Rapid vehicle logo region detection based on information theory

• We combine bottom-up and top-down approaches to detect vehicle logos.
• Our method is capable of searching for the vehicle logo in an image without learning.
• Our method is adaptable to different situations without adjusting the parameters.

Vehicle logo detection is an important task in intelligent transportation systems. In this paper, a novel method is proposed for detecting the vehicle logo in an image. Our method consists of three main steps. First, horizontal and vertical direction filters are applied to the original image to produce two new images. Then, a saliency map is generated from each image. Second, two clusters in the corresponding saliency map are formed to create a binary image. Finally, the vehicle logo is localized by searching the regions with the maximum useful information. Our method has two main contributions. One is that the vehicle logo can be detected rapidly without learning. The other is that our method is adaptable to different situations without adjusting the parameters. A series of experiments are performed on 970 images, which are captured from different real-time situations. Experimental results show that our method is also very fast and can achieve a high detection rate, which is suitable for real-time applications.

Figure optionsDownload as PowerPoint slide