U.S. speed limit sign detection and recognition from image sequences

In this paper, we present a novel visual speed limit signs detection and recognition system for American signs. Firstly, rectangle detector is used to search rectangle candidate. Secondly, the Improved Stroke Width Transform (ISWT) is introduced to seek stroke width inside each rectangle candidate. Thirdly, the modified Connected-Component labeling is used to group these pixels into digit candidates. Fourthly, a method is presented for segmenting the digit candidates in order to make every digit candidate contains only single digit. Finally, every segmented digit candidate is recognized respectively based on the Support Vector Machine using Majority Voting strategy (MV). We call the SVM using the MV strategy MV-SVM. And, all the recognized digits are verified whether combination of them is speed limit or not according to rule set. The presented system is tested in different conditions, including sunny, cloudy, rainy weather and night, and the experimental results demonstrate that it is much efficient for detecting and recognizing rectangular speed limit signs.