A feasible road segmentation method based on an improved Unet network model

A network model and road technology, applied in biological neural network models, neural learning methods, image analysis, etc., can solve problems such as gradient disappearance, achieve real-time accurate segmentation, and improve road segmentation accuracy.

Active Publication Date: 2019-06-04
TIANJIN UNIV +1
View PDF14 Cites 27 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] The invention provides a feasible road segmentation method based on the improved Unet network model. The invention introduces the residual module of ResNet in the feature extraction and deconvolution stages, which solves the problem of gradient disappearance in the model training pro...

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • A feasible road segmentation method based on an improved Unet network model
  • A feasible road segmentation method based on an improved Unet network model
  • A feasible road segmentation method based on an improved Unet network model

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0035] A feasible road segmentation method based on the improved Unet network model, see figure 1 and figure 2 , the method includes the following steps:

[0036] 101: Modify the convolutional layer in the Unet training model to a ResNet residual module, and use the cross-entropy loss function and the Lovász hinge loss function for backpropagation training;

[0037] 102: Use the trained model to segment the feasible region of the road scene without segmentation labels;

[0038] 103: Scale, contrast, and normalize the training picture and the corresponding segmentation mask label, and enhance the training sample, use the average histogram of the training data as a template, and perform histogram matching on the test data. Test, output the result of road segmentation.

[0039] Further, in step 101, the convolutional layer in the Unet training model is modified to a ResNet residual module, and the cross-entropy loss function and the Lovász hinge loss function are used for bac...

Embodiment 2

[0045] Combine below Figure 1-Figure 4 The scheme in Example 1 is further introduced, see the following description for details:

[0046] Step 1: Training data preprocessing

[0047]The original road image and the corresponding segmentation mask label are converted to a size of 101*101, and the contrast is enhanced. After the color image is grayscaled, the grayscale pixels are converted into 0-1 to reduce the scale of the input feature. Then increase the training sample size through horizontal, vertical, and 90-degree rotation enhancement processing.

[0048] Wherein, the specific operation of this step is well known to those skilled in the art, and will not be repeated in this embodiment of the present invention.

[0049] Step 2: Model Construction

[0050] Build a basic Unet network model (see figure 1 ), where the feature extraction stage and the deconvolution stage are divided into four stages of feature layers according to pooling and the last sample. The feature lay...

Embodiment 3

[0064] Combine below Figure 6-8 The scheme in embodiment 1 and 2 is further introduced, see the following description for details:

[0065] Step 1: Training data preprocessing

[0066] First, the training data is converted to a size of 101*101, and the contrast is enhanced. After the color image is grayscaled, the grayscale pixels are converted to between 0 and 1 to reduce the scale of the input features. Then increase the training sample size through horizontal, vertical, and 90-degree rotation enhancement processing.

[0067] Step 2: Model Construction

[0068] Construct the basic Unet network model, in which the feature extraction stage and the deconvolution stage are divided into 4 stages of feature layers according to pooling and the last sample. The feature layer of each stage contains 3 convolution structures, and the latter two The convolutional structure is replaced by two residual modules, where a normalization layer is added to the output of the second residual ...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses a feasible road segmentation method based on an improved Unet network model, and the method comprises the steps: modifying a convolutional layer in a Unet training model into aResNet residual module, and carrying out the back propagation training through employing a cross entropy loss function and a Lovsz hinge loss function; Performing feasible region segmentation on theroad scene without the segmentation label by adopting the trained model; Carrying out scale, contrast and normalization processing on the training picture and the corresponding segmentation mask mark,carrying out enhancement processing on the training sample, carrying out histogram matching processing on the test data by taking an average histogram of the training data as a template, carrying outtesting, and outputting a road segmentation result. According to the method, on the basis of not increasing the model complexity, the road segmentation precision is greatly improved, and real-time accurate segmentation of the feasible region of the road can be basically realized.

Description

technical field [0001] The present invention relates to the field of unmanned driving, in particular to a network model based on improved Unet [1] , combined with ResNet [2] The residual module of , a feasible road segmentation method for improving the Unet network model. Background technique [0002] The development of global driverless technology is ushering in a new round of upsurge. Major developed countries have taken various measures to support enterprises in developing and testing driverless cars. The research on unmanned driving technology has great practical significance, the most prominent of which is that it can improve the safety of people's travel. Road feasible area identification is an important part of unmanned driving research. Accurate and fast identification and segmentation of road feasible areas play a vital role in autonomous vehicle cruise control and path planning in unmanned driving. [0003] At present, the detection technology of road feasible a...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
IPC IPC(8): G06K9/00G06N3/04G06N3/08G06T7/155
CPCY02T10/40
Inventor 褚晶辉汤文豪王鹏李敏吕卫
Owner TIANJIN UNIV
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap