Enhancement of the Detection for Intelligent Vehicle Systems - Case Rain/Snow

Abstract

Intelligent vehicle systems based on vision clearly have remarkably progressed. However, they still suffer from the degradation of results quality in case of unfavorable acquisition conditions as fog, rain, and snow. Few works, where the aim is to enhance the visibility and reduce the effect of weather conditions like rain and snow, can be quoted in the context of intelligent vehicle systems. In this work, the authors established a method to enhance vehicles detection, in case of rain/snow. This method gathers a succession of phases: the extraction of the characteristics of Gaussian susceptible fields -type Gaussian -, the separation of these characteristics in two objects classes according to the criteria of the area, the modeling of the small area in a form of ellipses where one of those parameters ‘orientation’ was exploited, the distinction between the small objects  that represent rain / snow and those that represent the rest of details of big objects based on probability laws, the elimination of small objects representing rain/snow and the exploitation of the big objects for the detection of vehicles. This detection enhancement method allowed a considerable increase in the detection rate.