Location And Recognition Of Non-Standard DM Codes For Navigation Based On Neural Network

At present,the positioning method of two-dimensional code in pictures is of low accuracy,and the two-dimensional code itself has the problem of waste of internal information resources,so the use of two-dimensional code for navigation and positioning technology of unmanned vehicle(AGV)is not the mainstream.If a non-standard DM code is adopted in navigation and a new two-dimensional code positioning method is proposed,the above problems can be effectively solved and the cost of navigation system can be greatly reduced.In this paper,a region extraction method for non-standard DM codes based on SegNet neural network and a precise localization method combining edge detection and Radon transform are proposed.The basic steps of the region extraction method are as follows : establish a non-standard DM code image data set containing 5000 images and a SegNet neural network training label,and take 4000 images in the image set as the training set and 1000 images as the test set;build a neural network with eight-layer structure,and use the random gradient descent method to train the SegNet neural network with images in the training set,so that it can correctly identify the two-dimensional code region;the trained SegNet neural network was used to test the images in the test set,and the obtained two-dimensional code region was morphologically opened and closed to complete the region extraction process.Experimental results show that the algorithm proposed in this paper has higher robustness,stronger adaptability and better recognition rate than other algorithms based on BP neural network and convolutional neural network..The e trained SegNet neural network achieves 98% accuracy for the whole test set of two-dimensional code region extraction,and the two-dimensional code recognition rate reaches 99.5%.The basic steps of precision positioning method are : detect the edge of Canny operator in the extracted two-dimensional code area;carry out edge tracking and corner point detection on the detected image to determine the "L" shaped edge line of thetwo-dimensional code and the position coordinates of its three vertices;through the linear range of plus or minus 22.5 ° in the edge only calculate Radon transform determines the location of the fourth vertex coordinates.Experiments show that this method can accurately locate the non-standard DM code and reduce the computation by nearly 30%.Finally,a method of inverse mapping transformation is proposed to correct the distorted non-standard DM codes.And on the basis of Zxing library,a two-dimensional code recognition interface is completed by using C++ language programming,which successfully converts the two-dimensional code information into a computer-readable language and greatly reduces the reading time.Using the computer as the experimental platform only requires 168 ms.