traffic signs are crucial in advancing self-drivingcars and the broader automotive industry in the United States(US). Our study introduces a comprehensive US Traffic SignDataset, derived from US Division of Motor Vehicles (DMV)driver\u2032s license test exams, to support this endeavor. The studyaims to generate 12,500 augmented samples from the originaldataset through spatial and temporal augmentation techniques.We enhance the dataset with Gaussian noise using increasingstandard deviations of 25, 50, 75, and 100, which enables usto produce 12,500 samples for each noise level. Poisson noise,a next level augmentation technique, was applied to generatean additional 12,500 samples. Each subset comprises 50 distincttraffic signs, a unique class and each class has 250 samples.In total, our newly generated US Traffic Sign (USTS-75K)dataset consists of 75,000 traffic sign images. To demonstratethe generated dataset\u2019s classification capability, we implementedthe ResNet-50 deep learning (DL) model, achieving promisingresults with accuracies as high as 91%. The primary goalof this research work is to develop an authentic and diversedataset of US traffic signs, encompassing various illuminationsand positions, to enhance driving assistance systems and theself-driving automotive industry. The project holds significantpotential for improving the reliability and safety of autonomousvehicles and driving assistance technologies.