Intelligent camera management systems were developed to automatically record meetings for videoconferencing. These systems provided many benefits, such as reducing the production cost and conveniently documenting events. However, automatically recorded videos in general were not visually engaging. This paper presents a novel approach that intelligently controls camera shots and angles to improve the visual interest. We use 3D infrared images captured by a Kinect sensor to recognize active speakers and their positions in a meeting. A movable camera, constructed by placing a wireless PTZ (pan-tilt-zoom) camera on top of a motorized rail, can automatically move its position to frame an active speaker in the center of the screen. Without interrupting the meeting, a speaker can seamlessly switch video sources through gesture-based commands. We have summarized and implemented a set of heuristic rules to simulate a human director. These rules can be visually edited through a graphical user interface. The customization of a virtual director makes our system applicable in various scenarios. We conducted a user study, and the evaluation results justified the quality of an automated video.