Annie Wu¹, Yu Sun², ¹USA, ²California State Polytechnic University, USA

Autonomous vehicles are a potential solution to preventing crashes caused by human error. Although road signs are intended to attract drivers' attention and help them operate, drivers can still misinterpret signs, resulting in an accident. An autonomous vehicle system can implement artificial intelligence to detect and recognize known patterns in input graphics to minimize the human aspect of driving. In this study, we present an implementation of the CNN architecture to classify four regulatory instruments (stop, crosswalk, speed limit sign, and traffic light) using the TensorFlow library. We used a training dataset of 877 images of the four distinct classes to optimize the model. The goal of the study was to create a lightweight and accessible image classification model. Experimental results show a 92% model accuracy.

Machine Learning, Image classification, Autonomous vehicle