Detection of Traffic Signs Based on Support Vector Machine Classification Using HOG Features

Real time traffic sign recognition algorithms are in high demand as technology pushes for autonomous vehicles. Identifying traffic signs that regulate the flow of traffic provides another way to increase the safety of the driver and other road participants. In the present paper we propose an algorithm that introduces a filtering step which significantly reduces the processing times of an SVM based classification algorithm. Our approach uses the image processing techniques to identify regions of interest (ROIs) in an image, based on color information and on certain object properties. In the experiments, we evaluated our TSDR system on recordings of Romanian roads with traffic signs in various lighting conditions.