Model-based tracking on conveyor belts: Evaluation and practical results in the automotive industry

Model based tracking (MBT) of painted cars in the automotive mass production on conveyor belts with robots is a challenging task. Many disturbing sources that have an impact on the MBT exist, like the influence of the localized work illumination, the synchronization of the MBT to the conveyor belt, reflections in the paint, variants of the cars and the complexity of the used CAD models. By having such complex systems the mere assessment of the accuracy and stability of MBT approaches by literature can be hard. A real world application is necessary for a better understanding. Therefore, we present the evaluation of MBT for a robotic gap measurement system on painted cars. The influence of local lighting and car paint is analysed in detail regarding the MBT accuracy. To reduce complexity considering the car variants on a production line, we evaluated the MBT with different model setups and show the influence on the MBT results. Regarding MBT in a complex calibrated system that runs twenty-four-seven, a broken or slightly displaced camera should not have a huge impact like a loss of production. For this reason, we present a method to exchange a camera within minutes and without a loss of the overall accuracy. The method relies on a separate test specimen and is evaluated in detail. The presented evaluations can help researchers and the industry to better understand and assess the influence and correlations of different error sources or disturbing factors for the usage of MBT in complex conveyor belt based robotic applications