A novel algorithm to detect and extract the texts of road-sign plates in video scenes

Abstract EN

Traffic Road-Signs contain useful information for the road users; the operation of many of modern applications like the automatic or smart vehicle requires an automatic discrimination of the texts of the traffic road-sign.

Discrimination of text compose of several stages, the first of these stages is detection and extraction of the texts.

In this work, an algorithm is developed to detect, locate, and segment of the texts and the word in the video clips, that existing in the road signs in the city of Baghdad.

The proposed approach includes two stages, the first one is processing the image to locate and extract images of the road-sign and neglect the rest of the image, and the second stage is processing the image of the road-sign plate to detect and extract the texts without symbols and shapes.

The basic structure of the algorithm depends on the following functions: edge-detector, dilation, and filling-hole, morphological-opening.

The total recall values of 89%, the total precision value of 93%.The algorithm is, then, tested based on video clips, implementation of the algorithm based on video clips confirms its ability to detect the texts which may appear in the video scenes, recall-rate(r) is excellent with a total value of 94.5% and a total precision value of 86.5%.

The algorithm is tested to measure its validity to work under real-time operation, by processing one frame and exceed a set of next frames, the test appears that the algorithm is able to work under realtime operation with recall-rate(r) of (93%).