Rainfall Recognition in Various Conditions by In-Vehicle Camera Image for Driver Assistance

In this paper, we propose a rainfall recognition method in various conditions from in-vehicle camera images using extracted image feature characteristic to rain. As a driver assistance system using an in-vehicle camera, we have been trying to recognize weather, espcially rainfall in various conditions. We recognize the rainfall by detecting the changes of image features caused by raindrops on the windshield, making use of different methods for day and night. In daytime, we make raindrop templates by principal component analysis from various raindrop images, and detect raindrops by template matching. We have previously proposed a raindrop detection method from the sky region in the image. Int this paper, we propose a method that detects raindrops from the whole image by averageing multiple input images. In addition, higher accuracy of raindrop detection is expected by matching the detected raindrops between frames. In nighttime, we propose a method that quantifies lights reflacted by raindrops. As a result, the same detection accuracy with number of the previous method was obtained (precision rate 0.97, recall rate 0.51) for daytime without restricting the target region. In nighttime, we obtained a the success rate of 83% rainfall judgment. Effectiveness of this technique was shown from these results.