Person Detection in the Restaurant of the Future

Many computer vision techniques dealing with humans require that the location within an image of a person is known. In this research, we investigated into the subject of person detection in a real-world environment: the Restaurant of the Future. We evaluated several methods and selected one: person detection using face detection. In order to apply this we evaluated several known methods for face detection. Then we determined which aspects of the recordings from the Restaurant impacted the performance of the face detectors. All methods suffered strongly from the complex background in the Restaurant and the low quality of the recordings. The angled viewpoint had an impact on several of the methods while the video encoding used affected others. On the worst conditions, the Viola-Jones detector performed best. The analysis of the weaknesses of the face detectors will give future researchers a starting point in improving known or finding new methods.