Acquisition Of Image Content Processing Computer Science Essay

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

Object detection makes the region of interest in image data that identify real world objects. Segmenting and tracking regions of subjective size within a view allow the application to focus on more complex tasks like object recognition within a smaller spatial domain of the entire spatial scene which reduces the processing time required to identify the object of interest.

The Object tracking using camera is a vital task within the field of computer vision. The object tracking algorithm consists of two modules; the first module is to acquire the images, detect the moving objects, segment the object or track of such object from frame to frame and estimate the position of this object, then deliver this position to the second module, which is designed as a position controller to maintain the object in the view of camera.

1.1 Implementation

We intend to blend many disciplines of study including software development, process control, electrical engineering and even mechanical engineering to develop an automatic motorized object tracking system. We will be developing a complete system from the ground up including the software, hardware, firmware, and device drivers.

Our aim is to develop a robust algorithm using multiple metrics to segment and track objects within a single image or a sequence of images. The solution proposed is to retrieve geometric regions by integrating multiple measurements from a single image or from a sequence of images. Measurements might include information like edges, motion vectors, threshold information, and color spaces.

1.1.1 Acquisition of image content

The software will have the ability to process static images from storage as well as images captured in real-time from webcams. Because our system will be a generalized object tracking system, users will specify various constraints like color, size, and shape. Optionally, the user has the option to view the intermediate steps during image processing (enhancement, segmentation, etc).

Processing / Detection of the object

The image enhancements and object segmentation algorithms applied will be dependent upon the constraints given by the user during acquisition. Object segmentation will take place by integrating color, edge, differences, and motion information. Once the image manipulation process is complete, the system will look for connected regions and filter regions based on size.

Tracking System

The goal of the tracking system is to control the camera pan and tilt such that a detected object remains projected at the center of the image. The camera tracking system hardware will include a microcontroller based servo controller that interfaces to the software running on the computer. The servo controller will adjust the viewing field of the camera by applying the

Adjustment output as a pulsed width modulated signal to the servo motors. Adjustment output will be calculated by the software in the acquisition feedback loop to center the object found with the specified user constraints. A camera mount will be fabricated for the camera as well as housing for the servos this allows a range of motion for tracking moving objects.

1.2 Problems

The problems we face in tracking a moving object in real time are two.

1. Problem due to background change.

2. Time for computation

Problem due to background change

When we start the project and study different algorithms for object tracking system we came to realize that when object is tracking and the object is gone out from the frame of camera, we rotate the camera accordingly. But if we rotate the camera with the object the background changes and then there will be false detection and we can lose the moving object.

Firstly we solve this problem by starting with fixed background. Fixed background means that we just detect the movement from a fixed area and track it only on that area; if the object goes outside that area the camera will not track it anymore. But then it became more motion detection and less object tracking. So we came across the second solution and that is we just fix the rotation of camera in some specific degree. For example camera will move from 0 degree to 60 degree and there will no stopping in between these two degrees. And camera will move from 60 to 120 degree.

Now the camera can track the moving object in those fixed areas. If the object goes out from one area the camera try to locate its movement in the next area. For example if object goes out from the frame to right side then as soon as object goes out the camera will rotate to about 60 degree to right side for the next frame and tracking the object. But the now the background is change so the algorithm will use some information according to the new frame.

1.2.2 Time Computation Problem

This is another major problem for us because if the object moving too slow or too fast then it will be difficult to track it. If the object moving too slow then the camera may be gone to next frame but the object may come back (may be it move in to and fro motion). And if the object is moving too fast then the object goes out of the frame of camera and may be on the next frame or not, while we are still computing the movement of the object.

To solve this problem we came up with quick solution of the moving object and then if even the object is slowly moving we can control the motor according to it. So we to study different algorithm for the object tracking system and then compare them to find out which one is fast.


State Of The Art

There have been many object detection and tracking techniques developed till now. They can be generally divided in two categories: object based and non-object based. Object based technique is preferable for detecting and tracking object that are different from others in video frame such that interested object can be recognize easily. Most of the techniques uses creating a model that represent the require object, getting object properties and discriminating the target object. If the object properties are not use for detection and tracking then non-object based method can be used.

Object based technique can further divided into more classes: feature based, model based and motion based. In feature based algorithm, shape and colors of objects are use to differentiate between different object in the frame. In model based approach, high level semantic representation and domain knowledge is use to differentiate required object from other objects. Motion based method analyze the sequence of video frames and extract the motion over frames in order to detect the moving objects.

