Classification of Road Objects using Convolutional Neural Networks

Driving is the primary means of transportation for many people around the world. Whether the use is to assist human drivers or create autonomous driving, the use of machine learning can create safer road conditions. Drivers must consider other objects on the road, most commonly other vehicles, and pedestrians. These three components, road signs, pedestrians, and vehicles make up a large majority of objects that a driver will encounter when on the road. This research applies machine learning algorithms, specifically Convolutional Neural Networks (CNN), to classify these road objects. The goal is to create a classification model that can reliably classify road objects and classify the different road signs into individual classes. The results showed high accuracy in classifying the objects, even at lower resolutions and poor conditions. An accuracy of 99.13% was achieved on the test set of all classes and an accuracy of 97.45% on the German Traffic Sign Recognition Benchmark competition dataset from the 2011 International Joint Conference on Neural Networks (IJCNN).