7
Abstract Vision-based robot navigation systems allow a robot to explore and to navigate in its environment in a way that facilitates path planning and goal-oriented tasks. The vision sensor is mainly used for obstacle detection and avoidance, object detection and tracking, and interaction with users. Usually these systems do not depend solely on vision sensors but use other sensors like sonar and laser range finder. The paper considers an important issue for mobile robots navigation. This issue is the detection of obstacles in front of the robot within a corridor. We proposed and evaluated three algorithms for obstacle detection within a corridor environment using image processing techniques. I. Introduction Mobile Robots are now expected to be the next generation for assisting elderly and disabled persons as they present huge contributions in many industrial areas such as production lines at factories, surveillance systems, AGVs (autonomous guided vehicles), and indoor cleaning machines. One of the most researched areas in assistive technology is the development of intelligent wheelchairs. Intelligent wheelchairs are constructed by integrating artificial intelligence methods into a robotic wheelchair which has the ability to safely transport users to their desired destination. This work was originally inspired by the need for robotic assistance to contribute in developing an intelligent system to assist users with disabilities in core day-to-day activities. The final target of this research work is to develop a system that could be used to control the navigation process of a mobile robotic platform through an indoor corridor while detecting obstacles, avoiding them, avoiding walls, and planning a shortest path for the robot. A complete implementation and evaluation are performed for three obstacle detection algorithms. The first two algorithms work by detecting corridor region in front of the robot using the two boundary lines that normally exist between the walls and the ground. By detecting these two lines we could easily determine the ground region and we could later detect obstacles within this region only and neglect anything on the walls. There are many techniques proposed for detecting straight lines on an image for many purposes (power line detection, lane detection [1], etc.) such as Least Squares line fitting Method, Hough Transform, and Radon Transform. Line fitting can be a time consuming and computationally inefficient process, especially if there are many of such edge points. Therefore, The Hough Transform and the Radon Transform are. implemented and evaluated in this research for obstacle detection The third algorithm uses color segmentation approach to detect the obstacles in the corridor. Each of these three algorithms generates a grid-based map that could be merged with a path planning Algorithm to produce a safe path for the robot to follow. The rest of this paper is organized as follows: section II explains the used vision techniques, section III presents the environment assumptions for our algorithms, section IV explains the obstacle detection algorithms, and finally section V provide the conclusion. II. Vision Techniques There are many vision techniques have been used for vision based robot navigation system through the past two decades. In this research, many image processing techniques are used in the proposed algorithms such as the edge detection, Hough Transform, Radon Transform, and Color Segmentation. A. Edge Detection Edge detection is considered a primary method used in many computer vision systems that greatly reduces the amount of data for the vision system to preserve only the more important structural properties within an image. Edges in an image consist of either change in luminosity or change in color. One important consideration when using edge detection is its susceptibility to image noise. Noisy pixels create hard edges which introduce false information into the system. This problem could be solved using noise reduction filters and smoothing filters before the edge detection technique is performed to reduce the image noise, thereby reducing any false edge information. There is a large number of edge finding algorithms in existence [10]. B. Hough Transform The edge detection techniques tell us where edges are, but not what their shape are (geometric descriptions like line, arcs, etc.). The Hough Transform [2][3][4] is used to detect straight lines after edge detection operation. The main idea of Hough transform in line detection is to take into consideration the characteristics of the straight line not as image points x or y, but in terms of its parameters. Two representations for line parameters could be used with Hough transform. The first one is the slope-intercept form, the slope parameter (m) and the intercept parameter (c) as in equation (1): The 8th International Conference on INFOrmatics and Systems (INFOS2012) – 14-16 May Computational Intelligence and Multimedia Computing Track Faculty of Computers and Information - Cairo University MM-96 Basem M. ElHalawany, Hala M. Abdel-Kader, Adly TagEldeen, Alaa Eldeen Elsayed Vision-Based Obstacles Detection for a Mobile Robot Shoubra Faculty of Engineering Benha University Cairo, Egypt Zaki B. Nossair Helwan Faculty of Engineering Helwan University Helwan, Egypt

Vision-Based Obstacles Detection for a Mobile Robot Shoubra... · 2015. 5. 31. · Vision-based robot navigation systems allow a robot to explore and to navigate in its environment

  • Upload
    others

  • View
    0

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Vision-Based Obstacles Detection for a Mobile Robot Shoubra... · 2015. 5. 31. · Vision-based robot navigation systems allow a robot to explore and to navigate in its environment

Abstract

Vision-based robot navigation systems allow a robot to

explore and to navigate in its environment in a way that

facilitates path planning and goal-oriented tasks. The vision

sensor is mainly used for obstacle detection and avoidance,

object detection and tracking, and interaction with users.

