Traffic sign and license plate detection based on saliency, meanshift, and mathematical morphology
Loading...
Date
2015-09-01
Authors
Cheong; Wei Sheik
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
An object detection model that consists of cepstrum saliency, mean-shift segmentation,
and morphological shape estimation is proposed in this research. An improved
computational saliency method based on human visual attention is introduced. Cepstrum
saliency is based on the principles of de-convolution in the log-spectrum domain, and is
computationally fast with only single parameter to tune. Moreover, cepstrum saliency
exhibits color consistency under various illuminations, where the normalized RGB color
scheme can be used for color images. To further enhance the proposed object detection
model, non-parametric mean-shift and Otsu’s method are utilized for figure-ground
segmentation. Besides that, simple shape factors based on mathematical morphology are
introduced to identify the segmented objects by measuring shapes. To evaluate the
effectiveness and applicability of the proposed method, two problems in the transportation
section, i.e., traffic sign and license plate detection, were studied in detail. Based on two
publicly available and locally collected data sets, the proposed detection method
demonstrates a good equipoise between accuracy and speed. The simulation results indicate that it is seven times faster than shape descriptors in traffic sign detection, and has an average of less than 0.6 s in license plate detection as compared with template matching and machine learning methods. The findings indicate the usefulness of the proposed object
detection method in providing a unified framework for both traffic sign and license plate
detection problems; therefore contributing towards advancement in intelligent transportation systems.