PRESENTED BY:

ARZ KUMAR SATSANGI & VINEET VERMA

Since the beginning of the development of machine vision, researchers have realized its importance in the robotics field, as it provides a useful tool for both the environment detection and decision making during the automation process. Applications of robot path planning are indoor/outdoor guidance or industrial purposes where human intervention is not possible

 The

robot should determine autonomously a path which it thinks leads to the goal.
path to be transverse by the robot must be ensured to be free of obstacles. should be capable of adapting itself to the change in environment.

The

Robot

CASE 1: Here we present a case where a camera

continuously acquires information and depending upon the lanes interpreted area it positions itself Algorithm For Lane Following Capture image Turn Image to binary image Optimize image and remove noise Divide the image in two halves: left region and right region Check the ratio of lane strip area on the left side to the right side If the ratio of the left side of the image to the right side of the image is > 1.25 - Then Turn left Else if the ratio of the left side of the image to the right side of the image is< 0.8 - Then Turn right

Insight into image processing steps involved The following steps are needed to convert a RGB image to binary image:
Convert Find

the RGB image into gray scale image

the threshold value. If the value at the pixel position is greater than the threshold value then the value will be 1(white) else zero (black).

CASE 1 (Contd...) Algorithm for Traffic Light Detection Capture image Process pixels diagonally (5 to 6 lines along the diagonal) Find the R G B values of pixel If (R-value>210) && (G-value<175) && (B-value<175) Count the pixel as red. If multiple red pixels are found Then Convert the image into binary and check whether it is circular by calculating the correlation coefficient between the current image and the capture binary images of other traffic red lights

CASE 2: Now we present a situation where the information is NOT acquired adaptively by the robot

the system is designed to (1) Acquire an overhead image using the network camera that will provide a wider perspective. (2) Transmit the image to the server, retrieve its contours after image processing by the server, and (3) send the coordinates of the contours to the robot, point by point and contour by contour.

(4)The robot will follow the points to sketch the contours of the image on the road

Insight into various image processing techniques involved

1. Colored to gray scale conversion The captured image is first converted from colored to Gray scale format. This step helps to simplify and speed up the subsequent image processing procedures.

2. Brightness and contrast adjustment The brightness and contrast of the image are adjusted properly to improve the results of the upcoming stages of the image processing. 3. Gaussian filtering One of the major problems that affect the process of extracting the contours from the image is the existence of the Gaussian noise. The zero-mean Gaussian filter has, therefore, been used in the process to remove the noise.

Noise Reduction
STEP 1

4. Initial Thresholding

Thresholding is the process of segmenting the image into two or more segments depending on the intensity values of its pixels. Thresholding is used to remove the undesired noise in order to prepare the image for the next steps.

STEP 2

5. Edge detection The Sobel edge detection operation is adopted to detect contours (or edges) in the image. The process of detecting the edges is performed twice:
the

first time is for the thresholded image to detect the outer contours of the image, and the second time is for the smoothened image to find the inner contours of the image. Adding both images together guarantees the detection of those most important contours in the snapshot.

FOR THRESHOLDED IMAGE

FOR SMOOTHENED IMAGE

6. Median filtering One of the side effects of using the previous image processing procedures is what is called salt and pepper noise. This noise is represented by having regions in the image which consist of one or two pixels with black or white values. These regions normally do not represent a real edge and need to be eliminated. The Median filter is, therefore, adopted to remove the noise.

7. Thinning contours The main purpose of the thinning process is to reduce the redundant information about the detected contours of the image while preserving the essential information of the contours at the same time.

8. Identifying contour intersections Several contours of the image may intersect with each other, forming a complex curve segment. This affects the labeling and tracing of the contours (also called curves). Consequently, a curve labeling algorithm may detect the intersected contours as one single curve with more than two end points. In order to trace the contours in the image correctly, the intersecting pixels (points) shared by more than one contour are distinguished first. These points separate the contours into shorter ones before all the contours can be labeled. Contours labeling: The contours in the image are labeled at this stage in order to trace all contours one by one, independently and completely.

Contours tracing: For effective sketching (or path following in the points in each contour must be organized in the correct order. This order guarantees that the contour tracing starts from the first point and ends with the last point. The process for tracing the contours begins by : Scanning the pixels in the image to find the one having a contour label. Checking its neighboring pixels of the current pixel, (if only one of the neighbors belongs to the same contour, this pixel is the last or the first pixel of the contour) The next step is to add this pixel to a new list, and then to find the next neighboring pixel that belongs to the contour. After the new pixels are identified and added to the list, they are converted to background pixels. Repeat the procedure until all the pixels in the contour are found and all the contours in the image are scanned.

9. Merging broken contours As mentioned earlier, resetting the common pixels of those intersected contours can divide them into shorter ones. It is good for unique contour labeling but may not be efficient for contour tracing and following It starts iteratively by taking one endpoint of one contour, and scans the endpoints of other contours one by one, by comparing the distance between the two endpoints and the difference in directions of the two endpoints. If the result of this comparison is within a specified threshold, the two contours (the two lists of points) will be merged in the correct order of points.

A more advanced methodology

Now we present a method to navigate a mobile robot using a webcam. This method determines the shortest path for the robot to transverse to its target location, while avoiding obstacles along the way. The environment is first captured as an image using a webcam.Sensors (laser range finders are then utilized to identify the existence of obstacles within the environment.

A 3D image is now rendered which is input to a algorithm that generates voronoi diagrams which in turn helps to find a obstacle free path right in the beginning

Image processing steps

Step 1 Capture still image from webcam The captured image is stored as an array whose elements represent the light intensity Step 2 Remove the input device from memory. Step 3 Convert from RGB to grayscale mode. Step 4 Find edges of objects in the image. Step 5. Remove noise and fill image

 Now

the image is rendered as a 3D image combining laser range finder and webcams assistance Now using Voronoi diagrams and cell decomposition technique we can guide the robot to reach the goal

A Voronoi diagram

Few minor but important aspects GPS tracking is also a method chosen to localize the robot
In

order to send the correct control commands to the robot such that the chosen path is followed, the precise position of objects must be measured. Feedback is the comparison of the target and actual positions, and is a natural step in implementing a motion control system.

 www.wikipedia.com
Webcam-based

Mobile Robot Path Planning using Voronoi Diagrams and Image Processing Shahed Shojaeipour1,1, Sallehuddin Mohamed Haris1, Elham Gholami2 and Ali Shojaeipour2 Integrated image processing and path planning for robotic sketching A. Mohammeda, L. Wanga,*, R.X. Gaob