Shadow Suppression for Vehicle Target Detection in Open-air Expressway Scenes
Tong Zhou, Yuxuan Li, Mei Deng, Si-yuan Pang · 2021
The accurate extraction of the vehicle target area is key to the detection of expressway anomaly events based on video surveillance. In the expressway scenes, the existing vehicle shadow interference areas, often make the vehicle area distortion, expansion, connectivity or even lost. In addition, there is much more noise interference in the open-air images and most of the images' quality is lower, resulting in that the traditional shadow suppression method is still difficult to apply. In order to reduce the noise in open-air scenes, a new shadow suppression method is proposed, based on a combination of color grayscale feature and HLGP(Histogram of Local Gradient Binary Patterns) feature. The new algorithm first uses chromaticity and brightness similarity to determine the shadow preliminarily so as to solve the problem of misjudgment in the vehicle area. Then, based on the noise robustness LGBP and local gradient histogram feature, the HLGP feature is obtained, which is proved to be good robustness against illumination and applied to correct the results of color shading for reducing the false detection area. The experimental results indicate that the proposed method can overcome the noise interference and improve the accuracy of vehicle target detection.