Collaborative Project: Improvement of work safety at circular saw benches and dimension saws Subproject: Development and evaluation of a computer based approach to detect body arts in video image sequences and realisation of a prototype

Status:

completed 02/2007

Aims:

In a pilot study carried out in cooperation with the "Berufsgenossenschaftlichen Institut für Arbeitsschutz", it was shown that the head, shoulder and arm positions of a person standing at a circular saw can be recognized via the processing of electronic video images. The goal of this research project is to develop a fully functional prototype which generates a signal to automatically stop the machine wihtin 40 ms whenever a body part of the operator enters the specified danger zone of the wood processing machine. The recognition process which we plan to develop will utilize video data recordet on standard, off-the-shelf cameras, from which a triggering mechanism can be driven. This process shall be tested under real working conditions in order to be able to gain an accurate assessment of the system`s sturdiness and reliability.

Activities/Methods:

Within the framework of this subproject, a video-based sensor will be developed which is able to register the presence of parts of the body within a defined danger zone in the shortest possible time. For the implementation of the video-based detection process, a course - to - fine strategy will be followed. This means that beginning with the head and shoulder area of the person currently standing at the machine, a step-by-step detection and tracking of the arms, hands and, finally, fingers will be made using a specially-developed process of analysing video signals. In particular, actual conditions such as those usually encountered in workshops, will be taken into account. This includes the development of appropriate methods for dealing with different and, in particular, changing backgrounds, and which thus make possible the reliable differentiation between background and foreground. The same holds true for the consideration of variable lighting conditions. In addition, these methods should be fundamentally so designed as to enable the video-based system to detect and monitor multiple users. For each person within the visual field of the camera system, an independent thread will be generated, so that for every person an independent and autonomous process can be edited. The modelling of a person with his or her body parts and extremities presents a special challenge. For this a basic modelling approach will be developed which allows for an appropriate parameterization of the body parts using the two-dimensional data delivered by a conventional video camera to be made, so that the detection of the body parts can be made at every moment. The "human model" must be flexible enough in order to allow for arbitrary yet physiologically reasonable movements.

Last Update: