Research Of Sea Ice Detecting Based On Image Method

In cold areas marine project, detecting of sea-ice condition is an important content. Considering the actual situation, the measurement methods of sea ice intensity, thickness and speed based on video and image are proposed and implemented. Compared with the traditional methods, the equipment and system requirements are relatively low, and the efficiency and accuracy can both meet the needs of practical applications.In the intensity aspect, a method of MATLAB and .Net hybrid programming is applied to the monitoring system. It is proposed that the images can be compressed and gray-scale equalized before calculation. The two-dimensional Osto algorithm is also be simplified so the speed is accelerated a lot with a small loss of the accuracy. The software is developed based on the BS model, the efficient is improved in addition to the easily background updates.In the thickness aspect, the image contains the ice section has to be extracted from the video. Then the ice thickness can be calculated using the projection principle after calibration of the system. Since the total calculation amount is small, we developed the software using the CS model. The system calibration function is applied in the system and we make a error analysis in the end. As same as the intensity part, the thickness part of the system provides visual interactive graphical interface and another powerful tool for measurement and calculation of sea ice situation is provided.In the speed aspect, the sea ice video is decomposed to image sequence at first, and then the Harris corners are extracted as the feature points of the images. To match the feature points between different images, a simple gray-correlation function is selected as the matching method. After matching the feature points between images at different time, the distance and direction of the feature points moving can be calculated. Harris corner extraction method is simple, efficiency and flexible. Besides, the number of the feature points can be assigned by the user. However, obvious mismatches would be appeared if the images are fuzzy or unclear. So, a new method to eliminate the mismatches points using a greedy GMM model is proposed and implemented. Experiments and applications show that this method has a good robustness.