dc.description.abstract | With the advancement of scientific technologies, the cost of digital camera decreases rapidly. It is a trend to improve and uplift the living quality of people using image processing techniques. Automatic detection of text from video is one of the applications which is an essential task for understanding and indexing of video. In this thesis, a driver assistant system is designed by automatic detection of text on road sign (guid signs and limit signs) on highway to provide drivers information for navigation, such as location, direction, and speed limit. It may also alleviate the load of driver who may lose his/her attention looking at road signs while focusing on driving. For 2008 Olympic in Beijing, there will be many foreigner visiting China and not all of them understand Chinese language. Hence, the translation of text on road sign is another goal that can be accomplished.
In this thesis, a set of feature is devised to detect road signs. The proposed system consists of three modules. The first module finds the constituting colors of road signs using the color transform model and locates road sign candidates. In the second module, affine transformation is performed to restore road signs which are captured by camera in different positions to let every road sign seems to be vertical to the camera optical axis. Moreover, affine transformation can improve the accuracy in detecting texts embedded in road signs. As to the third module, it performs the task of detecting texts on road signs. The method we adopt is canny edge detector to obtain clearer edge information.
Experiments were conducted on a variety of situations. 20 video sequences (sunny*10 and cloudy*10) including light variations and straight or cursive road conditions were tested to verify the validity of the proposed method. The recall and precision rates in locating road sign are 91.1% and 80.8%, respectively. The recall and precision rates in detection text are 93.6% and 88.0%. | en_US |