Camera Based Automatic Behaviour Tracking among Children

Although object tracking solutions based on signal detection or sensor network have already been long known, these are only able to describe the direction and speed of the movement, although sometimes even with cm exactitude. Meanwhile they are not capable of providing information related to the direction of the sight of a person whose behaviour/movement is tracked, i.e. to what extent is the course he/she is following determined by an actual endpoint i.e. the destination and by any other circumstances. There have been several such researches conducted by human observers, but excessive number of such observations would be very costly. There is another problem inherent in former methods: track segments describing continuous movement in spaces that due to obstacles and overlays could be watched by several cameras should afterwards be manually consolidated. As to our knowledge, currently accessible solutions are not capable to identify completely the space sections covered by a person’s sight. In case video recordings could be processed by automated systems, the observation of a sufficiently large number of cases would enable the investigation of the interaction between the actual course of movement and the view of the environment; research results could directly be obtained on several aspects of spatial orientation. Our objective is to design and develop an experimental, multimedia-aided IT system that is capable of processing video camera recordings and thereby, in a highly automated way, distinguish observed objects and persons, watch and describe their spatial movements and calculate statistical distributions from such descriptions. With the help of the descriptions of spatial and time coordinates of tracks, we could detect regularities and irregularities intrinsic in tracks followed, in separation/grouping of the observed objects or persons. In the present article, the architecture of the planned system is described. The recording tool of the experimental system is a stereo camera system producing two time-synchronized video streams. With the appropriate optics, lighting and arrangement, this system is capable to continuously distinguish objects moving simultaneously and to provide instantaneous information appropriate for measuring the actual spatial location of the objects. The image analysing module identifies objects to be observed, identifies their actual position and records their places relative to each other in a track curve database. The track analyser module contains the procedures for the classification and analysis of movement curves. Finally, we are presenting an initial application carried out by the experimental system where children could be studied from the aspect of communal behaviour, trackfollowing/target orientation and efficiency.