基于大气反射-散射模型的复原图像中交通视频车灯检测

针对夜间复杂照明环境导致车灯检测率低的问题, 提出了一种基于大气反射-散射模型的复原图像中夜间交通视频车灯检测算法. 首先根据漫反射原理抑制路面漫反射光, 在对大气散射模型做了改进之后, 估计了大气散射模型中的大气光, 再根据暗原色先验理论估计环境光, 重新定义透射率, 从而得到了只含有车灯及反射区域的复原图像.为了进一步抑制该复原图像中的强光光晕, 再次利用暗原色先验理论重新估计环境光, 得到最终的复原图像. 最后对复原图像中的所有亮斑根据四类几何特征逐步筛选, 排除视野中的非车灯. 实验结果表明, 该方法在复杂雨雪天气、高密度及高速等不同情况下, 与同类先进算法相比具有较高的检测率, 较低的漏检率和误检率.

Headlights Detection in Traffic Videos Based on Atmospheric Reflection-scattering Model via Reconstructing Restoration Images

As the detection rate of headlights under complex lighting scenes in nighttime is relatively low, a novel algorithm of headlights detection based on atmospheric reflection-scattering model has been presented. The diffuse reflection light from road surface is firstly suppressed using the diffuse reflection principle. After improving the atmosphere scattering model, the airglow is estimated. The ambient light is then estimated with the dark-channel prior theory and the transmittance is defined again. A restoration image is obtained in which only headlights and reflection regions are contained. In order to suppress the halo produced by the highlights in the restoration image, the ambient light is estimated again with the dark-channel prior theory. All spots are then checked according to their four types of geometric characteristics step by step, non-headlights are finally filtered out. Experimental results show that under complicated weather, such as rain or snow, or under high population, or high-speed, the proposed algorithm has a higher detection rate, lower miss rate and false detection rate, compared with other similar advanced algorithms.