Road detection based on superpixels and convolution neural network

Abstract

The invention provides a road detection method based on superpixels and a convolution neural network, and the method comprises the steps: carrying out the image preprocessing through a simple linear iteration cluster algorithm, and segmenting an image into superpixels which are uniform in size; carrying out the automatic learning of characteristics which facilitate the classification most based on superpixel blocks through employing the convolution neural network, and training the network; carrying out the classification of a road region and a non-road region through employing the trained convolution neural network; and carrying out the optimization of a classification result through a Markov random field according to the relation between the superpixel. Compared with the prior art, the method can effectively detect the road region in a complex environment, and is better in robustness.

Description

technical field [0001] The invention relates to image processing, video monitoring and driving safety, in particular to a road detection method and device. Background technique [0002] Driver assistance systems can reduce the incidence of traffic accidents by giving drivers reminders and guidance. Vision-based road detection is the key to driver assistance systems, it can provide clues for obstacle detection, and it is the basis for scene understanding in driverless driving, which is conducive to path planning. [0003] Generally, roads can be divided into structured roads and unstructured roads. Structured roads refer to roads with obvious road marking lines, clear road boundaries, and special color information, such as expressways, urban roads, etc. The road detection problem can be simplified as road marking line detection. Mature. Unstructured roads refer to roads with a low degree of structure and no clear lane lines and road boundaries. The detection method for uns...

AI Technical Summary

Problems solved by technology

[0007] In order to achieve the above purpose, the present invention provides a road detection method based on superpixels and convolutional neural network for the problem of poor robustness of road detection in complex environments, the method comprising:

Images