The aim of this thesis is to develop and implement an algorithm that automatically detects and recognizes U.S. speed signs, from the grayscale images captured by a camera mounted on the interior mirror of a vehicle, as a part of designing smarter vehicles. The system operates in real-time within the computational limits of contemporary embedded general purpose processors. This system will assist the driver by providing the necessary information, regarding the assigned speed limits, right in front of him and provide additional safety measures by monitoring the vehicle’s speed.
The proposed method consists of two phases in it: a detection phase, in which all the possible speed signs in the input image are detected first, and a recognition phase, in which the detected regions are recognized and the information regarding the speed limits is extracted from them. The detection phase utilizes the region characteristics, such as aspect ratio and size, to hypothesize the speed sign locations in the input image. We have utilized the idea of connected component labeling technique and adapted it for the grayscale images, to divide the input image into a set of regions. The recognition phase calculates the invariant features of the inner parts of the detected regions using Hu’s moments. It verifies the hypothesis first, before extracting the assigned speed limit from the detected region using a feed forward neural network. The proposed method was experimented on a number of traffic images and the results show that the region characteristics are more immune to different noisy conditions such as partial occlusions, cluttered backgrounds and deformations.