Usually these systems do not depend solely on vision sensors

but use other sensors like sonar and laser range finder. The

paper considers an important issue for mobile robots

navigation. This issue is the detection of obstacles in front of

the robot within a corridor. We proposed and evaluated three

algorithms for obstacle detection within a corridor

environment using image processing techniques.

I. Introduction

Mobile Robots are now expected to be the next generation

for assisting elderly and disabled persons as they present

huge contributions in many industrial areas such as

production lines at factories, surveillance systems, AGVs

(autonomous guided vehicles), and indoor cleaning machines.

One of the most researched areas in assistive technology is

the development of intelligent wheelchairs. Intelligent

wheelchairs are constructed by integrating artificial

intelligence methods into a robotic wheelchair which has the

ability to safely transport users to their desired destination.

This work was originally inspired by the need for robotic

assistance to contribute in developing an intelligent system to

assist users with disabilities in core day-to-day activities.

The final target of this research work is to develop a

system that could be used to control the navigation process of

a mobile robotic platform through an indoor corridor while

detecting obstacles, avoiding them, avoiding walls, and

planning a shortest path for the robot.

A complete implementation and evaluation are performed

for three obstacle detection algorithms. The first two

algorithms work by detecting corridor region in front of the

robot using the two boundary lines that normally exist

between the walls and the ground. By detecting these two

lines we could easily determine the ground region and we

could later detect obstacles within this region only and

neglect anything on the walls.

There are many techniques proposed for detecting straight

lines on an image for many purposes (power line detection,

lane detection [1], etc.) such as Least Squares line fitting

Method, Hough Transform, and Radon Transform. Line

fitting can be a time consuming and computationally

inefficient process, especially if there are many of such edge

points. Therefore, The Hough Transform and the Radon

Transform are.

implemented and evaluated in this research for obstacle

detection

The third algorithm uses color segmentation approach to

detect the obstacles in the corridor. Each of these three

algorithms generates a grid-based map that could be merged

with a path planning Algorithm to produce a safe path for the

robot to follow.

The rest of this paper is organized as follows: section II

explains the used vision techniques, section III presents the

environment assumptions for our algorithms, section IV

explains the obstacle detection algorithms, and finally section

V provide the conclusion.

II. Vision Techniques

There are many vision techniques have been used for

vision based robot navigation system through the past two

decades. In this research, many image processing techniques

are used in the proposed algorithms such as the edge

detection, Hough Transform, Radon Transform, and Color

Segmentation.

A. Edge Detection

Edge detection is considered a primary method used in

many computer vision systems that greatly reduces the

amount of data for the vision system to preserve only the

more important structural properties within an image. Edges

in an image consist of either change in luminosity or change

in color. One important consideration when using edge

detection is its susceptibility to image noise. Noisy pixels

create hard edges which introduce false information into the

system. This problem could be solved using noise reduction

filters and smoothing filters before the edge detection

technique is performed to reduce the image noise, thereby

reducing any false edge information. There is a large number

of edge finding algorithms in existence [10].

