08-02-2010, 10:27 AM
[attachment=1823]
Low Cost Smart Cam Design
1.Introduction
This paper proposes a very low cost vision system that is able to perform image processing tasks with certain processing algorithms.
Digital cameras include dedicated digital image processing chips to convert the raw data from the image sensor into a color-corrected image organized in a standard image file format.
The images flow will be also enriched with a collection of information that ranges from the position of recognized objects to their color, size or shape.
The main contributions of the proposed work are the following: - The study and development of a very low cost
architecture for digital image processing; - The implementation of a set of optimized digital image processing algorithms to be used;
- The development of a color vision system that has been successfully applied to real-world case studies.
2.Previous Work
This camera has been developed in three different versions: CMUcam1, CMUcam2 , and CMUcam3.
CMUcam1 is a low-cost, low-power sensor for mobile robots to perform many different kinds of real-time processing, with a maximum resolution of 80x143 tracking 17 frames per second.
The CMUcam2 is characterized by a maximum resolution of 176 x 255 and it is able to track user dened color blobs up to 50 frames per second and it is also able to track motion using frame differencing at 26 frames per second.
CMUcam3 provides a exible and easy to use open source development environment. It supports CIF with resolution of 352x288 and is able to process images at a rate of 26 frames per second.
3.Digital Image Processing Algorithms
The following algorithms have been implemented in order to perform the image processing:
Background subtraction and foreground segmentation .
Blob detection .
Object tracking .
1. Background subtraction and foreground segmentation
This algorithm allows to obtain an image in which the background has been completely removed.
This elaboration can be performed both in a static scenario and in a dynamic one.
2. Blob Identication and Detection
The algorithm is represented in 2 phases:
In the rst phase an elaborated image can be processed in order to¦
- assign to adjacent pixels belonging to the same objects the same identication number;
- Lists of identication numbers refer to blob.
The second phase performs a second scan of the image, in order to¦
- map all the identication numbers that refer to the same object to the same blob identication letter.
3. Object Tracking
The rst step of the proposed object tracking algorithm is the evaluation of the list of blobs from detection phase.
- If this list is empty ,then it is possible to generate the output.
- Otherwise a blob of list has to be selected and evaluated , with respect to its color, shape, position, etc...
If the selected blob has been detected to be the primary target then it is possible to update the currently selected primary target and to remove the blob from the blobs list.
Otherwise, if the target has been detected to be a secondary target, it is possible to update the secondary targets list and to remove the blob from the blobs list.
The rst step is the detection of adjacent pixels belonging to the same objects as stated:
the pixel on the top left corner of the image will receive either a 0, if it belongs to the background, or a 1, if it belongs to an object;
any other pixel of the rst top row will receive:
a 0 if belonging to the background;
the same value of the adjacent pixel on its left if both the pixel on its left if both the pixels belong to an object;
new value if it belongs to an object but the adjacent pixel on its left belongs to the background ;
any other pixel of the other rows will receive:
a 0 if belonging to the background;
the same value of the adjacent pixel on its left if both the pixels belong to an object;
the same value of the adjacent pixel on its top if both the pixels belong to an object, but the adjacent pixel on the left of the lower one belongs to the background;
a new value (never used before on the same image) if it belongs to an object but the adjacent pixel on its left belongs to the background.
Figure 4.2 shows the result.
5. Experimental Results
One of the real scenarios in which the proposed low cost smart cam has been exploited in order to perform a recognition task is the test-tubes detection task.
The following couple of experiments with test-tubes are presented:
Detection of test-tubes presence or absence.
Tracking of the desired test-tube.
6. Conclusions
Solution for real-world case studies, since it is able to correctly detect, recognize and track the desired object.
The proposed smart cam is comparable to the previous models from the technological and the functional point of view,
Effective low cost solution for the detection and the tracking of objects, based on the elaboration on their color, position, shape and size.
Low Cost Smart Cam Design
1.Introduction
This paper proposes a very low cost vision system that is able to perform image processing tasks with certain processing algorithms.
Digital cameras include dedicated digital image processing chips to convert the raw data from the image sensor into a color-corrected image organized in a standard image file format.
The images flow will be also enriched with a collection of information that ranges from the position of recognized objects to their color, size or shape.
The main contributions of the proposed work are the following: - The study and development of a very low cost
architecture for digital image processing; - The implementation of a set of optimized digital image processing algorithms to be used;
- The development of a color vision system that has been successfully applied to real-world case studies.
2.Previous Work
This camera has been developed in three different versions: CMUcam1, CMUcam2 , and CMUcam3.
CMUcam1 is a low-cost, low-power sensor for mobile robots to perform many different kinds of real-time processing, with a maximum resolution of 80x143 tracking 17 frames per second.
The CMUcam2 is characterized by a maximum resolution of 176 x 255 and it is able to track user dened color blobs up to 50 frames per second and it is also able to track motion using frame differencing at 26 frames per second.
CMUcam3 provides a exible and easy to use open source development environment. It supports CIF with resolution of 352x288 and is able to process images at a rate of 26 frames per second.
3.Digital Image Processing Algorithms
The following algorithms have been implemented in order to perform the image processing:
Background subtraction and foreground segmentation .
Blob detection .
Object tracking .
1. Background subtraction and foreground segmentation
This algorithm allows to obtain an image in which the background has been completely removed.
This elaboration can be performed both in a static scenario and in a dynamic one.
2. Blob Identication and Detection
The algorithm is represented in 2 phases:
In the rst phase an elaborated image can be processed in order to¦
- assign to adjacent pixels belonging to the same objects the same identication number;
- Lists of identication numbers refer to blob.
The second phase performs a second scan of the image, in order to¦
- map all the identication numbers that refer to the same object to the same blob identication letter.
3. Object Tracking
The rst step of the proposed object tracking algorithm is the evaluation of the list of blobs from detection phase.
- If this list is empty ,then it is possible to generate the output.
- Otherwise a blob of list has to be selected and evaluated , with respect to its color, shape, position, etc...
If the selected blob has been detected to be the primary target then it is possible to update the currently selected primary target and to remove the blob from the blobs list.
Otherwise, if the target has been detected to be a secondary target, it is possible to update the secondary targets list and to remove the blob from the blobs list.
The rst step is the detection of adjacent pixels belonging to the same objects as stated:
the pixel on the top left corner of the image will receive either a 0, if it belongs to the background, or a 1, if it belongs to an object;
any other pixel of the rst top row will receive:
a 0 if belonging to the background;
the same value of the adjacent pixel on its left if both the pixel on its left if both the pixels belong to an object;
new value if it belongs to an object but the adjacent pixel on its left belongs to the background ;
any other pixel of the other rows will receive:
a 0 if belonging to the background;
the same value of the adjacent pixel on its left if both the pixels belong to an object;
the same value of the adjacent pixel on its top if both the pixels belong to an object, but the adjacent pixel on the left of the lower one belongs to the background;
a new value (never used before on the same image) if it belongs to an object but the adjacent pixel on its left belongs to the background.
Figure 4.2 shows the result.
5. Experimental Results
One of the real scenarios in which the proposed low cost smart cam has been exploited in order to perform a recognition task is the test-tubes detection task.
The following couple of experiments with test-tubes are presented:
Detection of test-tubes presence or absence.
Tracking of the desired test-tube.
6. Conclusions
Solution for real-world case studies, since it is able to correctly detect, recognize and track the desired object.
The proposed smart cam is comparable to the previous models from the technological and the functional point of view,
Effective low cost solution for the detection and the tracking of objects, based on the elaboration on their color, position, shape and size.