Kalman Tracking for Image Processing
Applications
Student:
Julius Oyeleke
Supervisor: Dr Martin Glavin
Co-Supervisor: Dr Fearghal Morgan
Objective of Project
1. To track a red ball over a frame of video
2. Image Thresholding
3. Find the centre point of the ball
4. The use of Kalman filtering to
•
•
track the red ball in the image.
predict the path of the ball in future as an aid of detection.
5. Display with Overlay
 OpenCV (computer vision library) is being used in this project
Why OpenCV
Real time computer vision.
Provides powerful function to assist in object
identification, motion tracking etc.
Virtually assist in any image processing
application.
C-based program computer vision repository.
Step1: Image Acquiring
 commission the OpenCV system to load
frames of video into memory.
IplImage* img = cvLoadImage( argv[1] );
//determines the file format to be loaded based on the file
name
cvNamedWindow(“C:/Users/julius/Desktop/FYP/redblue.bmp);
// opens a window on the screen that can contain and display
an image
cvShowImage( “redblue.bmp”, img );
// show a named window that already exist
Step1: Problem & Solution
Problem:
• Commissioning OpenCV to read images
• Installation of OpenCV 2.0
Solution:
• Uninstall OpenCV 2.0
• Install OpenCV 1.0
Step2: Image Thresholding
 convert the RGB frames to the HSV format.
//Create gray image
cvCvtColor(src,gray,CV_BGR2GRAY);
RGB
RGB
HSV
HSV
RGB
 threshold the HSV to identify the region of interest.
cvThreshold(gray,gray,150,255,CV_THRESH_BINARY);
//Threshold to make the gray black
RGB
HSV
Threshold
RGB
output to screen
Step2: Problems & Solutions
Problems:
• Circle Detection with OpenCV 1.0
• OpenCV 1.0 takes hue value to be 0-255
Solutions:
• Uninstall OpenCV 1.0
• Install OpenCV 2.0
• In OpenCV 2.0 hue value is 0-180 (works better for the red colour detection)
• OpenCV 2.0 has a better algorithm for circle detection.
C-make
• C-make helped in compiling OpenCV from the source code
• OpenCV 2.0 needs different files for different versions of
studio.
• One will need to complete visual studio 2008 for OpenCV 2.0
Example 1:
Example2
Step3: Centre Point detection
Finding the centre point of the red ball
•
Hough transform
Step4: Implementation of the Kalman
Filtering
Kalman TrackingPredicting the
path of the Red
ball
Centre point& predicted values
Step4: Problems & Solutions
Problems:
• Kalman not tracking & predicting properly
• OpenCV only has a 1-D example
• Program Crashed at the line
CvKalmanCorrect( Kalman, z_k ); // Correct Kalman filter state
Solutions:
• 2-D was needed for this project
• I added "if (circles->total > 0)
Step5: Display with Overlay
Displaying
with overlay
Conclusions
• Project was hampered by issues, most of which
were overcome.
• Ambitious goal of the project was fully fulfilled
• Further work would lead to a complete solution