B. Hough Transform

The edge detection techniques tell us where edges are, but

not what their shape are (geometric descriptions like line, arcs,

etc.). The Hough Transform [2][3][4] is used to detect

straight lines after edge detection operation. The main idea of

Hough transform in line detection is to take into

consideration the characteristics of the straight line not as

image points x or y, but in terms of its parameters. Two

representations for line parameters could be used with Hough

transform. The first one is the slope-intercept form, the slope

parameter (m) and the intercept parameter (c) as in equation

(1):

The 8th International Conference on INFOrmatics and Systems (INFOS2012) – 14-16 May Computational Intelligence and Multimedia Computing Track

Faculty of Computers and Information - Cairo University MM-96

Basem M. ElHalawany, Hala M. Abdel-Kader, Adly TagEldeen, Alaa Eldeen Elsayed

Vision-Based Obstacles Detection for a Mobile Robot

Shoubra Faculty of Engineering Benha University

Cairo, Egypt

Zaki B. Nossair Helwan Faculty of Engineering

Helwan University Helwan, Egypt

Page 2: Vision-Based Obstacles Detection for a Mobile Robot Shoubra... · 2015. 5. 31. · Vision-based robot navigation systems allow a robot to explore and to navigate in its environment

y mx c= + (1)

The second form that is preferred with the Hough

transform is the one that represent the line as in equation (2):

x cos y sin θ θ ρ+ =

(2)

The form in equation (2) is preferred to overcome the

drawback of the first form when vertical lines exist in the

image as the slope parameter (m) will be infinite.

By using equation (2), all lines in the image can be

transformed into the Hough space which gives a sinusoidal

curve that is unique for each pixel. The intersection point

between two curves in Hough space represents a line segment

in the image space that passes through both pixels. This can

be repeated for all the pixels on the edges in an edge detected

image as shown in fig.1.

Fig.1: Hough Space.

One important advantage of Hough Transform is its

resistance to noise in the image and its tolerance towards

holes in the detected line. On the other hand, one major

disadvantage of Hough Transform technique for detecting

straight lines is the large memory requirement and slowness.

C. Radon Transform

The Radon Transformation is a fundamental tool which is

used in various applications such as radar signals, medical

imaging, character Recognition, and line detection [5][6].

The 2D Radon transformation is the projection of the image

intensity along a radial line oriented at a specific angle as

shown in fig.2

Fig.2: Radon Transformation.

Let’s suppose we have a 2-D function f(x, y) as shown in

fig.2, By integrating along the line, whose normal vector is in

theta(θ) direction, we get a g(s, θ) function in equation (3)

which is the projection of the 2D function f(x,y) on the axis

s of θ direction and δ is the Dirac-delta function.

( ) ( ) ( )g s, f x, y . x cos y sin s dx dyθ δ θ θ= + −

(3)

As in Hough transform, the Radon operator maps from the

spatial domain (x,y) to the projection domain (s, θ), in which

each point corresponds to a straight line in the spatial domain.

Using this property, we used Radon transform for detecting

straight lines describing the road boundaries through the

corridor.

D. Color Segmentation

One of the most widely used methods for obstacle detection

is the segmentation of the area in front of the robot using

vision techniques. Segmentation in image analysis is the

division of an image into homogeneous areas of interest

based on some chosen image features [7][11]. A number of

different features can be used for segmentation, like color,

and texture. There are many segmentation approaches that

have been implemented within the past decades such as edge

detection segmentation, Histogram Threshold, Region-Based

segmentation, Fuzzy and Probabilistic techniques, Clustering

techniques, and Color segmentation techniques.

Color is a highly relevant feature for segmentation between

movable and non- movable areas in the corridor region,

which we are dealing with, in our scenario. There are two

steps required for color region segmentation, the first is

thresholding all pixels which share the same color and then

grouping those pixels together, throwing out any that don’t

seem to be in same area as majority of the pixels (region

growing). However we need a representation of color that

highlights the differences between the regions of interest.

One of the primary components to think about to get the

pixels that share the same color is to select the appropriate

color space (also called color model or color system) [8][9].

Color spaces are three-dimensional coordinate systems that

define colors. The most commonly used models are RGB,

HSV, YCbCr, L*a*b*, and CIE L*u*v*. There is no

agreement on which is the best.

III. Environment Assumptions

Some assumptions about the environment were taken.

Straight line indoor environment (Offices, Universities

corridors, etc.). All corridors marked with two Boundary

lines between the walls and the ground (common design) or

have different colors for wall and ground, see fig.3. and All

processed images are captured vertically by camera.

