Robust Marker Design And Recognition For Visual Location

Visual marker code is a kind of artificial visual feature designed to simplify the automatic detection of the machine.Visual marker code is widely used in many fields.With the increase in demand,higher requirements are placed on the design of visual mark codes,as well as the speed and robustness of recognition.This dissertation starts from two perspectives: increasing the robustness of the visual marker code during design and increasing the robustness of the visual marker code during detection and recognition.In the design process of visual marker code,based on the robust performance indicators,a random iterative method is used to design the visual marker code with good robustness.Considering that as the size of the marker code increases,the search space will be too large to search thoroughly,a mixed integer linear programming method is used to improve the generation process of the marker code and further increase the robustness of the visual marker code.The other process is the detection and recognition of visual marker code.First,a classic image processing method is given to complete the detection and recognition of visual marker code.Then,by analyzing the shortcomings of the classical methods,the detection and recognition methods are improved from the perspective of accelerating the speed of detection and recognition.Next,aiming at the complex situation of the visual marker code,especially the unrecognizable situation of the marker code when the boundary is occluded,a CNN(Convolution Neural Network)classification method for visual marker code detection and recognition is given.Finally,considering the requirements of camera pose information during actual positioning,we propose a border detection method based on target detection.Based on the original image processing,our method achieve a more robust visual marker recognition task.The final experimental results show that the robustness of visual marker code in this dissertation is better than April Tag.In the process of detection and recognition,the improvement of robustness is discussed from the perspective of boundary occlusion.The results show that the CNN classification method can achieve 97.1% accuracy,which can better handle the problem of visual marker boundary occlusion.When the boundary is occluded,the result shows that we can obtain the position of the visual marker in the image based on the border detection method.Therefore,this method can well solves the boundary occlusion problem of the visual marker code.