ICACT20220030 Slide.18
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
 |
We performed an environmental adaptive line detection by dynamically adjusting the HSV value.
To this end, an illumination/value matrix was created and applied to line detection.
However, matrix generation is a non-automatic process that is difficult to do.
Therefore, we plan to automatically perform this, such as Reinforcement learning.
|
|
ICACT20220030 Slide.17
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
 |
Next is a night when using a headlight.
In this case, the illumination is 0 lux and the value is about 180.
When headlight was used, there was no increase in illusion, while the value increased significantly.
This means that the values of illumination and value are different in that they are affected by headlight and the sun.
|
|
ICACT20220030 Slide.16
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
 |
The following figure shows the night without a headlight.
In this case, the illumination is about 0 lux and the value is about 40.
|
|
ICACT20220030 Slide.15
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
 |
The following is the result of lane detection considering illumination.
First, the result of lane detection during the day when the headlight is not turned on is shown in the following figure.
In this case, illumination is about 40 lux and value is about 180.
|
|
ICACT20220030 Slide.14
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
 |
The following is the work to create a matrix.
The time zone is 17:00 to 19:00 on August 13, 2021, and it is the time when the sunsets.
|
|
ICACT20220030 Slide.13
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
 |
Next is the implementation.
The specifications of HW and SW used are shown in the following figure.
|
|
ICACT20220030 Slide.12
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
 |
The following figure is an algorithm for dynamically adjusting the range of HSV values.
First, the algorithm fetches data and headlight information on current illumination.
It then brings each matrix according to the on/off of the headlight.
Finally, it calculates the value through the illumination data, matrix.
|
|
ICACT20220030 Slide.11
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
 |
The following figure is an algorithm for line detection.
First, the system imports the image and then applies the Gaussian filter and ROI.
After that, it dynamically adjusts the value using the generated matrix. A detailed description of this will be given in the following algorithm.
After the HSV range is determined, it extracts and terminates the edge, hough line representative line.
|
|
ICACT20220030 Slide.10
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
 |
The following is lane detection considering illumination.
Lane detection includes the following process.
First, the Gaussian filter is used to remove the grip from the image.
ROI is used to remove unnecessary parts of the image.
Edge detection is used to extract edges corresponding to the boundaries of colors in the image. In this paper, we use a canny edge.
Hough transformation calculates straight lines using the extracted edge information.
The representative line extracts a significant straight line from among the extracted several straight line information.
|
|
ICACT20220030 Slide.09
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
 |
The following figure shows the flow chart of the Creation illumination/value matrix.
First, in order to receive an image and illuminance value, it sets up OpenCV and I2C.
And this brings in the empty matrix and then the information and timestamp of the headlight.
Thereafter, it receives a pixel corresponding to the lane of the image as input and extracts HSV from the image.
The photoresistor uses the I2C protocol to obtain illuminance values.
After importing the two data, it creates a matrix and ends.
|
|
ICACT20220030 Slide.08
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
 |
The following is the Design of Dynamic Adjusting HSV of this paper.
First, we create an illumination/value matrix to dynamically adjust the lane color range.
To this end, the value is first obtained from the photo resistor.
And we compare the value of the lane detected by the actual camera with the value of the illuminance.
Image processing uses OpenCV.
Photoresist uses the I2C protocol.
|
|
ICACT20220030 Slide.07
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
 |
One study implemented line detection using raspberry pi.
In this paper, lanes were detected using OpenCV and Pools and unpaved roads were considered.
In another study, lane detection was implemented through an actual vehicle.
But Their studies still did not address the adaptation of the administration to environmental changes.
|
|
ICACT20220030 Slide.06
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
 |
Next is Related Works.
In one study, line detection was performed using Python.
In this paper, lane detection used techniques such as canny edge, hough line, etc, and the accuracy is 90%.
However, since this was done in the same time zone, it did not take into account the changes in the nightly environment.
Therefore, it is essential to add methods to adaptively change parameters.
|
|
ICACT20220030 Slide.05
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
 |
Actual lane detection needs to consider changing environment.
For example, weather, time, etc.
Exist studies only consider a single environment.
Our goal is Lane detection concerning the Illumination scenario.
To this end, line detection is developed in Classic Computer Vision.
To do this, we Compare the photoresistor with the value of HSV.
And we create an illumination/value matrix.
We dynamically adjust a range of values using the created matrix utilize created matrix.
|
|
ICACT20220030 Slide.04
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
 |
HSV includes Hue, saturation, and value configuration as color channels.
It can be used where the robot is controlled or shadowed. because it is similar to human vision.
Lane detection has Illumination Issue.
Illumination scenario is the case of Shading by trees and building, Rapid illusion changes, and Tunnel entry.
Lane detection systems shall be adaptable to Illumination circumstances.
Otherwise, problems such as recognition errors, noise, etc occur.
|
|
ICACT20220030 Slide.03
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
 |
Lane detection techniques utilize the following techniques.
A detailed explanation is described later.
|
|
ICACT20220030 Slide.02
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
 |
First, it is an explanation of the background of the study.
Lane detection is a technique for recognition when autonomous-driving vehicles drive on roads.
Typical lane detection is a visual-based approach using cameras.
It can be largely divided into two ways.
The first is a method using classical computer vision.
This includes methods of Cleaning, Feature Detection, Model Application, Tracking Integration, and Coordinates Translation.
The second is the learning-based method.
This includes the method of Extracting features using constructed learners.
In the case of classical computer vision, there is an advantage of Cost-effectiveness.
However, it is Easily affected by Illumination changes.
|
|
ICACT20220030 Slide.01
[Big slide for presentation]
|
Chrome Text-to-Speach Click!! |
 |
Hello, I'm Yong Hee Park, who will be presenting a paper.
The title of the paper is Implementation of Dynamic HSV Adjusting Method for Lane Detection
I will start the presentation.
|
