Automatic Reading of a Trade Gyrocompass Using Image Processing and Neural Networks

Gyrocompass is used to justify and aim (orient) the missiles; it is used in equipment and accuracy of launch position as well to find azimuth in particular points. Observations and reading in this kind of compass are done optically which suffers from low accuracy and optical illusion. In this paper a simple algorithm is suggested by utilizing methods based on processing digital image and neural networks for automatic reading of a trade gyrocompass. In the method at first, by analyzing the difference of sequential images taken from the compass display, the back frame is specified. Then the display region is extracted by edge detection and morphology operation. After that the segmented numbers are recognized one by one through the trained neural networks. Finally, the gyrocompass angle is calculated accompanied by accuracy of arc second by geometric analysis of relative position of scaled board and indicated index. The usage of suggested method in the step of back frame recognition has been evaluated on 6 real test video sequences and the next steps have also been evaluated on more than 300 real images.