A multi-scale fully convolutional network and method and device for visual guidance

A fully convolutional network and multi-scale technology, applied in the field of visual guidance, can solve the problems of complex hardware technology, high power consumption, and impossibility of popularization, and achieve high detection effect and real-time detection speed

Active Publication Date: 2020-06-19
成都快眼科技有限公司
View PDF4 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Infrared blind guide equipment is not easy to scatter and has a long transmission distance, but has the disadvantages of large power consumption, large power supply, and not easy to carry and use.
Although the guide robot can simulate the real guide dog to guide the blind safely, its hardware technology is complicated, the development cost is high, and it is bulky and inconvenient to carry and maintain, so it cannot be popularized
The portability of the guided cane is not high enough, and a large number of blind people think that the image of using the cane is not good, and they are unwilling to use the cane, and they cannot be promoted.
[0004] Although the above-mentioned blind-guiding equipment can meet certain blind-guiding needs of the blind, they cannot provide comprehensive and comprehensive blind-guiding tasks for the visually impaired.

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 multi-scale fully convolutional network and method and device for visual guidance
  • A multi-scale fully convolutional network and method and device for visual guidance
  • A multi-scale fully convolutional network and method and device for visual guidance

Examples

Experimental program
Comparison scheme
Effect test

specific Embodiment 1

[0048] A multi-scale fully convolutional network, including a classification network and a detection network, the classification network is used to extract the characteristics of the preset window; the detection network is used to score and return the preset window, using a multi-channel parallel The structure directly performs feature fusion on the 1*1 convolution layer, and splits the 5*5 convolution into two 3*3 convolution operations.

[0049] In this specific embodiment, because the parameters of the network model of inception are too large, it cannot meet the real-time requirements on the embedded platform. In order to reduce the parameters of the network model, on the basis of the network structure of GoogleNetInception, the 1*1 convolutional layer is directly fused with features, and the 5*5 convolution is split into two 3*3 convolution operations, Attached figure 2 , 3 shown. The invention reduces model parameters, deletes some redundant layers according to experi...

specific Embodiment 2

[0051]On the basis of specific embodiment 1, the specific method of directly performing feature fusion on the 1*1 convolutional layer is as follows: firstly, use a 1×1 convolution kernel to adjust the number of channels of the convolutional feature spectrum, and then use convolutional layers of different sizes The product kernel extracts convolution features of different scales, and finally fuses the features of different channels.

specific Embodiment 3

[0053] On the basis of the specific embodiment 1 or 2, the classification network cuts the input color picture size to 100*100 (unit: pixel) size, and then connects more than two convolution modules, each convolution module includes a volume Product operation, batch normalization operation and ReLU activation function with parameters; the classification network adopts filters of 3*3, 2*2 and 1*1 (unit: pixel) size, with a step size of 1, and set After the convolution module is set, the maximum pooling operation is added respectively. The size of the pooling area is 2*2 (unit: pixel), and the step size is 1; the image is classified by using the features of the set convolution module.

[0054] In this specific embodiment, the entire classification network such as figure 1 As shown, the present invention proposes a small network model, aiming at reducing the computational complexity of the model, and satisfying the real-time requirement of the algorithm while ensuring the accurac...

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 provides a multi-scale full convolutional network and a visual blind guiding method and device. The multi-scale full convolutional network comprises a classification network and a detection network. The classification network is used for extracting the characteristics of a preset window. The detection network is used for grading and regression of the preset window, is of a multi-channel parallel structure and can directly conduct feature fusion of a 1*1 convolution layer and divide a 5*5 convolution layer into two 3*3 convolution layers. The classification network and the detection network are trained to generate the multi-scale full convolutional network, and the generated multi-scale full convolutional network is used for detection of various road targets; and general obstacles nearby are detected through a depth map generation method, and the accurate distances between various detection targets and a device carrier as well as between the obstacles and the device carrier are worked out. Compared with the prior art, multi-scale full convolutional network can be realized on a mobile device with extremely high portability, the real-time detection speed can be achieved, and a good detection effect can stilled be kept under a complex scene.

Description

technical field [0001] The present invention relates to a visual guide technology, in particular to a multi-scale full convolution network and a visual guide method and device. Background technique [0002] There are a large number of visually disabled people in my country, and the situation is very serious. The government is investing huge financial resources in the construction of infrastructure for the blind, including medical technology and navigation system upgrades. With the help of current technology, the living conditions of blind people have been changed to a certain extent, and their quality of life has been improved. [0003] So far, there is no blind-guiding device specially made for the blind in the Chinese market, and the blind-guiding electronic products developed abroad are divided into two categories: autonomous and guided in terms of working methods. Among them, autonomous blind guide equipment is generally equipped with sensors, and blind people can hold ...

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): G06N3/08A61H3/06
CPCA61H3/061
Inventor 李宏亮
Owner 成都快眼科技有限公司
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