A method for constructing a road detection model and a road detection method

Abstract

A method for constructing a road detection model comprises the following steps of extracting the road area characteristics from a preset training image set by using a preset encoder to obtain tensorswith a first preset size corresponding to each training image; carrying out convolution layer operation on tensors of a first preset size corresponding to each training image by utilizing a preset decoder to generate a characteristic heat map of a second preset size, and generating training result images which are the same as the training images in size and are in one-to-one correspondence with the pixel points according to the characteristic heat map through deconvolution layer up-sampling operation; and adjusting model parameters of a preset encoder and / or a preset decoder according to the training result image and the preset training image set, and finally obtaining a required road detection model. According to the method, the blank of unmanned mine road detection is filled, the problemthat a road detection method based on Haar texture characteristics combined with vanishing points cannot be implemented due to unclear water stains and edges can be effectively solved, and the accurate positioning of the mine road based on pixel points can be realized.

Description

technical field [0001] The invention relates to the technical field of intelligent driving, in particular to a method for building a road detection model and a road detection method. Background technique [0002] The edge characteristics of the road in the open-pit mine environment are not obvious. During the driving process, it is necessary to identify roads with complex working conditions such as uphill, downhill, muddy roads, and many water marks. higher. [0003] In the existing technical solutions, there is no road detection method for roads in environments such as open-pit mines where the color characteristics are not obvious, there is no clear boundary, there are large water marks after rain, and the rut marks are deep. However, the general-purpose road detection method cannot specifically solve the performance and real-time road detection requirements suitable for open-pit mines. Contents of the invention [0004] In order to solve the above problems, the present...

AI Technical Summary

Problems solved by technology

[0003] The existing technical solutions do not have a road detection method for roads in environments such as open-pit mines where the color characteristics are not obvious, there is no clear boundary, there are large water marks after rain, and the ruts are deep

However, the general-purpose road detection method cannot meet the performance and real-time road detection requirements suitable for open-pit mines.

Images