OpenTrack - Automated Camera Control for Lecture Recordings

In this paper we present the current state of our research project which aims to develop an open-source framework for real-time scene analysis and automatic camera control in lecture recording scenarios. The system is designed to run in alliance with the Opencast Matter horn lecture capture system as well as stand-alone. A GPU-based scene segmentation technique using motion cues and background modeling has been implemented using OpenCL. Moving objects are tracked by their centroids and bounding boxes and a dynamic appearance model is used to give persons a relative identity. Development started on a scriptable virtual camera operator module that will drive PTZ-cameras. Other applications of the scene analysis are possible.