Storm Motion Analysis Based On Doppler Radar Image

Due to the greenhouse effect in recent years, severe weather such as thunderstorms, hail and tornadoes often happened, which has larger damage to human beings'production and life. The Doppler radar warning system based on the modern computer technology and information-processing technology is the most effective means for severe weather's accurate detection, analysis and forecasting. Therefore, China has built this warning system in the regions where severe weather often happens. Experience is accumulated in the practical application and the accuracy rate is improved constantly. The key technology in this research is to study and explore the data processing and algorithm optimization of Doppler radar image in the storm movement.Based on computer vision theory, the optical flow as a movement estimation method is adopted to analyze the motion vector filed of radar image at first time, and the level set method in the curve evolution theory is also used for tracking the storm contour at the first time.After processing the five years'observation data from Beijing and Tianjin radar, this thesis selects some typical examples as the experimental data and achieves the improvement and experiment of all the algorithms on the VC++ development platform, in the Windows environment.In the use of optical flow method, the best parameters of different optical flow algorithm are determined while analyzing the motion vector field of radar image through experiments, these algorithms are the matching method, the global smoothing method and the local smooth method, meanwhile, this thesis improves the local method using block-interpolation method, significantly increases the computing speed. The vector fields generated by the three methods are also compared, the experimental results show that the local method is better on smoothing and accuracy. Finally, qualitative and quantitative comparison of the above three methods are analysed using an improved extrapolation forecasting method and the table evaluation method, the results also indicate that the local method is superior to the other methods.In the use of level set method, this paper solved the problem of inconsecutiveness after curve (border) evolution,and experimentally proved that radar images tracking can get an excellent result using the improved level set method.