Fig.3: Sample corridor images

IV. Obstacle Detection Algorithms

Novel obstacle detection algorithms are still open research

area till now [12].Three different algorithms are proposed for

detecting obstacles within the corridor area inside buildings

using vision as the only sensor for the robotic platform. Each

algorithm uses a collection of image processing and

mathematical techniques. The first algorithm which called

obstacle detection using Hough Transform uses edge

detection, Hough Transform, and morphological operation.

The second algorithm which called obstacle detection using

Radon Transform is very similar to the first one but the

Radon Transform is used instead of Hough Transform. The

third one which called obstacle detection using color

segmentation uses color segmentation and morphological

The 8th International Conference on INFOrmatics and Systems (INFOS2012) – 14-16 May Computational Intelligence and Multimedia Computing Track

Faculty of Computers and Information - Cairo University MM-97

Page 3: Vision-Based Obstacles Detection for a Mobile Robot Shoubra... · 2015. 5. 31. · Vision-based robot navigation systems allow a robot to explore and to navigate in its environment

operations. In the next sub-sections, these algorithms are

presented with the results obtained for each algorithm.

The first and second algorithms need to perform edge

detection on the gray image of the area in front of the robot.

Therefore, we implemented and tested many methods of edge

detection using Matlab to get suitable results for next stages

of our algorithms. Sobel, Prewitt, Roberts, and Canny edge

detectors are tested. Canny edge detector offered the best

results. This was not a surprise as canny edge detection gives

thin lines for its edges by using non-maximal suppression.

Canny also utilizes hysteresis with thresholding to link strong

and weak edge pixels. Therefore, a decision is taken to make

it as the edge detector operator through this research.

Figure.4 represents the results of using canny’s edge

detector on image samples of the real corridor where our

algorithm is tested. The values of three parameters of canny

algorithm (Sigma, Max Threshold “Thmax”, and Min

Threshold level “Thmin”) are changed to get the suitable

parameters to be used.

After examining the results of using different values of the

three parameters, the values shown in table 1 were found

more suitable as they extracted the important details needed

to detect the two corridor lines and ignored unimportant

details in the captured samples of the corridor.

Table1: selected canny’s parameters

Sigma Thmax Thmin

3 0.4 0.25

(a)Sigma=3,

Thmax=0.8,Thmin=0.25 (b) Sigma=1,

Thmax=0.8,Thmin=0.25

(c) Sigma=2, Thmax=0.7, Thmin=0.1

(d) Sigma=3, Thmax=0.7, Thmin=0.1

(e)Sigma=1,

Thmax=0.4,Thmin=0.25 (f)Sigma=3,

Thmax=0.4,Thmin=0.25

Fig.4: Results of Canny Edge Detector

A. Obstacle Detection using Hough Transform

This algorithm uses many image processing functions to

detect the obstacles within the corridor area as seen in the

flowchart given in fig.5.

Fig.5: Flowchart for obstacle detection using Hough

transform.

Through this section, an explanation is introduced for

the each step of the flowchart in fig.5 which is supported by

samples of the obtained results.

a. Reading the color image from samples folder or

getting real-time images from the camera attached to

the robotic platform.

b. Convert the color images into grayscale. Fig.6 shows

both the original colored and the converted one into

grayscale.

Fig.6: Original colored and Grayscale image

c. Get the edge points of the grayscale image. The

results shown in fig.7 are done using the canny’s

edge with the selected parameters in table 1.

Fig.7: Edge points for the image

d. Reduce the number of edge points by eliminating the

upper edge points between the wall and wall-floor

boundary. This reduces the processing time for next

Hough calculation stage as shown in table 2. The

remaining pixels are shown in fig.8

e. Calculating Hough Transform Parameters for the

new edge points to get a voting matrix for the

candidate straight lines.

