Object tracking and prediction in video sequences

OData support
Dr. Loványi István
Department of Control Engineering and Information Technology

The field of image processing is gaining more and more importance in our world. As computing power becomes more cheap and available, several different real-time object tracking solutions are being made available.

The most obvious and well-known field of real-time image processing is automated navigation, but remotely related industries as commercial advertising can benefit from image recognition, too. When a company creates an advertisement video, or uses product placement to gain publicity through movies or TV shows, the actual time a given trademark appeared on air needs to be tracked to provide measurements with which the fulfillment of a contract can be controlled.

I designed a tool with the goal of performing real-time trademark matching using the combination of two very different object tracking algorithms, augmented with prediction and filtering of measurements.

Trying to achieve real-time performance on high-definition image sequences with a frame rate of 25 FPS, I examined the possibilities of parallel programming, and implemented parts of my software to utilize the power of modern parallel computing devices.


Please sign in to download the files of this thesis.