Non-object based approaches use other information then the target object. It does not evaluate whether the sole object is the target object: instead, it uses some other properties such as context information which includes the information on location, time, people activity and surrounding environment.

Remaining chapter discussed the different approaches for image processing.

2.1 Image Processing Techniques

There are number of image processing techniques are mentioned in the thesis. These techniques are for different purposes such as image smoothing, edge detection and contour detection. In the following they are described briefly.

2.1.1 Median Filter

This filter is commonly use for reducing small noise in an image. Small noise is very appropriate and its gray level value is quite different from its neighbors. This technique eliminates the noise by changing the gray value of the target area according to neighbors' value. An example is given in figure 2.1: the value of neighboring pixels is 115, 119, 120, 123, 124, 125, 126, 127 and 150. The median value of this frame is 124. So the value of the targeted pixel is replaced with 124. This filter use to make the frame smoother. In this example the neighbor is 3x3. By increasing the neighborhood size more smoothness is achieved.

Figure 2.1

Morphological Operators

Dilation is one of the two basic morphological operators. It combines two set of vectors with the help of vector addition. In figure 2.2 gives an example of dilation operator. Dilation effect is to expand the boundaries of foreground object. As a result foreground of the object increase and the holes in those object become smaller.

A= {(1, 0), (1, 1), (1, 2)}

B= {(-1, -1), (-1, 0), (-1, 1), (0, -1), (0, 0), (0,1), (1, -1), (1, 0), (1,1)}

AB= {(0, -1), (0, 0), (0, 1), (0, 2), (0, 3), (1, -1), (1, 0), (1, 1), (1, 2), (1, 3), (2, -1), (2, 0), (2, 1), (2, 2), (2, 3)}

A B dilation

Figure 2.2: An example of dilation

An opposite of dilation, the erosion is another morphological operator. Erosion combines two sets using vector subtraction. This is used to carry away the boundaries of foreground object. Thus the foreground area reduces, which increases the holes within those area. Figure 2.3 is an example of erosion operator.

A= {(1, 0), (1, 1), (1, 2), (1, 3), (2, 0), (2, 1), (2, 2), (2, 3), (3, 0), (3, 1), (3, 2), (3, 3)}

B= {(-1, -1), (-1, 0), (-1, 1), (0, -1), (0, 0), (0,1), (1, -1), (1, 0), (1,1)}

AB= {(1, 1), (1, 2)}

A B Erosion

Figure 2.3: Example of erosion

LAPLACE Operator

Gray level changing in image is illustrated in gradient images. They are normally use for edge detection and can be done by some gradient generator like LAPLACE operator. LAPLACE is a gradient generator which produces four convolution masks which is nearly second derivative. These fixed masks are:

To make a gradient magnitude image, different masks are applied to an input image and the value of the targeted pixel is replaced by the result generated from the convolution. For example a value of the current pixel is changes from 100 to 179.


Figure 2.4: LAPLACE example


1. Gonzalez, R and Woods, R "Digital Image Processing", 2nd Edition, Prentice Hall,


2. Cretual, A, Chaumette, F, and Bouthemy, P "Complex Object Tracking by Visual

Servoing Based on 2D Image Motion", Proceedings of the IAPR International

Conference on Pattern Recognition. ICPR'98, (1998), pp 1251-1254

3. Starck, JL, Bijaoui, A, Valtchanov, I, and Murtagh, F, "A Combined Approach for Object

Detection and Deconvolution", Astron. Astrophys. Suppl. Ser. Vol. 147, (2000), pp 139-


4. Mahamud, S and Hebert, M "the Optimal Distance Measure for Object Detection",

Carnegie Mellon University

5. Verbeek, PW and Vliet, LJ "Line and Edge Detection by Symmetry Filters", Pattern

Recognition Group Delft, Delft University of Technology, The Netherlands

6. Song, JQ, Cai, M, and Lyu, MR "Edge Color Distribution Transform: An Efficient Tool

for Object Detection in Images", Chinese University of Hong Kong, Hong Kong, China

7. Lefevre, S, Vincent, N and Proust, C "Multi-Resolution: A Way to Adapt Object

Detection to Outdoor Backgrouds", Université de Tours, France

8. Sukmarg, O and Rao KR "Fast Object Detection and Segmentation in MPET

Compressed Domain", University of Texas at Arlington, USA

9. Schneiderman, H "A Statistical Approach to 3D Object Detection Applied to Faces and

Cars", Robotics Institute Carnegie Mellon University Pittsburgh, (2000)

Appendix A