Research On Image Matching Algorithm Based On Deep Learning

Image matching is one of the key technologies in the field of image processing.In recent years,with the rapid development of technology,image matching technology has been applied to robot autonomous navigation,single and double vision,remote sensing positioning,image stitching and many other fields.In the multi-objective environment,the accuracy of the traditional matching algorithm is seriously degraded due to the rich external texture,which directly affects the positioning accuracy based on feature point matching.Therefore,the matching accuracy is extremely high.In order to improve the matching accuracy in multi-target environment,based on deep learning,a de-missing matching algorithm is proposed.The whole algorithm process includes two processes:rough matching and fine matching.The coarse matching identifies the target in the multi-target image by the Yolov3 target detection algorithm,and obtains several prediction frames and label information,and directly removes the matching pairs with different labels but matching success,and also eliminates the matching between the prediction box and the prediction frame..The fine matching is performed on the basis of the matching pairs selected by the coarse matching.First,the feature point B of the image to be matched is rotated around its image center P,and the rotation angle?is the angle between the principal divection point of the feature point B and the feature point A whose matching is successful,the angle is positive,clockwise rotation,and the angle is Negative,counterclockwise rotation,get point C,then calculate the angle?₃ of the matching point A and the line segment OA composed of the matching image center point O,and calculate the PC line segment angle?₄ between the C point and the P point obtained by the rotation,and calculate The angle difference is preset by a certain angle threshold.If it is greater than the angle threshold,it is culled,and vice versa.For all matching pairs in the rough match,the above operations are performed to obtain the final matching pair.Finally,based on the Sift algorithm and based on the OpenCV open source algorithm,the standard test set is simulated.The experimental results show that the algorithm greatly improves the accuracy of the Sift image matching algorithm.