Abstract

In this thesis a novel system is presented capable of detecting and tracking multiple people wearing similar clothing using multiple synchronized and calibrated cameras. The system improves on existing systems in the following way: Instead of segmenting objects using color models this research uses face detection, and geometrical classification to segment objects from one another.
The problem is tackled by developing some innovative algorithms and methods. It designs a parallel implementation of the Viola and Jones face detector to detect faces in different orientations. The information from multiple cameras about tracked objects is combined to produce better results. An innovative algorithm for object-location estimation given multi-view classification data is developed and a novel classification algorithm is presented based on previous position and orientation off the tracked object. Finally a system is constructed that combines the fields of face detection, tracking and multi-perspective reasoning.With these algorithms a running prototype of the system is created. This prototype has been successfully tested using a simulated environment and also with real-live camera footage.