Traffic visibility estimation based on dynamic camera calibration

In order to solve the problems of costly instruments and small detection area in visibility detection, a traffic visibility estimation algorithm was proposed by combing light transmission model in fog with camera calibration geometrical optics model. The algorithm calculated the distance from point in road area to the camera by calibrating internal and external parameters dynamically, obtaining the atmospheric extinction coefficient and estimating the traffic visibility using scene transmittance. Firstly, interesting area was searched by activity map, and the fog weather was recognized by fitting curve of area average pixels whether meet the edge spread function or not. Secondly, transmittance value of each point in traffic scene was calculated by dark channel prior, and four group points of transmission difference maximum were selected to calibrate the internal parameters of the camera. Thirdly, vanishing point and boundary of the road were extracted to calibrate the camera external parameters dynamically. Finally, traffic visibility was estimated by using the distance from the point in road area to the camera and corresponding scene transmittance. In this paper, visibility estimation results are compared with corresponding data got by manual and physical equipment. We also verify the effectiveness and real-time of this proposed method. ©, 2015, Science Press. All right reserved.