Research On Traffic Image Enhancement Technology Under Low Visibility Condition

With the rapid development of information technology, intelligent transportation has become an inevitable trend for the development of transportation management. A traffic Monitoring System is an important part of an Intelligent Transportation System(ITS) and it has many advantages: being networked, intelligent, systematic and automated, etc. It allows traffic managers not to manage at spot or conduct a frame-by-frame analysis of video recordings. It can find the offenders directly by live images. But in practical applications, traffic monitoring systems are set based on sunny visibility. Influenced by low visibility conditions such as haze and night, it is hard to capture the high quality images and make effective image processing.In order to improve the low visibility image contrast so that the collected images can be used to effectively identify vehicles, highway sign and marking, this paper studied the 2 kinds of haze and night images enhancement. We proposed algorithms for haze image enhancement and night image enhancement. The main research achievement and contents are as follows:1) According to the characteristics of the fog haze traffic image, this paper proposes an image enhancement algorithm based on wavelet transform and retinex by two bilateral filters. Frist let the V layer of an HSV image be decomposed by a wavelet algorithm. Then the low frequency part is processed by a coefficient enhancement. The high frequency part is applied with a threshold denoising-algorithm. This process not only reduces algorithm complexity, it also keeps the image details. When dealing with adjacent pixels, the bilateral filtering approach considers the vicinity relation and luminance similarity between image pixels. The filtering eliminates mustiness in image edges and Halo artifacts in images. Then, the V layer is reconstructed, and the image contrast is enhanced. At last, a saturation enhancement makes the image more bright-colored. Experiment results show that our algorithm effectively improves the brightness and contrast than other enhancement algorithms. In addition, it keeps more edge details, and also enhances the brightness to a certain degree. In summary, it has a very good effect on traffic images.2) Night traffic images, especially those traffic intersection images lack of sufficient light, are more complicated than haze images. Due to a serious shortage of light, these images contain large amounts of dark connected area. We propose an image enhancement method to handle the night traffic images. Unlike the haze image enhancement algorithm, before the V layer is decomposed by wavelet, we let it take a global contrast enhancement. This process enhances the overall brightness of the input image. The inverse image has similar characteristics like a fog haze image. So the haze image enhancement method can be used for processing the inverse image. After a wavelet decomposition, the low frequency part and its inverse image are manipulated by a MSR algorithm of adaptive weights, respectively. The MSR algorithm can better adjust brightness. In order to better show fixed scenes such as road surfaces in a night image, we extract the image background from a day image on the same scene. The low frequency part of the day image is added to the low frequency part of the night image. In this way, the image’s fixed scene is easily extracted. Since we process only the low frequency part, the details of the night image would not be effected. Experiment results show that this algorithm effectively improves the brightness and contrast. It also makes the image color appear natural and eliminates the Halo artifacts in images.