Tracking and indexing moving object in multitude environment

• Main Author: Abdul Kadir, Muhd. Khairulzaman
• Format: Thesis
• Language: English
• Published: 2007
• Subjects: TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://eprints.utm.my/id/eprint/6406/1/MuhdKhairulzamanAbdulKadirMFKE2007.pdf; http://eprints.utm.my/id/eprint/6406/; http://dms.library.utm.my:8080/vital/access/manager/Repository/vital:62215

The project will be focusing on tracking multiple people in various environments specifically for outdoor scene. This will also involve in indexing each different person in the same area or background. Tracking 1 moving object is an easy works but if tracking involved more than 1 people the process will become harder. The complexity also is more complicated if it involves moving object occlusion and illuminations change in the image frame. In this tracking and indexing system, background subtraction model for each frame is being used for extracting the moving object from the background. But, before background subtraction model is executed, each frame (background and current) will be filtered by using Gaussian filter for reducing small noise in the frame and morphological filter process is perform after background subtraction. Next, each moving object will be labeled so as to differentiate each different people that exist in the same background or environment. This was done by using feature-based model method which used area, center point of each moving people and the average of RGB pixels value as recognition. All the work will be done in a grayscale image and applied to every frame. Without loss of generality, the indexing algorithm will be done up to 4 people within the same background with different types of actions (different posture) and different type of conditions (walking slow and faster). The results of this system are 100% accurate for 1 and 2 moving people without any errors. But, if the moving objects are from 3 to 4 people, the accuracy reduces around 25% due to the feature not robust enough in differentiating it.