Evaluation of an Intelligent Collision Warning System for Forklift Truck Drivers in Industry

摘要

The number of collisions caused by lift trucks in the area of intralo-gistics is still increasing despite the availability of collision avoidance systems. Commercially available products for collision avoidance mounted on forklifts often issue false alarms activated by minimum distances during daily work. Consequently, the drivers sooner or later turn those systems off. A collision warning system based on computer-vision methods combined with a time-of-flight camera delivering 2D and 3D data can overcome inflationary warnings in warehouse situations. The 3D data delivered is used to identify objects by clustering as well as to get information about the movement of objects in forklift's path. Machine-learning algorithms use the 2D data mainly to detect people in the path. Distinguishing people from non-human objects makes it possible to establish a two-level warning system able to warn earlier if humans are endangered than in collision situations in which no humans are in sight. This system's general functionality has already been proven in lab tests. To transfer the academic results to application in an industrial environment, the same test procedure has been executed during daily work in a warehouse at a company in the production sector. In this paper, the authors aim to list the differences and commonalities between the academic and industrial runs.