Read Colored-Image

Colored-to-Grayscale Conversion

Perform Edge Detection

Detect All Lines Using Hough-Transform

Select specific lines of the corridor

Get Area between the two lines

Threshold the area (to get obstacles)

Perform Morphological Operations

Build a grid-based Map

The 8th International Conference on INFOrmatics and Systems (INFOS2012) – 14-16 May Computational Intelligence and Multimedia Computing Track

Faculty of Computers and Information - Cairo University MM-98

Page 4: Vision-Based Obstacles Detection for a Mobile Robot Shoubra... · 2015. 5. 31. · Vision-based robot navigation systems allow a robot to explore and to navigate in its environment

Table 2: Time reduction by removing upper pixels Sample

No

Time without

using reduction

Time using

reduction

% reduction

in time

1 0.1452 0.1130 22.17%

2 0.1115 0.1030 7.6%

3 0.1319 0.1175 10.917%

4 0.1256 0.1136 9.5541%

5 0.1517 0.1358 10.4812%

6 0.1510 0.1329 11.9867%

7 0.1338 0.1332 0.4484%

8 0.1394 0.1092 21.664%

9 0.1373 0.1067 22.2869%

Fig.8: Reduced edge points.

f. Calculate the highest values in the voting matrix (or

accumulator) which represent the candidate line

pieces and plot as shown in fig.9. Line pieces in

green, starting points of the piece in yellow, and

ending points in blue.

Fig.9: Candidate straight line pieces

g. Filter the candidate Hough lines based on the

orientation angle to select the two-corridor Lines.

Vertical and near vertical Lines are removed from

the list and the rest are divided into right and left

category.

h. Select only one line from each category (if one of

the two categories has no lines on it this means that

the robot is facing one of the two corridor walls or

heading with angle that shows only one line as in

fig.10.b.

i. Complete these two line pieces to get the

intersections with the image boundaries and

intersection between them (if any within the image

range).

(a) Two complete lines (b) One complete line

Fig.10: corridor lines

j. Build a mask for the corridor region based on the

data of the two lines and intersection points (i.e.

Region of Interest). This area shows the region in

the image that need to be processed to get the

obstacles and free region for movement and

navigation. Fig.11 shows many samples from

different images.

Fig.11: Mask of the corridor region

k. Perform a threshold for the selected area which

converts it to binary image. Then follow this by

morphological operations to detect obstacles within

this region

Fig.12: Binary image of the corridor region l. Label the connected components in the binary image

into different object regions and get some properties

of each object such as the “centroid” marked with

BLUE asterisk and the RED “Bounding Box”

around its area as shown in fig.13.

Fig.13: Corridor region with marked obstacle

m. Use the previous results shown in fig.13 to build a

grid-based map that could be used later by path

planning algorithms. This could be done by

quantizing the image resulted from earlier steps.

Fig.14 shows a sample of such map created by our

system that is followed a path planning performed

using A* algorithm.

Fig.14: Grid-based Map

Figure.15 represents more samples from the real corridor

at the left side and the detected obstacles at the right side.

Obstacle Detection using Radon Transform

In this algorithm, we use the same steps on the flowchart

of obstacle detection using Hough Transform but we replace

the process of detecting line segments using Hough transform

by the use of Radon Transform. The radon transform is used

to implement the function of Hough transform to detect line

segments. Then we analyze the resultant possible line

segments to get the right and left corridor lines that were used

The 8th International Conference on INFOrmatics and Systems (INFOS2012) – 14-16 May Computational Intelligence and Multimedia Computing Track

Faculty of Computers and Information - Cairo University MM-99

Page 5: Vision-Based Obstacles Detection for a Mobile Robot Shoubra... · 2015. 5. 31. · Vision-based robot navigation systems allow a robot to explore and to navigate in its environment

earlier as boundaries. Figure.16 shows samples from these

results. We use these two lines for corridor region detection

and we could apply the morphological operation to detect the

obstacles within it.

(a) Single obstacle

(b) Single obstacle

(c) Single obstacle

(d) Two obstacles

(e) Two obstacles Fig.15: More Results for Obstacle Detection using Hough Transform

This algorithm provides us with the same results obtained

by the method of obstacle detection using Hough transform.

The algorithm detects the obstacles correctly but it is slower

than first method because the integration process in radon

transform consumes more time as shown in table 3 after

performing both methods on the same image samples.

Obstacle Detection using Color Segmentation

This algorithm focuses on color image segmentation using a

color thresholding technique to differentiate between

movable free areas of the corridor ground and the obstacles

within it.

In these experiments, we investigated two color models

other than the RGB to determine the best one for detecting

the objects within the corridor.

