Lane recognition method based on full convolutional neural network

Abstract

In order to solve the problem of lane recognition in current vehicle automatic driving, a neural network model is planned to recognize a lane as a final solution. Compared with a typical computer vision technology in which a gradient threshold value, a color threshold value and other manual programming methods are used for processing pictures, the deep learning model can achieve a better effect inthe aspect of extracting key characteristic values of the images. In particular, a full convolutional neural network model, namely a neural network without any full connection layer is constructed; important specific pixel groups are found by using a filter, lane feature information in an image can be extracted very effectively, and finally output information can be displayed in a lane area recognized by an original image in a highlight manner, so that the purpose of lane recognition is achieved.

Description

technical field [0001] This patent belongs to the field of automobile automatic driving perception technology, in particular, it involves the scheme of identifying lanes with deep learning methods. Background technique [0002] Lane detection remains a fertile area of ​​machine vision research. Therefore, many methods have been proposed in the world to accomplish this task. However, the Hough transform method is still one of the most commonly used methods. In these methods, the input image is first preprocessed using a Canny edge detector or a steerable filter to find edges. The classic Hough transform is used to find straight lines in binary images, which usually correspond to lane boundaries. But with the Hough transform, it is often difficult to determine whether a line corresponds to a lane boundary. In color segmentation methods, RGB images are often converted to HSI or custom color spaces. In these alternative color spaces, the luma and chrominance components of a...

AI Technical Summary

Problems solved by technology

Moreover, these methods can exhibit performance degradation problems under unfamiliar conditions

Images