Construction And Realization Of Night Traffic Video Lamp Restoration Model

Nighttime traffic accidents occur frequently in recent years, and the accident rates at night have been higher than that in daytime.Due to the complex environmental factors, it becomes difficult to locate the nighttime vehicles accurately. Therefore, nighttime vehicles detection and tracking research becomes one of the difficult and attractive filed in the intelligent transportation system. The interplay of various lights in the nighttime traffic environment enhances the reflected light of subject’s surface and it leads to the difficult extraction of vehicle headlights. The darker illumination light affects the detail extraction of vehicles, such as color, contour etc. The factors mentioned above increase the limits of vehicle detection and tracking.This article focuses on the complex nighttime traffic situation, and then extracts the headlight’ features by analyzing the specific scenarios details according to the reality, and ultimately achieves the vehicles location and tracking.The main feature of nighttime vehicle detection is headlights or taillights, by detecting them to achieve detecting and tracking moving vehicles in videos. This is because, in generally, the characteristics of the lamp is relatively stable, its’ illuminance values almost not affected by outside light. But under the nighttime complex environment, such as interference of strong reflection lights in rain and snow pavement, floor interference of brighter lights, sheltered vehicle lights, distant lights adhesion, etc., lights or similar lights seriously interfere headlights extraction, accurate extract all lights in perspective has great difficulties. To solve those problems, this article propose a novel algorithm of headlights detection based on atmospheric reflection-scattering model. This method analyze all types of lights’composition and mechanism under nighttime complex traffic environment, construct atmospheric reflection-scattering model. Diffuse reflection light from road surface is firstly suppressed based on the diffuse reflection principle, it is the main cause of brighter background. Then similar lights will be further suppressed based on atmospheric scattering model, atmospheric particulates’influence that follow Bouguer attenuation, make the process of light propagation in the air of this part lights has a certain degree attenuation. The image that surveillance camera received is the result of light attenuation, in order to get the restored image, it’s necessary to suppress attenuation of light and similar lights when propagating in the atmosphere.When using atmospheric scattering model, it need to estimate the unknown parameters in this model, such as ambient light, airglow and atmospheric transmittance:The ambient light is estimated based on the dark-channel prior theory; Atmospheric light in nighttime environment has dropped weaker and its estimated value close to zero, almost equal to constant; Transmission refers to the reflected light that after transmission and attenuation, then reaches surveillance cameras. Herein transmittance is defined as the proportion of ambient lights cent of all lights. Finally, according to the modified atmospheric scattering model, the latest restored image will be obtained. Extract all the lights in this recovery image, further screening will be needed, so that lights and strong road reflection lights could be separated, achieve the purpose of detecting lights.