Study On Visibility Estimation Using Imagery From Freeway Monitoring System

Low visibility not only affects the capacity of road, but also easily leads to accidents, threatening the life and property safety, especially can arise much greater safety hazards in the freeway which has higher speed. Traditional methods of detecting visibility includes human eye observation and measuring with specific meteorological instruments. The former has a strong subjective. The latter has a high cost on installation and maintenance, so laying the instruments in the freeway with high density is difficult. With the traffic monitoring system being widely used, the collected video resources can provide the data foundation for study visibility detection based on imagery. Therefore, the visibility estimation in freeway using imagery has became an important issue.This paper mainly studies on daytime visibility estimation. On basis of relevant theories, the paper analyzes the demand and testing feasibility of visibility detection in the freeway, then proposes the visibility estimation algorithm based on imagery, which locates the farthest point of view on imagery by using imagery brightness contrast, and gets the actual value of visibility from the farthest point of view by camera calibration. Then, points out two key issues of the algorithm: the detection for region of interest(sky & road) and camera calibration.Firstly, the methods on detecting the region of interest are studied, and this process aims at building a foundation for brightness calculation and camera calibration. Adaptive thresholding and region growing method are used for the sky region detection, and the both methods can achieve a better result on the sky areas detection; In the road area detection, the method of road edge detection based on linear model is proposed, then basing on Hough transform, an algorithm to reduce redundancy in detecting edges is also proposed. This method can detect a better road region and identify road format.Secondly, the camera calibration in freeway is studied based on the pinhole camera imaging mode. Additionlly combining with the size of freeway pavement structure, the camera calibration is achieved based on lane dividing lines. Then, the brightness parameters of the sky and road are calculated based on the image respectively, and the farthest view point is fixed by using human visual contrast threshold which is 0.05, so that the value of the visibility can be acquired with camera calibration.Finally, the visibility estimation algorithm is verified by two sets of imagery. The results in experiments show:(1)The algorithm is applicable for standard definition(SD) and high definition(HD) imagery;(2) The algorithm can meet the demand for highway visibility detection;(3)The result is better base on HD imagery;(4)The detection accuracy is higher when the road is dry.