Training method of lane line detection model, image processing method and device

Abstract

The embodiment of the invention discloses a training method of a lane line detection model, an image processing method and a device. The method comprises the steps of obtaining a lane line sample annotation image; performing semantic segmentation on the lane line sample annotation image based on a first convolutional neural network model to obtain a first lane line semantic segmentation result; semantic segmentation is carried out on the lane line sample annotation image based on a second convolutional neural network model, a second lane line semantic segmentation result is obtained, and the number of convolutional layers of the first convolutional neural network model is larger than the number of convolutional layers of the second convolutional neural network model; calculating an output gap loss based on the first lane line semantic segmentation result and the second lane line semantic segmentation result; calculating self-attention feature loss among a plurality of convolutional layers in the second convolutional neural network model; and adjusting network parameters of the second convolutional neural network model based on the output gap loss and the self-attention feature loss to obtain a lane line detection model.

Description

technical field [0001] The present application relates to the technical field of image processing, in particular to a training method for a lane line detection model, an image processing method and a device. Background technique [0002] In the field of automatic driving, the understanding of the surrounding environment is crucial to safe driving, and real-time lane line position detection is a very necessary link, which can provide functions such as lane departure warning. [0003] The method based on traditional image processing has great limitations. With the increase of the complexity of road scenes, such as strong light, low visibility, shadows, occlusion and other problems, the effect is greatly reduced, and the calculation complexity is high, and the operation efficiency is low; based on The method of deep learning often needs to build a huge model to achieve high detection accuracy, but the additional computing overhead brought by it cannot meet the real-time detecti...

Images