This thesis presents a new approach, based on human visual perception, for detecting and recognising traffic signs under different viewing conditions. Traffic sign recognition is an important issue within any driver support system as it is fundamental to traffic safety and increases the drivers' awareness of situations and possible decisions that are ahead. All traffic signs possess similar visual characteristics, they are often the same size, shape and colour. However shapes may be distorted when viewed from different viewing angles and colours are affected by overall luminosity and the presence of shadows. Human vision can identify traffic signs correctly by ignoring this variance of colours and shapes. Consequently traffic sign recognition based on human visual perception has been researched during this project. In this approach two human vision models are adopted to solve the problems above: Colour Appearance Model (CIECAM97s) and Behavioural Model of Vision (BMV). Colour Appearance Model (CIECAM97s) is used to segment potential traffic signs from the image background under different weather conditions. Behavioural Model of Vision (BMV) is used to recognize the potential traffic signs. Results show that segmentation based on CIECAM97s performs better than, or comparable to, other perceptual colour spaces in terms of accuracy. In addition, results illustrate that recognition based on BMV can be used in this project effectively to detect a certain range of shape transformations. Furthermore, a fast method of distinguishing and recognizing the different weather conditions within images has been developed. The results show that 84% recognition rate can be achieved under three weather and different viewing conditions.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:524071 |
| Date | January 2005 |
| Creators | Hong, Kunbin |
| Publisher | Middlesex University |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://eprints.mdx.ac.uk/6547/ |
Page generated in 0.0019 seconds