RGB is the most widely used color model in computer

hardware and cameras. This model represents the color as a

three independent components red, green, and blue. RGB is

an additive model and combination of R, G and B values

generates a specific color. This model is hard to comprehend

for humans because the human way of understanding and

describing color is not based on combination of red, green,

and blue. In addition, The RGB is perceptually non-uniform.

This means that the difference of two RGB values does not

correspond to the perceptual difference [9].

Table 3: Processing time for many samples using Hough Transform and Radon Transform

Sample

No

Hough Time

(second)

Radon Time

(second)

1 0.1352 0.1950

2 0.1342 0.2099

3 0.1221 0.1798

4 0.1149 0.1643

Fig.16: Radon Transform Line Detection Results

HSV color model is suitable for human description of

colors as we use three parameters to describe the color by its

Hue (H), Saturation (s), and Value (V).The Hue is attribute

that describe the color purity or tint (pure red, yellow, or

green) whereas saturation is a measure of the degree to which

a pure color is diluted by white light, (e.g. red is saturated and

pink is less saturated). The value defines how bright the color

is or the quantity of light received by the sensor. The HSV

color model overcomes the limitations of the RGB color

space because it describes human vision quite accurately and

because it mainly uses only the hue to describe color. The

other two components (i.e., saturation and value) are

significant only when describing black, white, gray, and the

various shades of the colors. HSV model has some

disadvantages such as the complexity in calculations and the

Hue is undefined for gray. Equation 4 describes how to get

the Hue (H) value from RGB.

3 (G - B)H = tan[ ]

(R - G) + (R - B) (4)

Color Segmentation in HSV Space

A sample from the area in front of the robot is captured in

HSV space. We use the Hue, Saturation and a combination of

Hue and Saturation as a threshold value to get the best results.

Figure.17 shows samples of the results which prove that HSV

is not valid when dealing with corridor having a gray color.

When using both Hue and Saturation together for

thresholding the corridor area, better results are obtained but

the obstacles could not be extracted alone for gray grounds.

Some false obstacles are detected also due to HSV failure to

represent gray floor. Therefore, another color space is needed

The 8th International Conference on INFOrmatics and Systems (INFOS2012) – 14-16 May Computational Intelligence and Multimedia Computing Track

Faculty of Computers and Information - Cairo University MM-100

Page 6: Vision-Based Obstacles Detection for a Mobile Robot Shoubra... · 2015. 5. 31. · Vision-based robot navigation systems allow a robot to explore and to navigate in its environment

overcome the weak point of HSV model for representing the

gray objects.

(a) Original RGB (b) Segmentation using “H”

(c) using “S” (d) Segmentation using “HS”

Fig.17: Color thresholding in HSV

Color Segmentation in YCbCr Space

The YCbCr color space is widely used for digital video.

In this model, luminance information is stored as a single

component (Y), and chrominance information is stored as

two color-difference components (Cb and Cr). Cb represents

the difference between the blue component and a luminance

value. Cr represents the difference between the red

component and a luminance value. The conversion from

RGB to YCbCr illustrated in equation 5.

0.257 0.504 0.098 R 16

-0.148 - 0.291 0.439 G 128

0.439 - 0.368 - 0.071 B 128

Y

Cb

Cr

= +

(5)

Our Obstacle Detection using color segmentation

algorithm benefits from the efficient YCbCr model. This

algorithm performs a color-space conversion to YCbCr

system followed by a color thresholding on the converted

image by comparing each pixel to a sample taken from the

corridor area in front of the robot. Comparison is performed

using the two color channels of the YCbCr system (Cb & Cr).

As shown in fig.18, the algorithm could easily perform color

thresholding for the gray corridor region.

This algorithm could easily detect doors within the

corridor by the same algorithm as shown in fig.18.e. This will

be easier if the building environment be prepared for robot

navigation by using distinct colors for corridors, walls, and

doors. We advice also that each door be marked with distinct

colored mark to distinguish between different rooms.

V. Conclusion

Obstacle detection is a very important issue in the field of

robot navigation and path planning. The process of detecting

the obstacles in the way of the mobile robotic platform could

be performed using many sensors such as laser rang finders,

Infrared sensors, and vision sensors. By improving the results

obtained by each type of these sensors, we could make a

fusion of these results to get a robust obstacle detection

process that facilitate the goal-oriented tasks of the robot.

