Implementing Machine Learning Algorithms on GPUs for Real-Time Traffic Sign Classification
Department of Electrical Engineering, Stanford University
Stanford University, CS 229 Machine Learning project, 2014
This paper investigates traffic sign classification, which is an important problem to solve for autonomous driving. Linear discriminant analysis and convolutional neural networks achieved an accuracy of 98.25% and 98.75% respectively when classifying eight different types of traffic signs. The CNN was implemented on a GPU for real-time traffic sign classification: testing time for the CNN on a GPU was 4 ms/image, which was 7.5x as fast as running LDA on a CPU and 60.2x as fast as running CNN on a CPU. Additionally, different types of classification errors and the effects of adding a new sign to the dataset were explored.
March 12, 2015 by hgpu
