Real-time daytime road marker recognition using features vectors and neural network


Md Sani, Zamani and Loi, Wei Sen and Ab. Ghani, Hadhrami and Besar, Rosli (2016) Real-time daytime road marker recognition using features vectors and neural network. In: 2015 IEEE Conference on Sustainable Utilization And Development In Engineering and Technology (CSUDET). IEEE Xplore, pp. 1-6.

[img] Text
Restricted to Repository staff only

Download (419kB)


Road markers provide vital information to ensure traffic safety. Different sets of markers are normally used between the highways and the normal road. At the normal road for example, the double lane markers are used to indicate the hazardous area, where overtaking is prohibited while broken marker lane indicate otherwise. To avoid traffic accidents and provide safety, these markers should be accurately detected and classified, which is best solved via vision detection approach. Marker type classification is however affected by the changing sun illumination throughout the day. In this paper, real-time recognition of these markers is developed using the artificial neural network (ANN) to alert the users while driving. The accuracy of the scheme is observed when different input features (geometrical and texture) and image pixels are fed for recognizing broken and double lane markers. A very high accuracy result with low error rate is obtained at 98.83% (10-fold cross validation) accuracy detection using additional features, compared with ~95% by using only the image pixels as the input vector and average processing time is at ~30ms per frame.

Item Type: Book Section
Uncontrolled Keywords: Roads, Videos, Feature extraction, Image edge detection, Artificial neural networks, Cameras, Image color analysis
Subjects: T Technology > TA Engineering (General). Civil engineering (General)
Divisions: Faculty of Engineering and Technology (FET)
Depositing User: Ms Rosnani Abd Wahab
Date Deposited: 15 Nov 2017 18:17
Last Modified: 21 Dec 2020 05:55


Downloads per month over past year

View ItemEdit (login required)