Visibility Estimation Using Traffic Surveillance Video

Since atmospheric visibility has a significant impact on the behaviors of people and cars in the traffic, visibility monitoring has been an important task of intelligent transportation systems. Poor visibility conditions not only adverse to the health of people, but also easily cause traffic accidents, which threaten the safety of public transportation. Therefore, the detection and warning of low visibility has particular importance. Traditional methods of measuring visibility include human eye observation and measuring with specific instruments. The former has a strong subjectivity which leads to a poor accuracy, and the latter is costly with a high operational complexity. Hence, the improvement of the visibility measurement technology has practical significance and wide applications.Based on the studies of visibility associated physical model, this thesis presented the visibility estimation methods based on image understanding under different application scenarios and derived its corresponding measurement principle. After that, we proposed a visibility classification method using traffic surveillance video. To be specific, the algorithm extracts two features, the average Sobel gradient (ASG) and the dark channel ratio (DCR) from the image, and uses them to construct a visibility level estimation model, and finally employs a histogram correction module and a sunnv detector module to revise the result. A lot of experiments are performed to demonstrate the validity of this method.In addition, this thesis developed a road environmental information integration platform to reflect the road environmental information more comprehensively. The integrated information includes visibility level, temperature and humidity. The platform has three main functions:Firstly, it can estimate the atmospheric visibility level with the traffic surveillance video as input. Secondly, it can read the temperature and humidity which is detected by the temperature and humidity sensor through the USB interface. Thirdly, it can realize integration and visualization of the detected temperature, humidity and visibility level by video watermarking.