Learning about objects in the meeting rooms from people trajectories

In ambient intelligence object recognition is an important step towards behaviour analysis and the understanding interactions between people and the environment. Existing methods focus on a detailed analysis of image content using colour, shape, texture and motion analysis (direct recognition). In this paper we present a method for recognizing furniture, i.e. chairs, tables and the walking area in a meeting room using the estimated trajectories of people (indirect recognition). We use Support Vector Machines (SVMs) to classify the activities into three categories: sitting, standing and walking to create two occupancy maps for sitting and walking spaces according to Bayesian theory. The positions of the chairs and tables are inferred from these maps. We compared the recognition of chairs and tables to ground truth data on meeting scenarios. The performance of this method is good.