Datenbestand vom 19. Juli 2019
Tel: 089 / 66060798
Mo - Fr, 9 - 12 Uhr
Fax: 089 / 66060799
aktualisiert am 19. Juli 2019
978-3-8439-0036-2, Reihe Elektrotechnik
New active safety Approaches - Left behind Occupant Detection and physiological Parameter Sensing
178 Seiten, Dissertation Bergische Universität Wuppertal (2010), Softcover, B5
The objectives of the developed safety approaches are focused on two upcoming safety devices. Heartbeat and respiration information are needed for the development of a driver state monitor (DSM). The DSM is intended for the observation of the driver’s concentration on the actual traffic. It should prevent accidents due to drowsiness and fatigue.
Another field of work is defined by a research of the National Highway Traffic Safety Administration (NHTSA). It records an increasing risk of death by heat stroke for unattended left behind children in parked cars on summer days. To prevent these avoidable deaths the NHTSA evaluates several concepts of left behind occupant recognition devices.
Content of the thesis is the development of two new sensing approaches to fulfill future safety needs of the automotive market.
The first one is based on high sensitive analogue accelerometers that monitor vibrations occurring at the car chassis. Investigations showed a recognizable signal produced by human beings seated in the parked vehicle. Its origin is medically known as human tremor. The human tremor is an unintentional, rhythmic, oscillating muscle movement which can not be suppressed by the individual itself.
The second approach is founded on a re-engineered series product called “passive occupant detection system” model B (PODS-B). It contains a silicon oil filled bladder mat which is integrated in the front passenger seat. The mat includes a connected pressure sensor to measure the applied loading force on the seat cushion. With the help of developed electronics physiological parameters (heartbeat and respiration movement) are extracted.
After the evaluation of both sensing concepts the classification by machine learning techniques is prepared. This involves a feature extraction, normalization and correlation-based selection to gain proper classification spaces for each approach. Additionally a fused space of both sensors is constructed. Subsequently four data mining algorithms (SVM, k-NN, J48, PNN) are evaluated on the given feature bases to determine the best for each task.
Finally a conclusion of the thesis is given and further steps up to the first automotive A-samples are mentioned.