Due to the nature of the problem, it is hard to objectively

evaluate the performance of the proposed methods. To say

where the clear movable area starts and ends is a matter of

definition and will probably vary depending on the observer.

It is hence more or less impossible to present an absolute

quantitative measure of performance. We believe the best

way to test the system’s performance is to run it on a group of

real images from the operating scenario and study the output

subjectively.

Through the experimental work, the three algorithm were

tested on many images captured from the real corridors we

testing the algorithm in. It is found that they detect all

obstacles in front of the robot. But in cases, at which there is

exist an obstacle very close to the walls, this obstacle is not

detected as a standalone object but it appears as a part of the

unmovable area of the walls. This will not be a problem in

our scenario as it already marked as an unmovable area but

when we need to perform a target tracking, a modification

should be done.

(a) YCbCr image

(b) Binary image after color thresholding

(c) Removing small objects& Filling Holes

(d) free-corridor area & obstacles within it

(e) Color thresholding for the target door

Fig.18: Color thresholding in YCbCr

References

[1] Ji-Hun Bae, "Monocular vision-based lane detection using segmented regions from edge information ", Ubiquitous Robots and Ambient Intelligence (URAI), 2011 8th International Conference.

[2] Wojciech Wojcikiewicz, “Hough Transform Line Detection in Robot Soccer”, Heriot Watt University, March 2008.

[3] Muhammad Azwan and Mohd Rizal Arshad, "A Feature-based Lane Detection System using Hough Transform Method", 8th Seminar on Intelligent Technology and Its Applications, Kampus ITS Sukolilo Surabaya, Indonesia, May 2007

[4] Takeshi Saitoh, Naoya Tada and Ryosuke Konishi, “Indoor Mobile Robot Navigation by Center Following based on Monocular Vision”, Tottori University, Japan,2008

[5] P. Toft, “The Radon Transform-Theory and Implementation”, Ph.D. thesis, Dept. Math. Model, Tech. Univ. Denmark, Lyngby, 1996.

[6] A.S.M Shihavuddin, Kabir Ahmed, Md. Shirajum Munir and Khandakar Rashed Ahmed , “Road Boundary Detection by a Remote Vehicle Using Radon Transform for Path Map Generation of an Unknown Area”, IJCSNS International Journal of Computer Science and Network Security, VOL.8 No.8, August 2008.

[7] Urban Bergquist, “Colour vision and Hue for Autonomous Vehicle Guidance”, M.Sc thesis, Linköping University, 2000.

[8] H.J.C. Luijten, “Basics of color based computer vision implemented in Matlab”, Technische Universiteit Eindhoven, June 2005.

[9] Mohamed E.Helala, “Quantitative Image Search Based On Feature Integration”, M.Sc thesis, Benha University, Egypt, October 2010.

[10] Raman Maini, Dr. Himanshu Aggarwal, “Study and Comparison of Various Image Edge Detection Techniques”,

The 8th International Conference on INFOrmatics and Systems (INFOS2012) – 14-16 May Computational Intelligence and Multimedia Computing Track

Faculty of Computers and Information - Cairo University MM-101

Page 7: Vision-Based Obstacles Detection for a Mobile Robot Shoubra... · 2015. 5. 31. · Vision-based robot navigation systems allow a robot to explore and to navigate in its environment

International Journal of Image Processing (IJIP), Volume (3) : Issue (1),March 2009.

[11] Bakar, M.N.A, "Development of a doctor following mobile robot with Mono-vision based marker detection ", Applied Power Electronics Colloquium (IAPEC), 2011 IEEE.

[12] K. Kungcharoen, "Development of object detection software for a mobile robot using an AForce.Net framework", 9th International Conference, ICT and Knowledge Engineering (ICT & Knowledge Engineering), Jan. 2012.

The 8th International Conference on INFOrmatics and Systems (INFOS2012) – 14-16 May Computational Intelligence and Multimedia Computing Track

Faculty of Computers and Information - Cairo University MM-102