Research On Algorithms Of Video Image Visibility Detection

Visibility is an important index of meteorological observation,the condition of very low visibility would have great impact on meteorological observation, transportation, military and agriculture fields etc. In these fields,the greatest impact is the traffic field, highway often occur traffic accidents due to the low visibility weather,a considerable part of these are major traffic accidents of multi car chain collision,these pose a great threat to the safety of people's life and property. In order to solve this problem, it requires visibility detection equipment installed on the road intensively,to ensure traffic safety through the timely warning in low visibility conditions, traditional visibility detection equipment can not meet the requirements because of the complex operation and expensive price. Video visibility detection has the advantages of low cost, simple operation and high reliability by many researchers favor. In this paper, I have done some improvements on video image visibility algorithm on the basis of existing research results, the main research work is as follows:The thesis studied the dual differential luminance algorithm, completed the measures of the visibility by this algorithm, and analyzed the causes of error, when the object is behind the black shade and the sky background gray distribution is not uniform, the traditional dual differential luminance algorithm will cause larger error detection, an improved algorithm has been proposed to solve this problem, segment the sky region of the image using region growing algorithm, and average the pixels' gray of this region, experimental results show that the improved algorithm can detect more accurate visibility.The thesis studied the dark channel prior algorithm, use dark channel prior knowledge to calculate the transmittance, and use the camera calibration technology to achieve 3D reconstruction using the camera calibration technology, get the distance information, then calculate the visibility. In order to solve this problem of the estimated transmission value is smaller than the actual value, propose an algorithm to adaptive the segmentation threshold using k-means clustering algorithm in order to segment the sky region accurately, and fix the sky area transmission. For the atmosphere optical brightness value that using to estimate the transmission is estimated value will be larger than the actual value when the pure white objects in the image, this thesis proposed an improved algorithm using the thresholds to estimation the atmosphere optical brightness value. Experiments show that these improvements can do well in getting more accurate visibility.Do the experiments to detect visibility in the dual differential luminance algorithm, the improved dual differential luminance algorithm, the dark channel prior algorithm and the improved dark channel prior algorithm, use the video image data includes several kinds of weather, the experiments show that the improved dual differential luminance algorithm and improved dark channel prior algorithm compared with the traditional algorithms both are more accurate and reliable, comparison of these algorithms, the conclusion is the most accurate detection algorithm is the improved dark channel prior algorithm, and it can adapt to all kinds of different weather.