ZURUECK HOCH VOR INHALT SUCHEN

» Back to overview
Proposing Institution

Institute for Materials Handling, Material Flow, Logistics; TUM
Project Manager

Armin Lang
Boltzmannstraße 15
85748 Garching
Abstract
In the project "PräVISION", we try to find methods in order to prevent occupational accidents in intralogistic environments caused by forklift drivers. The idea is to develop a warning system, which is able to predict upcoming collisions by using latest algorithms of computer vision. The main challenge is to avoid unnecessary warnings but to warn whenever it is necessary. This challenge shall be mastered by combining collision detection with people detection algorithms. The latter ones will be used to be able to warn earlier, when humans are in the driveway of the forklift and endangered. When no human is in sight, the system should warn at that time, when it's the last change for driver to brake. To be able to distinguish between humans and warehouse accomodation, we use amongst others neural networks. Because the time-of-flight camera we use for capturing the environment is mounted on the top of the forklift, we can't use pre-trained models for people detection. The most algorithms are trained with stationary captured frontal pictures of humans. At first, we do have the situation of a moving camera, that is in addition not mounted horizontally but in a specific angle on the top of the truck. At second, we get infarered images from the time-of-flight camera, which differ quite much from ordinary color (rgb) images. So, we have to train the neural networks ourself to be able to detect people in our application case.

Impressum, Conny Wendler