Article Preview
Top1. Introduction
Automatic Traffic Sign Recognition (TSR) is becoming an essential component of the new cars (Greenhalgh & Mirmehdi, 2012a; Greenhalgh & Mirmehdi, 2012b; Ruta et al., 2010; Xu, 2009; Kassani et al., 2016), it is an important application for Advanced Driver Assistance Systems (ADAS) (Wahyono, 2014; Hsieh et al., 2014; Rublee et al., 2011; Grana et al., 2013). The inability to observe road signs due to various conditions might result in road accidents. Many car accidents are caused by drivers’ lack of awareness or fatigue. Thus, early warning of the driver will keep him more attentive. Consequently, the integration of an automated embedded system of traffic signs recognition increases safety conditions.
A road traffic sign recognition approach can be composed of two stages: detection and recognition (Hsieh et al., 2014; Jung et al., 2016; Seo et al., 2015; Timofte et al., 2014). The purpose of the first step is to extract the signs from the whole scene. The second stage tries to recognize the symbols accordingly to a predefined database. Traffic sign detection and recognition in unstructured scenes have been drawing the attention of many researches, both theoretically and technically (Gudigar et al., 2016; Salti et al., 2015). One of the foremost difficulties of TSR applications is the understanding of the environment and orientation of the signs in real scenes. The first challenge is to consider urban scenes understanding under complex conditions (time of the day, weather, noise, occlusions, lighting changes, shadows, distortion, etc.) as presented in Figure 1.
Figure 1. Problems of traffic sign detection: Over-Illumination, under-illumination, occlusion, rotation and deterioration of the signs, deficiency, complexity scene and weather situations.
The proposed system is divided into two main stages: the detection and the recognition stages. In the detection step, the image is transformed to the HSV color space, and the Maximally Stable Extremal Regions (MSERs) algorithm (Greenhalgh & Mirmehdi, 2012a), (Greenhalgh & Mirmehdi, 2012b), (Salti et al., 2015) is used to detect all possible candidates. For the recognition step, the Oriented fast and Rotated Brief (ORB) descriptor (Rublee et al., 2011), (Grana et al., 2013) is used to extract features serving the recognizing process and enables the identification of signs. Thus, we try to integrate the methods while ensuring a compromise between processing time and accuracy. In the following, we justify the choice of HSV color space, MSER algorithm and ORB feature.