Tunnel identification model and method based on multi-scale edge feature detection

Abstract

The invention discloses a tunnel identification model and method based on multi-scale edge feature detection, and the model comprises a data set construction module which is used for constructing a data set for model training by using the images of various types of driving roads; a training sample loading module which is used for loading training samples in batches during training, cutting the images into the same size during loading, and inputting the images into a recognition network; and an identification network which is constructed based on a residual network Resnet34, uses an edge binaryimage as a label during network training, leads the network to learn edge features during image feature learning, and cooperates with the multi-scale edge features so as to predict the tunnel more accurately.

Description

technical field [0001] The invention relates to the technical field of computer image processing, in particular to a tunnel recognition model and method based on multi-scale edge feature detection. Background technique [0002] At present, the main method for processing road images captured by cars in autonomous driving and intelligent monitoring is semantic segmentation, which classifies all pixels in the image and obtains the location information of environmental objects such as roads, pedestrians, and cars. Help the car to dodge and drive. However, image semantic segmentation methods usually use fully convolutional network methods such as FCN, UNet and other neural networks. After inputting the road image, these networks output a segmented image of the same size, which is for each pixel in the image about different targets and Background classification. Although these networks combine deep information with shallow information, they do not make full use of the multi-scale...

AI Technical Summary

Problems solved by technology

However, image semantic segmentation methods usually use fully convolutional network methods such as FCN, UNet and other neural networks. After inputting the road image, these networks output a segmented image of the same size, which is for each pixel in the image about different targets and Background classification. Although these networks combine deep information with shallow information, they do not make full use of the multi-scale information contained in the output of different layers of the network, so there will be deficiencies in the recognition of certain targets, such as When the road environment changes, such as a tunnel in front, the recognition accuracy of the network will not be ideal

Images