Abstract— In a visual driver assistance system, traffic sign detection and recognition are important functions. This paper presents automatic traffic sign detection and recognition systems based on neural networks and particle swarm optimization. Our system is able to detect and recognize all types of traffic signs used in Thailand, namely, prohibitory signs (red or blue), general warning signs (yellow) and construction area warning signs (amber). Traffic signs provide drivers with important information that help them to drive more safely and easily by guiding and warning them. The systems consist of four main stages: 1) color filtering according to color of RGB pixels; 2) color segmentation and traffic sign detection by black-white color transformation; 3) feature extraction; 4) traffic sign recognition based on classification techniques. Experiments show that system has high accuracy of traffic sign detection and recognition for the traffic signs used in Thailand. Keywords—Traffic sign detection and recognition systems, color filtering, color segmentation, neural networks, particle swarm optimization. I. INTRODUCTION RAFFIC-SIGN detection and recognition have been an important issue for research recently because they are becoming increasingly important for the development of intelligent vehicles. The first work in this area can be traced back to the late 1960s and significant advances were made in the 1980s and 1990s. An example of this early work was the Traffic Sign Recognition (TSR) project that was begun in 1988. In this project, color was ignored because insufficient processing power was available at that time to handle color information. The main work was based on the detection and extraction of relevant shapes, known as form primitives, from the images [1]. The idea of computer vision-based driver assistance attracted worldwide attention when video processing became more attainable. Originating from large-scale projects developed in the USA, Europe [2,3,4] and Japan [5], intensive research on traffic sign recognition is now being conducted by both academic and industrial groups all over the world, often in close association with the car industry [6]. In the majority of published work, a two-stage sequential T. Surinwarangkoon is with the Department of Business Computer, Suan Sunandha Rajabhat University, Bangkok, Thailand (phone: 087-276-9617; fax: 02-160-1494; e-mail: thongchaisurin@ yahoo.com). S. Nitsuwat is with the Department of Mathematics, King Mongkut’s University of Technology North Bangkok, Bangkok, Thailand (e-mail: sns@kmutnb.ac.th). E. J. Moore is with the Department of Mathematics, King Mongkut’s University of Technology North Bangkok, Bangkok, Thailand (e-mail: ejm@kmutnb.ac.th). approach has been adopted. The first stage aims at traffic sign detection, and the second stage aims at traffic sign recognition. For the traffic sign detection stage, a common approach is to first define the acceptable appearance of signs and the geometrical relationships between their parts with respect to color and shape [7]-[10] and then to use this information to identify the region of the roadside image that could contain a traffic sign. Different approaches have been used for different stages of the detection problem, such as color segmentation, control theory and feature extraction. For the traffic sign recognition stage, a pixel-based approach is often used and the class of a detected sign might be determined by cross- correlation template matching [11], [12] or classification techniques [7], [13]. In [14], [15] a different strategy was developed based on the idea of representing a candidate sign as a set of similarities to stored prototype images. Related research on the detection and extraction of text characters on road traffic panels can be found in [16]. A vehicle detecting and tracking system has recently been proposed for dynamic environments where shapes and sizes of vehicles are changing all the time [17]. For automatic traffic sign recognition systems, it is necessary to create templates of characteristic patterns for different classes of sign. A classification technique defines an object into a class by using object features. For traffic sign recognition, there are broadly 3 major methods, namely, color- based, shape-based, and classification techniques such as neural-network based recognition [18]. A computational intelligence technique, called particle swarm optimization (PSO), inspired by social behaviour simulation, has been proposed by Eberhart and Kennedy [19]. PSO is a simple but powerful optimization algorithm. In the last decade PSO algorithms have been developed and successfully applied to many problems in image analysis. In fact, image analysis tasks can often be reformulated as the optimization of an objective function. This paper focuses on the task of automatic traffic sign recognition from video and the use of that information in a driver assistance system. We use detection and recognition methods based on color filtering, color segmentation, neural networks and particle swarm optimization. Color filtering and color segmentation are used in the traffic sign detection stage. Two classification techniques: neural networks and particle swarm optimization, are used in the traffic sign recognition stage. Experimental tests have been carried out on the four main types of traffic signs used in Thailand to compare the A traffic sign detection and recognition system Thongchai Surinwarangkoon, Supot Nitsuwat, and Elvin J. Moore T INTERNATIONAL JOURNAL OF CIRCUITS, SYSTEMS AND SIGNAL PROCESSING Issue 1, Volume 7, 2013 58