Research Of Circular Marker Recognition Algorithm Based On Deep Learning

Circular markers were commonly used in the field of computer vision measurement due to their distinct geometric features and higher positioning accuracy.At present,many methods have been proposed for the recognition of circular markers.However,most of these methods are based on traditional image processing techniques for recognition,which is easily affected by external interference,causing the inability to accurately identify markers.For this reason,this paper combines deep learning technology to study the recognition problem of circular markers.The main research contents and innovations are as follows:(1)A circular marker detection algorithm based on improved Tiny-yolov3 is studied: Because of the problem that the traditional methods are susceptible to interference when detecting circular markers,this paper introduces the Tiny-yolov3 algorithm,and realizes robust detection of circular markers by improving its network structure and corresponding loss constraints.Experiments show that compared with traditional marker detection algorithms,the algorithm proposed in this paper is higher detection accuracy and significantly improved robustness.(2)A center extraction algorithm of circular marker based on contour segmentation is studied:Aiming at the problem that traditional algorithms are difficult to detect the edge of markers and affect the center extraction when it is subject to interference by abnormal brightness or motion blur,this paper studies a marker contour segmentation model to deal with.The experiment results show that the algorithm studied in this paper can accurately and stably extract the contour of the circular marker and calculate its center when it is disturbed by the above.Compared with the traditional algorithms,the proposed algorithm is not only more robust,but also more accurate in extracting the center of the marker.(3)A center recognition algorithm of circular marker based on the end-to-end model is studied:The circular marker recognition based on the traditional step-by-step method is not only complicated in the operation process,but also has the risk of introducing external noise.To solve this problem,this paper studies an end-to-end model which maps the center of the marker directly through the network.The experiments based on synthetic and measured data show that the proposed algorithm can identify the center of circular markers better than the traditional step-by-step center recognition algorithm.At the same time,when using this algorithm for identification,the recognition process is greatly simplified and the center of the marker can be obtained in only one step.The recognition speed of the algorithm is significantly improved.