Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

A Visual Terrain Classification Method Based on Multiple Coding and Feature Fusion

A technology of terrain classification and feature fusion, which is applied in the directions of instruments, computing, character and pattern recognition, etc., can solve the problems of unsatisfactory accuracy and stability of underlying features, reduce feature space correlation, improve classification effect, The effect of accuracy and speed improvement

Active Publication Date: 2019-05-31
SANITARY EQUIP INST ACAD OF MILITARY MEDICAL SCI PLA
View PDF1 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] Various color, texture and gradient features have been applied to visual terrain recognition, and have achieved certain results, but the accuracy and stability of these underlying features are still unsatisfactory.

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 Visual Terrain Classification Method Based on Multiple Coding and Feature Fusion
  • A Visual Terrain Classification Method Based on Multiple Coding and Feature Fusion
  • A Visual Terrain Classification Method Based on Multiple Coding and Feature Fusion

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0029] The embodiment of the present invention provides a visual terrain classification method based on multiple coding and feature fusion, see figure 1 , the method includes the following steps:

[0030] 101: Establish a visual terrain dataset as an evaluation test object, perform key point detection on the visual terrain image, describe the key points of local information in the image using SIFT features, and extract local underlying visual information;

[0031] 102: Use PCA-Whitening to perform feature preprocessing on SIFT features to reduce the correlation of feature spaces;

[0032] 103: Establish K-means (K-means clustering) dictionary and GMM (Gaussian mixture model clustering) dictionary respectively according to the result of feature preprocessing;

[0033] 104: Use sparse coding for the K-means dictionary to describe the 0-order local features of the visual terrain image; use differential coding for the GMM dictionary to describe the 1st-order and 2-order local fea...

Embodiment 2

[0039] The following is combined with specific calculation formulas, examples and figure 2 The scheme in embodiment 1 is described in detail, see below for details:

[0040] (1) Establishment of visual terrain dataset

[0041] Because there is no general-purpose data set in the field of terrain recognition, the embodiment of the present invention produced a data set DS1. As an evaluation test object, the data set DS1 contains a total of 8 different typical terrain road surfaces: asphalt road, mud, grass, ceramic tiles, gravel , large gravel, sand and leaf cover. The pictures are acquired under different lighting and weather conditions, with a uniform resolution of 256×256, each category contains at least 300 pictures, and the data set has a total of 2700 pictures. Various types of typical road surfaces such as figure 2 shown.

[0042] The embodiments of the present invention do not limit the resolution of pictures, the number of pictures, and the types of pictures, which ...

Embodiment 3

[0094] Below in conjunction with concrete experiment, the scheme in embodiment 1 and 2 is carried out feasibility verification, see the following description for details:

[0095] see image 3 , In 2012, Filichkin and Byl of the University of California-Santa Barbara (UCSB) Robotics Institute used SIFT features, combined with the word bag method, to generate compact image mid-level descriptors, and input them into support vector machines Classification and recognition in (SVM) has been greatly improved compared with previous methods, and has become the current standard method in the field of terrain recognition. Its research results have also been applied to the US military robot - LittleDog (LittleDog), which has become an important functional load. . In the embodiment of the present invention, this method is used as a baseline method, and a comparison of effects is given at the end.

[0096] The embodiment of the present invention extracts SIFT and GIST features, adopts PC...

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 visual terrain classification method based on multiple coding and feature fusion. Firstly, SIFT feature is extracted to describe the local information of the image, and PCA-Whitening is used to perform feature preprocessing to reduce the correlation of feature space, and then establish K-means dictionary and GMM dictionary, realize the re-description of visual terrain features through sparse coding and differential coding methods, and use GIST features to supplement the global information of visual terrain. Finally, use multi-core learning method for feature fusion to form a compact visual terrain feature expression, input linear support In the vector machine (SVM), the result of visual terrain classification is obtained. The invention comprehensively describes local multi-level information and global features of visual terrain by using multiple encoding and multi-feature fusion methods, and verifies the effectiveness of the method by comparing with the baseline method.

Description

technical field [0001] The invention relates to the field of visual terrain classification, in particular to a visual terrain classification method based on multiple coding and feature fusion. Background technique [0002] Different from the indoor structured environment, the robot must face different road environments in the wild. Soft, muddy, and uneven roads may bring danger to the robot. These dangerous road surfaces are collectively referred to as non-geometric hazards. In this regard, the robot must have accurate perception and classification capabilities in order to make appropriate path planning, different gait planning and motion control strategies. If the terrain feature information where the robot is located cannot be accurately identified, it may cause the robot to make wrong gait planning and dynamic control decisions, making the robot unable to achieve the desired movement, or even dangerous. [0003] The methods for identifying non-geometric terrain features...

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
Patent Type & Authority Patents(China)
IPC IPC(8): G06K9/62
CPCG06F18/23213G06F18/2411G06F18/2415G06F18/253
Inventor 吴航刘保真孙景工苏卫华张文昌苑英海安慰宁秦晓丽
Owner SANITARY EQUIP INST ACAD OF MILITARY MEDICAL SCI PLA
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com